Merge remote-tracking branch 'gitea/develop' into feat/501-mcp-apikey-auth

apps/server/src/app.module.ts

@@ -25,7 +25,7 @@ import { CacheModule } from '@nestjs/cache-manager';
 import KeyvRedis from '@keyv/redis';
 import { LoggerModule } from './common/logger/logger.module';
 import { ClsModule } from 'nestjs-cls';
-import { NoopAuditModule } from './integrations/audit/audit.module';
+import { AuditModule } from './integrations/audit/audit.module';
 import { ThrottleModule } from './integrations/throttle/throttle.module';
 import { McpModule } from './integrations/mcp/mcp.module';
 import { SandboxModule } from './integrations/sandbox/sandbox.module';
@@ -55,7 +55,7 @@ try {
       middleware: { mount: true },
     }),
     LoggerModule,
-    NoopAuditModule,
+    AuditModule,
     CoreModule,
     DatabaseModule,
     EnvironmentModule,

apps/server/src/collaboration/yjs.util.spec.ts

@@ -529,4 +529,107 @@ describe('replaceYjsMarkedText', () => {
     expect(result).toEqual({ applied: false, currentText: 'abcdef' });
     expect(text.toDelta()).toEqual(before);
   });
+
+  // #496: apply must NOT silently strip the replaced run's inline formatting.
+  // Build a paragraph and format the marked range with extra marks, then assert
+  // the replacement carries them.
+  function buildFormatted(
+    runs: Array<{ text: string; attrs?: Record<string, any> }>,
+  ): { fragment: Y.XmlFragment; text: Y.XmlText } {
+    const ydoc = new Y.Doc();
+    const fragment = ydoc.getXmlFragment('default');
+    const para = new Y.XmlElement('paragraph');
+    fragment.insert(0, [para]);
+    const text = new Y.XmlText();
+    para.insert(0, [text]);
+    text.insert(0, runs.map((r) => r.text).join(''));
+    let offset = 0;
+    for (const run of runs) {
+      if (run.attrs) text.format(offset, run.text.length, run.attrs);
+      offset += run.text.length;
+    }
+    return { fragment, text };
+  }
+
+  it('preserves the original run formatting (bold + link) on the replacement', () => {
+    const { fragment, text } = buildFormatted([
+      { text: 'see ' },
+      {
+        text: 'old',
+        attrs: {
+          comment: { commentId: 'c1', resolved: false },
+          bold: true,
+          link: { href: 'https://x.test' },
+        },
+      },
+      { text: ' end' },
+    ]);
+
+    const result = replaceYjsMarkedText(fragment, 'c1', 'old', 'new');
+
+    expect(result).toEqual({ applied: true, currentText: 'new' });
+    // The comment anchor AND the bold/link marks survive the delete+insert.
+    expect(text.toDelta()).toEqual([
+      { insert: 'see ' },
+      {
+        insert: 'new',
+        attributes: {
+          comment: { commentId: 'c1', resolved: false },
+          bold: true,
+          link: { href: 'https://x.test' },
+        },
+      },
+      { insert: ' end' },
+    ]);
+  });
+
+  it('mixed formatting under the mark: replacement takes the DOMINANT (longest) run, NOT the leading one', () => {
+    // Leading run is SHORT + plain ("x", 1 char); the following run is LONGER +
+    // bold ("bolded", 6 chars), same commentId. The longest run is deliberately
+    // NOT first: a "first-wins" pick would carry plain (no bold), so asserting
+    // bold on the result only holds if the code genuinely selects the LONGEST run.
+    const { fragment, text } = buildFormatted([
+      { text: 'x', attrs: { comment: { commentId: 'c1', resolved: false } } },
+      {
+        text: 'bolded',
+        attrs: { comment: { commentId: 'c1', resolved: false }, bold: true },
+      },
+    ]);
+
+    const result = replaceYjsMarkedText(fragment, 'c1', 'xbolded', 'Z');
+
+    expect(result).toEqual({ applied: true, currentText: 'Z' });
+    expect(text.toDelta()).toEqual([
+      {
+        insert: 'Z',
+        attributes: { comment: { commentId: 'c1', resolved: false }, bold: true },
+      },
+    ]);
+  });
+
+  it('mixed formatting under the mark: on a length tie the FIRST run wins', () => {
+    // Two equal-length runs (2 chars each) with different formatting, same
+    // commentId. The reduce keeps the accumulator on a tie, so the FIRST run
+    // (italic) prevails over the later bold one.
+    const { fragment, text } = buildFormatted([
+      {
+        text: 'AA',
+        attrs: { comment: { commentId: 'c1', resolved: false }, italic: true },
+      },
+      {
+        text: 'BB',
+        attrs: { comment: { commentId: 'c1', resolved: false }, bold: true },
+      },
+    ]);
+
+    const result = replaceYjsMarkedText(fragment, 'c1', 'AABB', 'Z');
+
+    expect(result).toEqual({ applied: true, currentText: 'Z' });
+    expect(text.toDelta()).toEqual([
+      {
+        insert: 'Z',
+        attributes: { comment: { commentId: 'c1', resolved: false }, italic: true },
+      },
+    ]);
+  });
 });

apps/server/src/collaboration/yjs.util.ts

@@ -145,6 +145,10 @@ type MarkedSegment = {
   length: number;
   text: string;
   markAttrs: Record<string, any>;
+  // The FULL attribute set of this delta run — the `comment` mark plus any
+  // inline formatting (bold/italic/code/link/…). Captured so apply can carry the
+  // original run's formatting onto the replacement instead of dropping it.
+  attributes: Record<string, any>;
 };
 
 /**
@@ -202,6 +206,7 @@ export function replaceYjsMarkedText(
             length,
             text: insert,
             markAttrs: markAttr,
+            attributes,
           });
         }
         offset += length;
@@ -251,15 +256,25 @@ export function replaceYjsMarkedText(
     return { applied: false, currentText: joinedText };
   }
 
-  // 3. All guards passed: delete the marked run and re-insert newText with the
-  // same comment attributes at the same offset. Atomic within the caller's
-  // transaction.
+  // 3. All guards passed: delete the marked run and re-insert newText at the
+  // same offset. Atomic within the caller's transaction.
   const start = segments[0].offset;
   const len = segments.reduce((sum, s) => sum + s.length, 0);
-  const markAttrs = segments[0].markAttrs;
+
+  // Carry the ORIGINAL run's formatting onto the replacement (#496): inserting
+  // with only the `comment` mark silently dropped bold/italic/code/link of the
+  // replaced text. Yjs applies one flat attribute set to the whole insert, so
+  // when the marked run mixes formatting we pick the DOMINANT segment (the one
+  // covering the most characters) and apply its attributes — a v1 that preserves
+  // the common single-format case exactly and, for a mixed run, keeps the
+  // prevailing style rather than losing all of it. `attributes` already carries
+  // the `comment` mark (every collected segment is filtered on it above), so the
+  // anchor is preserved by copying the run's attribute set verbatim.
+  const dominant = segments.reduce((a, b) => (b.length > a.length ? b : a));
+  const insertAttrs = { ...dominant.attributes };
 
   node.delete(start, len);
-  node.insert(start, newText, { comment: markAttrs });
+  node.insert(start, newText, insertAttrs);
 
   return { applied: true, currentText: newText };
 }

apps/server/src/core/ai-chat/ai-chat-stream-registry.service.ts

@@ -1,42 +1,122 @@
 import { Injectable, Logger, OnModuleDestroy } from '@nestjs/common';
 
 /**
- * In-memory run-stream registry (#184 phase 1.5). A durable agent run tees its
- * SSE frames here (via `pipeUIMessageStreamToResponse({ consumeSseStream })`)
- * so a LATE tab — one that reloaded, or opened after the starter dropped — can
- * attach through `GET /ai-chat/runs/:chatId/stream`, replay the frames buffered
- * so far, and then follow the live tail as a normal streamer.
+ * In-memory run-stream registry (#184 phase 1.5, step-aligned retention #491). A
+ * durable agent run tees its SSE frames here (via
+ * `pipeUIMessageStreamToResponse({ consumeSseStream })`) so a LATE tab — one that
+ * reloaded, or opened after the starter dropped — can attach through
+ * `GET /ai-chat/runs/:chatId/stream`, be handed the TAIL past the step it already
+ * has persisted, and then follow the live tail as a normal streamer.
  *
  * This is deliberately single-process and best-effort: it holds nothing the DB
  * does not (the run + assistant row are the source of truth), so a process
  * restart simply drops in-flight entries and the client falls back to its
  * restore + degraded-poll path. The async `attach` return type is the seam for a
  * future phase-2 cross-process backend (Redis) — the interface does not change.
+ *
+ * ── #491 step-aligned retention (the OOM fix) ────────────────────────────────
+ * The old registry buffered up to 32MB of raw SSE frames PER active run (V8 ~2×
+ * in memory) and, on attach, blasted the WHOLE buffer to the socket synchronously
+ * with no drain — a handful of marathon runs on a 1GB container OOM'd. #491 caps
+ * the ring at a few MB (env-tunable, default 4MB) and keeps it there by ROTATING:
+ *
+ *  - Every buffered frame is STAMPED with a step number at tee (see ingestFrame).
+ *    Convention: the stamp of a frame is the number of `finish-step` parts seen
+ *    BEFORE it (starting at 0). The finish-step frame itself carries the current
+ *    value, THEN the counter increments. So a frame stamped `s` is the content of
+ *    the (s+1)-th step — 0-based step index `s` — and the stamp aligns EXACTLY
+ *    with `metadata.stepsPersisted`: a client whose persisted `stepsPersisted` is
+ *    N has steps 0..N-1 on disk (and in its seed) and needs the tail `stamp >= N`.
+ *
+ *  - The ring rotates ONLY on a CONFIRMED persist of step N
+ *    (`confirmPersistedStep`), dropping frames with `stamp < N` (those steps are
+ *    now on disk and a fresh client seed carries them). A NON-confirmed step is
+ *    never rotated away, so a persist FAILURE just makes the ring cover MORE
+ *    (auto-safe). This is the anti-inversion rule: a naive "rotate in .then()"
+ *    that rotated after an UNwritten step would drop a step nobody has → silent
+ *    hole. Rotation is gated on a real, successful persist.
+ *
+ *  - If the ring still exceeds its byte cap after rotation (a single fat step, or
+ *    a lagging persist), the OLDEST frames are evicted to stay bounded. Evicting a
+ *    not-yet-persisted frame opens a GAP: an attach whose N falls at or below an
+ *    evicted step answers 204 and the client degrades to restore+poll. The gap is
+ *    NOT sticky — the coverage floor is recomputed from the ring, so a later
+ *    persist that rotates past the holey steps clears it.
+ *
+ * ── attach numbering / coverage (the wire convention) ────────────────────────
+ * The step marker N comes ONLY FROM THE CLIENT (a query param). The server never
+ * reads the row to derive N — a server-side N from a stale seed would open a
+ * silent one-step hole. N is the client's persisted `stepsPersisted` (a COUNT):
+ *   - the tail it needs = frames with `stamp >= N`;
+ *   - coverage is OK ⟺ `coverageFloor(entry) <= N`, where coverageFloor is the
+ *     smallest step FULLY present in the ring (its smallest retained stamp, bumped
+ *     by one when that leading step was only partially evicted by overflow). If
+ *     `coverageFloor > N` the ring starts AFTER the client's frontier (a hole, or
+ *     the client's seed simply lagged behind a rotation) → 204 → the client
+ *     refetches (a larger N) and re-attaches.
+ * The N cutoff is applied in ALL branches, INCLUDING the finished-retained replay.
+ *
+ * ── same-tick invariants (unchanged, still load-bearing) ─────────────────────
+ * invariant 1: only the matching run may mutate/observe an entry (runId check).
+ * invariant 2: retention deletes ONLY its own entry (a replacement may own the key).
+ * invariant 3: open() over a live entry mirrors the done-path (subscribers released).
+ * invariant 4: the tail SLICE + subscriber registration happen in ONE synchronous
+ *              tick inside attach() — no await between them — so a concurrently
+ *              ingested frame is EITHER in the snapshot (buffered before the sync
+ *              block, and the just-added subscriber never sees it) OR fanned out to
+ *              the paused subscriber's `pending` (ingested after) — never both and
+ *              never neither: no loss, no duplication. NOTE (#491): the controller
+ *              now AWAITS the drain-respecting tail write BEFORE calling start(), so
+ *              frames ingested during that await accumulate in `pending`; this is
+ *              bounded by the subscriber cap (an overflow degrades start() to an
+ *              end(), a 204-equivalent). It is the SYNCHRONOUS snapshot+registration
+ *              — not a same-tick start() — that makes this correct.
+ * invariant 5: the controller wires close-cleanup BEFORE any write.
+ * invariant 6: no cross-run replay — the `anchor` (the client's assistant row id)
+ *              must match this run's assistant id, or a foreign run's transcript
+ *              would be appended to the client's message.
  */
 
 /** How long a finished entry is retained for late attach (replay + immediate end). */
 export const RUN_STREAM_RETAIN_FINISHED_MS = 30_000;
 
 /**
- * Per-run replay buffer cap. Past this the buffer is dropped (attach -> 204, and
- * the client falls back to its restore + degraded-poll path, #430).
- *
- * Raised from 4MB to 32MB (#430): marathon autonomous runs (11-25 min observed)
- * stream far more than 4MB of SSE frames, so a live disconnect mid-run would find
- * an already-overflowed buffer and could only degrade-poll instead of re-attaching
- * to the live tail. 32MB comfortably covers those runs while staying bounded.
- *
- * Memory cost: this is the WORST-CASE retained size PER ACTIVE run (the buffer is
- * freed on finish + retention, or dropped immediately on overflow). With the small
- * number of concurrent autonomous runs a single workspace realistically has, 32MB
- * each is an acceptable ceiling; the overflow->204->degraded-poll fallback remains
- * the backstop for anything larger, so correctness never depends on this bound.
+ * DEFAULT per-run replay ring cap (#491, down from 32MB). SSE frames carry
+ * UNcompacted tool outputs + framing overhead (×1.5–2 vs the persisted parts), so
+ * a "2–3 large reads + reasoning" step routinely blows past 2MB; 4MB comfortably
+ * holds a step or two of TAIL, which is all a resuming client needs (steps below
+ * its persisted frontier come from the seed, not the ring). The ring stays bounded
+ * because it rotates on every confirmed persist; this cap is only the ceiling for
+ * the un-persisted tail between rotations. Env-tunable via
+ * AI_CHAT_RUN_STREAM_MAX_BUFFER_BYTES (bytes); a 0/invalid value falls back to this.
  */
-export const RUN_STREAM_MAX_BUFFER_BYTES = 32 * 1024 * 1024;
+export const AI_CHAT_RUN_STREAM_MAX_BUFFER_BYTES = 4 * 1024 * 1024;
 
-// 2x the replay cap: a just-written full-replay burst alone can never trip the
-// per-subscriber cap (see controller); only a genuinely stalled socket can.
-export const SUBSCRIBER_MAX_BUFFERED_BYTES = 2 * RUN_STREAM_MAX_BUFFER_BYTES;
+// 2× the ring cap: a just-written full-tail burst alone can never trip the
+// per-subscriber cap (see controller); only a genuinely stalled socket can. This
+// derivative relationship is preserved even when the ring cap is env-overridden.
+export const SUBSCRIBER_MAX_BUFFERED_BYTES = 2 * AI_CHAT_RUN_STREAM_MAX_BUFFER_BYTES;
+
+/**
+ * A finish-step boundary frame is exactly `data: {"type":"finish-step"...}\n\n`
+ * (verified empirically against ai@6.0.207 — each UI-message-stream part is a
+ * single `data: {json}\n\n` event, never split across `data:` lines, and `type`
+ * is always the first key). A prefix match is cheaper than JSON.parse-per-frame
+ * and has no false positives: a literal `"type":"finish-step"` inside a text
+ * delta is JSON-escaped (`\"type\":...`), and the frame would start with
+ * `data: {"type":"text-delta"` anyway.
+ */
+const FINISH_STEP_FRAME_PREFIX = 'data: {"type":"finish-step"';
+
+/** Resolve the ring cap from the environment, falling back to the default. */
+function resolveMaxBufferBytes(): number {
+  const raw = process.env.AI_CHAT_RUN_STREAM_MAX_BUFFER_BYTES;
+  if (!raw) return AI_CHAT_RUN_STREAM_MAX_BUFFER_BYTES;
+  const parsed = Number(raw);
+  return Number.isFinite(parsed) && parsed > 0
+    ? Math.floor(parsed)
+    : AI_CHAT_RUN_STREAM_MAX_BUFFER_BYTES;
+}
 
 export interface RunStreamCallbacks {
   onFrame: (frame: string) => void;
@@ -44,6 +124,9 @@ export interface RunStreamCallbacks {
 }
 
 export interface RunStreamAttachment {
+  // The synthetic `start` frame (carrying { runId, chatId }) followed by the
+  // buffered TAIL filtered to `stamp >= N`. The controller writes these to the
+  // socket in chunks respecting drain, then calls start().
   replay: string[];
   finished: boolean;
   start(): void; // drain pending frames (order preserved) and go live
@@ -53,14 +136,19 @@ export interface RunStreamAttachment {
 interface Subscriber extends RunStreamCallbacks {
   started: boolean;
   pending: string[];
-  // Byte size of `pending`, capped at SUBSCRIBER_MAX_BUFFERED_BYTES. `start()` is
-  // called in the SAME tick as `attach()` today (see attach), so `pending` never
-  // holds more than one microtask of frames — but the async `attach` signature is
-  // a phase-2 seam: an await between attach and start would let a stalled paused
-  // subscriber buffer the WHOLE run here. The cap is the structural backstop.
+  // Byte size of `pending`, capped at the subscriber cap. `start()` is called in
+  // the SAME tick as `attach()` today, so `pending` never holds more than one
+  // microtask of frames — but the controller writes the (potentially large) tail
+  // respecting drain BEFORE start(), so a stalled socket can accumulate here; the
+  // cap is the structural backstop (an overflow degrades start() to an end()).
   pendingBytes: number;
   overflowed: boolean;
   pendingEnd: boolean;
+  // The client's step frontier N: this subscriber only receives frames with
+  // `stamp >= minStamp` (the tail past what it already persisted). Live frames
+  // always satisfy this (their stamp is the current, highest step), so it only
+  // filters the rare out-of-order below-frontier frame.
+  minStamp: number;
 }
 
 interface Entry {
@@ -68,8 +156,20 @@ interface Entry {
   // The persisted assistant row id of this run (set at bind; undefined if the
   // seed failed). Used by the attach anchor check (invariant 6).
   assistantMessageId?: string;
+  // Parallel arrays: frames[i] is the SSE string, stamps[i] its step number.
   frames: string[];
+  stamps: number[];
   bytes: number;
+  // The running step counter used to stamp the NEXT frame (number of finish-step
+  // frames seen so far).
+  currentStamp: number;
+  // The highest confirmed `stepsPersisted`: frames with stamp < persistedFloor are
+  // on disk (safe to drop, never re-buffered). Monotonic (confirmPersistedStep).
+  persistedFloor: number;
+  // The highest stamp EVICTED by an overflow (unsafe) drop, -1 if none. Used to
+  // detect a partially-evicted leading step when computing the coverage floor.
+  overflowThroughStamp: number;
+  // Sticky-for-logging only: at least one unsafe (overflow) eviction happened.
   overflowed: boolean;
   finished: boolean;
   subscribers: Set<Subscriber>;
@@ -80,6 +180,10 @@ interface Entry {
 export class AiChatStreamRegistryService implements OnModuleDestroy {
   private readonly logger = new Logger(AiChatStreamRegistryService.name);
   private readonly entries = new Map<string, Entry>(); // key: chatId
+  // Env-resolved caps (per instance) so a deployment can tune the ceiling without
+  // a code change. The subscriber cap keeps the documented 2× relationship.
+  readonly maxBufferBytes = resolveMaxBufferBytes();
+  readonly subscriberMaxBufferedBytes = 2 * this.maxBufferBytes;
 
   /**
    * Register a fresh entry at the START of a run (before any frame), so a tab
@@ -105,7 +209,11 @@ export class AiChatStreamRegistryService implements OnModuleDestroy {
     this.entries.set(chatId, {
       runId,
       frames: [],
+      stamps: [],
       bytes: 0,
+      currentStamp: 0,
+      persistedFloor: 0,
+      overflowThroughStamp: -1,
       overflowed: false,
       finished: false,
       subscribers: new Set<Subscriber>(),
@@ -150,6 +258,34 @@ export class AiChatStreamRegistryService implements OnModuleDestroy {
     void pump();
   }
 
+  /**
+   * Confirm that step `stepsPersisted` (a COUNT: steps 0..stepsPersisted-1) is on
+   * disk for this run, and ROTATE the ring: drop the buffered frames of those
+   * now-persisted steps (stamp < stepsPersisted). This is the ONLY thing that
+   * rotates the ring, and it is called ONLY after a genuinely SUCCESSFUL per-step
+   * persist (see ai-chat.service updateStreaming). A failed persist never calls
+   * it, so the ring covers more (auto-safe). Identity-checked (invariant 1) and
+   * monotonic (a stale lower count is ignored).
+   */
+  confirmPersistedStep(
+    chatId: string,
+    runId: string,
+    stepsPersisted: number,
+  ): void {
+    const entry = this.entries.get(chatId);
+    if (!entry || entry.runId !== runId) return;
+    if (!Number.isFinite(stepsPersisted) || stepsPersisted <= entry.persistedFloor)
+      return;
+    entry.persistedFloor = stepsPersisted;
+    // Clean rotation: drop the persisted steps from the head. These frames are on
+    // disk + carried by a fresh client seed, so this NEVER opens a gap.
+    while (entry.frames.length > 0 && entry.stamps[0] < stepsPersisted) {
+      entry.bytes -= Buffer.byteLength(entry.frames[0]);
+      entry.frames.shift();
+      entry.stamps.shift();
+    }
+  }
+
   /**
    * Terminate a run's entry from the OUTER catch of the stream method (a failure
    * before/while wiring the pipe, so `done` will never arrive). Identity-checked
@@ -162,36 +298,77 @@ export class AiChatStreamRegistryService implements OnModuleDestroy {
   }
 
   /**
-   * Attach to a run's stream. Async only for the phase-2 Redis seam — the body
-   * runs synchronously so the replay snapshot and the subscriber registration
-   * happen in ONE tick with no await between them (invariant 4): a frame ingested
-   * concurrently cannot slip into the gap and be lost or duplicated.
+   * Attach to a run's stream from the client's step frontier `n` (its persisted
+   * `stepsPersisted`). Async only for the phase-2 Redis seam — the body runs
+   * synchronously so the tail SLICE and the subscriber registration happen in ONE
+   * tick with no await between them (invariant 4).
    *
    * Returns null (-> the caller answers 204) when:
-   *  - there is no entry, or it overflowed (replay is gone);
-   *  - expect=live with an anchor that does not match this run's assistant id
-   *    (invariant 6: a stripped tab must never replay a FOREIGN run's transcript);
-   *  - the run finished and the caller did not expect a live tail.
-   * A finished run with expect=live yields a replay-only attachment (no
-   * subscriber registered). Otherwise a paused subscriber is registered and the
-   * caller replays `replay`, then calls start() to drain and go live.
+   *  - there is no entry;
+   *  - the `anchor` does not match this run's assistant id (invariant 6);
+   *  - the ring does not cover the client's frontier (coverageFloor > n): a hole
+   *    from overflow, or the client's seed simply lagged behind a rotation. The
+   *    client then refetches (a larger n) and re-attaches.
+   *
+   * Otherwise the attachment's `replay` is a synthetic `start` frame (the run-fact
+   * on re-attach) followed by the buffered tail filtered to `stamp >= n`. For a
+   * FINISHED run this is replay-only (no subscriber) and ends after the replay —
+   * with n = N_final that tail is just the run's `finish` frame, so the client
+   * closes the stream. For a LIVE run a paused subscriber is registered; the
+   * caller writes the replay (respecting drain) then calls start() to drain the
+   * pending frames and go live.
    */
   async attach(
     chatId: string,
-    expectLive: boolean,
     anchor: string | undefined,
+    // The client's persisted step frontier. `null` = a NOT-tail-aware client (no
+    // `n` query param) — a legacy/parameterless tab that expects the old
+    // "finished -> 204 -> poll" contract; distinct from `0` (a tail-aware client
+    // with nothing persisted yet).
+    n: number | null,
     cb: RunStreamCallbacks,
   ): Promise<RunStreamAttachment | null> {
     const entry = this.entries.get(chatId);
-    if (!entry || entry.overflowed) return null;
+    if (!entry) return null;
     // Invariant 6: cross-run replay is forbidden. Before bind, assistantMessageId
     // is undefined and mismatches any anchor -> 204 -> client restore+poll path.
-    if (expectLive && anchor && entry.assistantMessageId !== anchor) return null;
-    if (entry.finished && !expectLive) return null;
-    if (entry.finished && expectLive) {
+    if (anchor && entry.assistantMessageId !== anchor) return null;
+    // #491 regression guard (#137/#161 dup): a NOT-tail-aware client (no `n`)
+    // resuming a FINISHED run must 204 and poll — the old `finished && !expectLive`
+    // gate. Without this, a missing `n` collapsing to frontier 0 would serve the
+    // WHOLE tail of a finished, NON-rotated run (coverageFloor 0), and a
+    // parameterless client that never stripped its transcript would APPEND that
+    // full replay onto the steps it already shows -> duplicated text. A tail-aware
+    // client (n present, incl. n=0) still gets the tail past its frontier.
+    if (entry.finished && n === null) return null;
+    // A finished entry with NOTHING in the ring (aborted before the first frame,
+    // or fully overflowed) has no tail to deliver -> 204 -> the client polls.
+    if (entry.finished && entry.frames.length === 0) return null;
+    // A LIVE run with no `n` (legacy parameterless) replays from step 0 (the old
+    // behavior); a tail-aware client resumes from its frontier.
+    const frontier = n ?? 0;
+    const floor = this.coverageFloor(entry);
+    if (floor > frontier) {
+      this.logger.warn(
+        `run-stream attach gap for run=${entry.runId}: coverageFloor=${floor} ` +
+          `> client frontier=${frontier} -> 204 (client refetches + re-attaches)`,
+      );
+      return null;
+    }
+
+    const startFrame = this.buildStartFrame(chatId, entry.runId);
+    const sliceTail = (): string[] => {
+      const out: string[] = [startFrame];
+      for (let i = 0; i < entry.frames.length; i++) {
+        if (entry.stamps[i] >= frontier) out.push(entry.frames[i]);
+      }
+      return out;
+    };
+
+    if (entry.finished) {
       // Replay-only: the run is done, no subscriber is registered.
       return {
-        replay: entry.frames.slice(),
+        replay: sliceTail(),
         finished: true,
         start: () => undefined,
         unsubscribe: () => undefined,
@@ -206,15 +383,12 @@ export class AiChatStreamRegistryService implements OnModuleDestroy {
       pendingBytes: 0,
       overflowed: false,
       pendingEnd: false,
+      minStamp: frontier,
     };
+    // Register + snapshot in the SAME synchronous block (invariant 4). No await
+    // separates them, so a concurrently ingested frame cannot be lost/duplicated.
     entry.subscribers.add(sub);
-    // Snapshot in the SAME synchronous block as the registration (invariant 4).
-    const replay = entry.frames.slice();
-    // CONTRACT: the caller MUST call start() in the SAME tick as this attach()
-    // returns — no await between them. While a subscriber is paused, every frame
-    // is buffered in sub.pending; a delayed start() lets a whole run accumulate
-    // there. The pendingBytes cap (see ingestFrame) is the structural backstop if
-    // that contract is ever broken (e.g. the phase-2 Redis await seam).
+    const replay = sliceTail();
     return {
       replay,
       finished: false,
@@ -263,24 +437,83 @@ export class AiChatStreamRegistryService implements OnModuleDestroy {
     this.entries.clear();
   }
 
-  /** Buffer + fan-out a single frame. See invariant/overflow semantics inline. */
+  /** The synthetic `start` frame the tail is prefixed with — the source of the
+   *  run-fact (runId/chatId) on re-attach. A `start` frame does NOT reset the
+   *  client's message parts (ai@6.0.207 createStreamingUIMessageState), so it is
+   *  safe to prepend even when the sliced tail begins mid-message. */
+  private buildStartFrame(chatId: string, runId: string): string {
+    return `data: ${JSON.stringify({
+      type: 'start',
+      messageMetadata: { runId, chatId },
+    })}\n\n`;
+  }
+
+  /**
+   * The smallest step FULLY present in the ring: its smallest retained stamp, or
+   * (when the leading step was only partially evicted by an overflow) one past it.
+   * When the ring is empty it is the current step (only the live tail is coming).
+   * An attach at frontier `n` is covered ⟺ coverageFloor <= n.
+   */
+  private coverageFloor(entry: Entry): number {
+    // Empty ring: only the live tail is coming. The floor is the current step,
+    // but never below persistedFloor — a confirmed persist can rotate the ring
+    // empty while currentStamp still lags a beat behind on another connection, so
+    // max() keeps the invariant STRUCTURAL (a client with n = persistedFloor is
+    // always covered) rather than timing-dependent.
+    if (entry.frames.length === 0)
+      return Math.max(entry.currentStamp, entry.persistedFloor);
+    const min = entry.stamps[0];
+    return entry.overflowThroughStamp >= min ? min + 1 : min;
+  }
+
+  /**
+   * Buffer (step-stamped) + fan-out a single frame. The stamp is the number of
+   * finish-step frames seen BEFORE this one; a finish-step frame carries the
+   * current value and THEN increments the counter (so its stamp equals the 0-based
+   * index of the step it closes). Only frames at/above persistedFloor are buffered
+   * (already-persisted steps are on disk); the ring is then trimmed to the byte
+   * cap, an unsafe eviction opening a gap. Fan-out is always live (filtered per
+   * subscriber by its frontier).
+   */
   private ingestFrame(entry: Entry, frame: string): void {
-    entry.bytes += Buffer.byteLength(frame);
-    if (!entry.overflowed) {
+    const size = Buffer.byteLength(frame);
+    const stamp = entry.currentStamp;
+    if (frame.startsWith(FINISH_STEP_FRAME_PREFIX)) {
+      entry.currentStamp = stamp + 1;
+    }
+
+    // Buffer for replay only if this step is not already persisted+rotated away.
+    if (stamp >= entry.persistedFloor) {
       entry.frames.push(frame);
-      if (entry.bytes > RUN_STREAM_MAX_BUFFER_BYTES) {
-        // The crossing frame was already counted AND (below) fanned out; only the
-        // replay buffer is dropped. After overflow no more frames are buffered,
-        // but live fan-out continues.
-        entry.overflowed = true;
-        entry.frames = [];
-        this.logger.warn(
-          `run-stream buffer overflow for run=${entry.runId}; ` +
-            `late attach will 204 until the run ends`,
-        );
+      entry.stamps.push(stamp);
+      entry.bytes += size;
+      // Enforce the ring cap. Evicting a not-yet-persisted frame (stamp >=
+      // persistedFloor) opens a GAP; a leftover persisted frame (< floor) is a
+      // safe drop. Keep evicting until the ring is back under the cap.
+      while (entry.bytes > this.maxBufferBytes && entry.frames.length > 0) {
+        const evStamp = entry.stamps[0];
+        entry.bytes -= Buffer.byteLength(entry.frames[0]);
+        entry.frames.shift();
+        entry.stamps.shift();
+        if (evStamp >= entry.persistedFloor) {
+          if (evStamp > entry.overflowThroughStamp)
+            entry.overflowThroughStamp = evStamp;
+          if (!entry.overflowed) {
+            entry.overflowed = true;
+            this.logger.warn(
+              `run-stream ring overflow for run=${entry.runId}: an un-persisted ` +
+                `step was evicted to stay under ${this.maxBufferBytes}B; a late ` +
+                `attach at an evicted step will 204 until a later persist confirms`,
+            );
+          }
+        }
       }
     }
+
+    // Fan out live, filtered to each subscriber's frontier (a subscriber only
+    // wants the tail past the step it already persisted).
     for (const sub of entry.subscribers) {
+      if (stamp < sub.minStamp) continue;
       if (sub.started) {
         try {
           sub.onFrame(frame);
@@ -289,12 +522,12 @@ export class AiChatStreamRegistryService implements OnModuleDestroy {
         }
       } else {
         sub.pending.push(frame);
-        sub.pendingBytes += Buffer.byteLength(frame);
-        if (sub.pendingBytes > SUBSCRIBER_MAX_BUFFERED_BYTES) {
+        sub.pendingBytes += size;
+        if (sub.pendingBytes > this.subscriberMaxBufferedBytes) {
           // The paused subscriber's buffer overflowed — only possible if start()
-          // was delayed past the same-tick contract (the phase-2 await seam).
-          // Drop it rather than buffer the whole run; on start() it degrades to an
-          // immediate end (a 204-equivalent) instead of replaying a partial.
+          // was delayed (the controller's drain-respecting tail write, or the
+          // phase-2 await seam). Drop it rather than buffer the whole run; on
+          // start() it degrades to an immediate end (a 204-equivalent).
           sub.overflowed = true;
           sub.pending = [];
           entry.subscribers.delete(sub);

apps/server/src/core/ai-chat/ai-chat-stream-registry.spec.ts

@@ -1,19 +1,27 @@
 import {
   AiChatStreamRegistryService,
-  RUN_STREAM_MAX_BUFFER_BYTES,
+  AI_CHAT_RUN_STREAM_MAX_BUFFER_BYTES,
   RUN_STREAM_RETAIN_FINISHED_MS,
-  SUBSCRIBER_MAX_BUFFERED_BYTES,
   RunStreamCallbacks,
 } from './ai-chat-stream-registry.service';
 
 /**
- * Unit tests for the in-memory run-stream registry (#184 phase 1.5). The registry
- * is the whole of the resumable-transport contract: replay ordering, paused ->
- * live hand-off, overflow, retention, the anchor check (invariant 6), and the
- * mirror-the-done-path replace semantics (invariant 3). Every enumerated case in
- * the issue's task 1.5 has a test here.
+ * Unit tests for the in-memory run-stream registry (#184 phase 1.5, step-aligned
+ * retention #491). The registry is the whole of the resumable-transport contract:
+ * step-stamped retention, tail-only attach at the client's frontier N, the
+ * confirmed-persist ring rotation (and the anti-inversion rule), the memory bound,
+ * the overflow gap, paused -> live hand-off, retention, the anchor check
+ * (invariant 6), and the mirror-the-done-path replace semantics (invariant 3).
  */
 
+// Real ai@6 UI-message-stream SSE frames are `data: {json}\n\n`, one part each.
+const sse = (part: Record<string, unknown>): string =>
+  `data: ${JSON.stringify(part)}\n\n`;
+const finishStep = (): string => sse({ type: 'finish-step' });
+const textDelta = (id: string, delta: string): string =>
+  sse({ type: 'text-delta', id, delta });
+const finish = (): string => sse({ type: 'finish' });
+
 // A ReadableStream whose frames the test pushes explicitly, plus close/error.
 function makePushStream(): {
   stream: ReadableStream<string>;
@@ -58,6 +66,9 @@ function collector(): {
   };
 }
 
+// The tail past the synthetic start frame (replay[0] is always the start frame).
+const tail = (replay: string[]): string[] => replay.slice(1);
+
 describe('AiChatStreamRegistryService', () => {
   const CHAT = 'chat-1';
   let registry: AiChatStreamRegistryService;
@@ -71,7 +82,21 @@ describe('AiChatStreamRegistryService', () => {
     registry.onModuleDestroy();
   });
 
-  it('replays frames in arrival order (live attach)', async () => {
+  it('prepends a synthetic start frame carrying { runId, chatId }', async () => {
+    registry.open(CHAT, 'run-1');
+    const src = makePushStream();
+    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+    src.push('a');
+    await flush();
+
+    const c = collector();
+    const att = (await registry.attach(CHAT, 'assist-1', 0, c.cb))!;
+    const start = JSON.parse(att.replay[0].replace(/^data: /, '').trim());
+    expect(start.type).toBe('start');
+    expect(start.messageMetadata).toEqual({ runId: 'run-1', chatId: CHAT });
+  });
+
+  it('replays the buffered tail (from frontier 0) in arrival order (live attach)', async () => {
     registry.open(CHAT, 'run-1');
     const src = makePushStream();
     registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
@@ -81,13 +106,13 @@ describe('AiChatStreamRegistryService', () => {
     await flush();
 
     const c = collector();
-    const att = await registry.attach(CHAT, false, undefined, c.cb);
+    const att = await registry.attach(CHAT, 'assist-1', 0, c.cb);
     expect(att).not.toBeNull();
-    expect(att!.replay).toEqual(['a', 'b', 'c']);
+    expect(tail(att!.replay)).toEqual(['a', 'b', 'c']);
     expect(att!.finished).toBe(false);
   });
 
-  it('late attach gets the full prefix as replay plus the live tail', async () => {
+  it('late attach gets the buffered prefix as tail plus the live tail', async () => {
     registry.open(CHAT, 'run-1');
     const src = makePushStream();
     registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
@@ -96,17 +121,16 @@ describe('AiChatStreamRegistryService', () => {
     await flush();
 
     const c = collector();
-    const att = (await registry.attach(CHAT, false, undefined, c.cb))!;
-    expect(att.replay).toEqual(['a', 'b']);
+    const att = (await registry.attach(CHAT, 'assist-1', 0, c.cb))!;
+    expect(tail(att.replay)).toEqual(['a', 'b']);
     att.start();
-    // Live tail arrives after start().
     src.push('c');
     src.push('d');
     await flush();
     expect(c.frames).toEqual(['c', 'd']);
   });
 
-  it('a paused subscriber receives frames buffered during pause in order, then live (no loss/reorder)', async () => {
+  it('a paused subscriber receives frames buffered during pause in order, then live', async () => {
     registry.open(CHAT, 'run-1');
     const src = makePushStream();
     registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
@@ -114,81 +138,45 @@ describe('AiChatStreamRegistryService', () => {
     await flush();
 
     const c = collector();
-    // Attach (paused). Frames that arrive BEFORE start() must queue, not drop.
-    const att = (await registry.attach(CHAT, false, undefined, c.cb))!;
-    expect(att.replay).toEqual(['a']);
+    const att = (await registry.attach(CHAT, 'assist-1', 0, c.cb))!;
+    expect(tail(att.replay)).toEqual(['a']);
     src.push('b'); // arrives while paused -> pending
     src.push('c');
     await flush();
     expect(c.frames).toEqual([]); // nothing delivered yet (paused)
-    att.start(); // drains pending in order
+    att.start();
     expect(c.frames).toEqual(['b', 'c']);
-    src.push('d'); // now live
+    src.push('d');
     await flush();
     expect(c.frames).toEqual(['b', 'c', 'd']);
   });
 
   it('a run that finishes while a subscriber is paused ends it on start()', async () => {
     registry.open(CHAT, 'run-1');
+    registry.bind(CHAT, 'run-1', 'assist-1', makePushStream().stream);
     const c = collector();
-    const att = (await registry.attach(CHAT, false, undefined, c.cb))!;
-    // Terminate the run while the subscriber is still paused.
+    const att = (await registry.attach(CHAT, 'assist-1', 0, c.cb))!;
     registry.abortEntry(CHAT, 'run-1');
     expect(c.ended()).toBe(0); // paused: not ended yet
     att.start();
     expect(c.ended()).toBe(1); // start() drains + ends
   });
 
-  it('finished + expect=live returns a replay WITHOUT registering a subscriber', async () => {
-    registry.open(CHAT, 'run-1');
-    const src = makePushStream();
-    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
-    src.push('a');
-    src.push('b');
-    src.close();
-    await flush();
-
-    const c = collector();
-    const att = (await registry.attach(CHAT, true, undefined, c.cb))!;
-    expect(att.finished).toBe(true);
-    expect(att.replay).toEqual(['a', 'b']);
-    // No subscriber registered: start()/unsubscribe are no-ops and the entry has
-    // zero subscribers.
-    const entry = (registry as any).entries.get(CHAT);
-    expect(entry.subscribers.size).toBe(0);
-    att.start();
-    expect(c.frames).toEqual([]);
-  });
-
-  it('finished WITHOUT expect=live returns null', async () => {
-    registry.open(CHAT, 'run-1');
-    const src = makePushStream();
-    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
-    src.push('a');
-    src.close();
-    await flush();
-
-    const c = collector();
-    expect(await registry.attach(CHAT, false, undefined, c.cb)).toBeNull();
-  });
-
-  it('anchor mismatch with expect=live returns null (and null before bind sets assistantMessageId)', async () => {
+  it('anchor mismatch returns null (and null before bind sets assistantMessageId)', async () => {
     registry.open(CHAT, 'run-1');
     const c = collector();
     // Before bind: assistantMessageId is undefined -> mismatches any anchor.
-    expect(
-      await registry.attach(CHAT, true, 'assist-1', c.cb),
-    ).toBeNull();
+    expect(await registry.attach(CHAT, 'assist-1', 0, c.cb)).toBeNull();
 
     const src = makePushStream();
     registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
     src.push('a');
     await flush();
     // Wrong anchor -> null (cross-run replay forbidden, invariant 6).
-    expect(await registry.attach(CHAT, true, 'other-id', c.cb)).toBeNull();
+    expect(await registry.attach(CHAT, 'other-id', 0, c.cb)).toBeNull();
   });
 
-  it('matching anchor with expect=live attaches', async () => {
+  it('matching anchor attaches', async () => {
     registry.open(CHAT, 'run-1');
     const src = makePushStream();
     registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
@@ -196,97 +184,60 @@ describe('AiChatStreamRegistryService', () => {
     await flush();
 
     const c = collector();
-    const att = await registry.attach(CHAT, true, 'assist-1', c.cb);
+    const att = await registry.attach(CHAT, 'assist-1', 0, c.cb);
     expect(att).not.toBeNull();
-    expect(att!.replay).toEqual(['a']);
+    expect(tail(att!.replay)).toEqual(['a']);
   });
 
-  it('overflow: attach returns null, but the LIVE subscriber keeps receiving (incl. the crossing frame)', async () => {
+  it('a throwing onFrame ejects only that subscriber; the ingest loop stays alive', async () => {
     registry.open(CHAT, 'run-1');
     const src = makePushStream();
     registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
 
-    // A live (started) subscriber attached before the flood.
+    const bad = collector();
+    const badAtt = (await registry.attach(CHAT, 'assist-1', 0, {
+      onFrame: () => {
+        throw new Error('boom');
+      },
+      onEnd: bad.cb.onEnd,
+    }))!;
+    badAtt.start();
+
+    const good = collector();
+    const goodAtt = (await registry.attach(CHAT, 'assist-1', 0, good.cb))!;
+    goodAtt.start();
+
+    src.push('a'); // bad throws on this frame -> ejected
+    src.push('b'); // good still receives both
+    await flush();
+
+    const entry = (registry as any).entries.get(CHAT);
+    expect(entry.subscribers.size).toBe(1);
+    expect(good.frames).toEqual(['a', 'b']);
+  });
+
+  it('open() over a LIVE entry ends started subscribers once; a late done never touches the new entry (invariant 3)', async () => {
+    registry.open(CHAT, 'run-1');
+    const src = makePushStream();
+    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+    src.push('a');
+    await flush();
+
     const c = collector();
-    const att = (await registry.attach(CHAT, false, undefined, c.cb))!;
+    const att = (await registry.attach(CHAT, 'assist-1', 0, c.cb))!;
     att.start();
 
-    // Cap-relative so it survives a buffer-cap change (#430): a quarter-cap frame
-    // means 5 frames comfortably exceed the replay cap; the last one crosses.
-    const chunk = 'x'.repeat(Math.floor(RUN_STREAM_MAX_BUFFER_BYTES / 4));
-    for (let i = 0; i < 5; i++) src.push(chunk + i);
-    await flush();
-
-    const entry = (registry as any).entries.get(CHAT);
-    expect(entry.overflowed).toBe(true);
-    expect(entry.bytes).toBeGreaterThan(RUN_STREAM_MAX_BUFFER_BYTES);
-    // The live subscriber received ALL 5 frames, including the crossing one.
-    expect(c.frames).toHaveLength(5);
-    expect(c.frames[4]).toBe(chunk + 4);
-
-    // A NEW attach after overflow gets null (replay buffer is gone).
-    const c2 = collector();
-    expect(await registry.attach(CHAT, false, undefined, c2.cb)).toBeNull();
-  });
-
-  it('a paused subscriber whose pending buffer overflows is dropped and ends on start(); other subscribers keep receiving', async () => {
-    registry.open(CHAT, 'run-1');
-    const src = makePushStream();
-    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
-
-    // A: paused (start() deliberately delayed to simulate the phase-2 await seam).
-    const a = collector();
-    const attA = (await registry.attach(CHAT, false, undefined, a.cb))!;
-    // B: live (started) — its delivery must be unaffected by A's overflow.
-    const b = collector();
-    const attB = (await registry.attach(CHAT, false, undefined, b.cb))!;
-    attB.start();
-
-    // Cap-relative so it survives a buffer-cap change (#430): a quarter-of-the-
-    // per-subscriber-cap frame means 5 frames exceed A's paused-pending cap while
-    // B streams every frame live.
-    const chunk = 'x'.repeat(Math.floor(SUBSCRIBER_MAX_BUFFERED_BYTES / 4));
-    for (let i = 0; i < 5; i++) src.push(chunk + i);
-    await flush();
-
-    const entry = (registry as any).entries.get(CHAT);
-    // A was dropped from the subscriber set on overflow; B (started) remains.
-    expect(entry.subscribers.size).toBe(1);
-    expect(a.frames).toEqual([]); // paused + overflowed: nothing was delivered
-    // B received every frame live (delivery unaffected by A's overflow).
-    expect(b.frames).toHaveLength(5);
-
-    // A's start() (arriving late) degrades to an immediate end, not a partial replay.
-    attA.start();
-    expect(a.frames).toEqual([]);
-    expect(a.ended()).toBe(1);
-  });
-
-  it('open() over a LIVE entry ends started subscribers exactly once and a late done does not touch the new entry (invariant 3)', async () => {
-    registry.open(CHAT, 'run-1');
-    const src = makePushStream();
-    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
-    src.push('a');
-    await flush();
-
-    const c = collector();
-    const att = (await registry.attach(CHAT, false, undefined, c.cb))!;
-    att.start(); // started subscriber on run-1
-
-    // run-2 starts on the same chat while run-1's tee is still reading.
     registry.open(CHAT, 'run-2');
-    expect(c.ended()).toBe(1); // exactly one onEnd from the replace
+    expect(c.ended()).toBe(1);
 
     const newEntry = (registry as any).entries.get(CHAT);
     expect(newEntry.runId).toBe('run-2');
     expect(newEntry.finished).toBe(false);
 
-    // The old tee now completes: its late done must NOT double-end nor delete the
-    // new entry.
     src.push('b');
     src.close();
     await flush();
-    expect(c.ended()).toBe(1); // still exactly one
+    expect(c.ended()).toBe(1);
     const still = (registry as any).entries.get(CHAT);
     expect(still).toBe(newEntry);
     expect(still.runId).toBe('run-2');
@@ -299,7 +250,6 @@ describe('AiChatStreamRegistryService', () => {
     src.push('a');
     await flush();
     const entry = (registry as any).entries.get(CHAT);
-    // Frames were NOT ingested (bind bailed), assistantMessageId untouched.
     expect(entry.frames).toEqual([]);
     expect(entry.assistantMessageId).toBeUndefined();
   });
@@ -310,32 +260,276 @@ describe('AiChatStreamRegistryService', () => {
     const entry = (registry as any).entries.get(CHAT);
     expect(entry.finished).toBe(false);
   });
+});
 
-  it('a throwing onFrame ejects only that subscriber; the ingest loop stays alive', async () => {
+/**
+ * #491 step-stamped retention: the boundary detector, tail-only slicing at the
+ * client's frontier N, the confirmed-persist rotation (+ anti-inversion), the
+ * overflow gap, the memory bound, and the finished-retained tail. All observable
+ * against the REAL registry driven through open/bind/ingest.
+ */
+describe('AiChatStreamRegistryService step-aligned retention (#491)', () => {
+  const CHAT = 'chat-s';
+  let registry: AiChatStreamRegistryService;
+
+  beforeEach(() => {
+    registry = new AiChatStreamRegistryService();
+    jest.spyOn((registry as any).logger, 'warn').mockImplementation(() => {});
+  });
+  afterEach(() => registry.onModuleDestroy());
+
+  const entryOf = () => (registry as any).entries.get(CHAT);
+
+  it('stamps frames by finish-step count, aligned with stepsPersisted', async () => {
     registry.open(CHAT, 'run-1');
     const src = makePushStream();
     registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+    // step 0 content, its finish-step, step 1 content, its finish-step, finish.
+    src.push(textDelta('t0', 'a')); // stamp 0
+    src.push(finishStep()); // stamp 0 (the finish-step frame carries the pre value)
+    src.push(textDelta('t1', 'b')); // stamp 1
+    src.push(finishStep()); // stamp 1
+    src.push(finish()); // stamp 2
+    await flush();
+    const e = entryOf();
+    expect(e.stamps).toEqual([0, 0, 1, 1, 2]);
+    expect(e.currentStamp).toBe(2);
+  });
 
-    const bad = collector();
-    const badAtt = (await registry.attach(CHAT, false, undefined, {
-      onFrame: () => {
-        throw new Error('boom');
-      },
-      onEnd: bad.cb.onEnd,
-    }))!;
-    badAtt.start();
+  it('does NOT treat a text delta that merely quotes "finish-step" as a boundary', async () => {
+    registry.open(CHAT, 'run-1');
+    const src = makePushStream();
+    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+    // A model that literally types "type":"finish-step" — JSON-escaped in the frame.
+    src.push(textDelta('t0', '"type":"finish-step"'));
+    await flush();
+    expect(entryOf().currentStamp).toBe(0); // no false boundary
+  });
 
-    const good = collector();
-    const goodAtt = (await registry.attach(CHAT, false, undefined, good.cb))!;
-    goodAtt.start();
-
-    src.push('a'); // bad throws on this frame -> ejected
-    src.push('b'); // good still receives both
+  it('tail-only: attach at N slices frames with stamp >= N', async () => {
+    registry.open(CHAT, 'run-1');
+    const src = makePushStream();
+    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+    src.push(textDelta('t0', 'a')); // 0
+    src.push(finishStep()); // 0
+    src.push(textDelta('t1', 'b')); // 1
+    src.push(finishStep()); // 1
+    src.push(textDelta('t2', 'c')); // 2 (in-progress)
     await flush();
 
-    const entry = (registry as any).entries.get(CHAT);
-    expect(entry.subscribers.size).toBe(1); // bad ejected, good remains
-    expect(good.frames).toEqual(['a', 'b']);
+    const c = collector();
+    // Client persisted 2 steps -> wants the tail from step 2.
+    const att = (await registry.attach(CHAT, 'assist-1', 2, c.cb))!;
+    expect(tail(att.replay)).toEqual([textDelta('t2', 'c')]);
+  });
+
+  it('attach in the MIDDLE of a step (N between finish-steps) slices from that step', async () => {
+    registry.open(CHAT, 'run-1');
+    const src = makePushStream();
+    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+    src.push(textDelta('t0', 'a')); // 0
+    src.push(finishStep()); // 0
+    src.push(textDelta('t1', 'b1')); // 1
+    src.push(textDelta('t1', 'b2')); // 1 (still step 1, no finish-step yet)
+    await flush();
+
+    const c = collector();
+    const att = (await registry.attach(CHAT, 'assist-1', 1, c.cb))!;
+    // Step 0's frames are dropped from the tail; the whole in-progress step 1 is kept.
+    expect(tail(att.replay)).toEqual([textDelta('t1', 'b1'), textDelta('t1', 'b2')]);
+  });
+
+  it('rotates the ring ONLY on a confirmed persist (drops stamp < N)', async () => {
+    registry.open(CHAT, 'run-1');
+    const src = makePushStream();
+    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+    src.push(textDelta('t0', 'a')); // 0
+    src.push(finishStep()); // 0
+    src.push(textDelta('t1', 'b')); // 1
+    await flush();
+    expect(entryOf().stamps).toEqual([0, 0, 1]);
+
+    // Confirm step 0 persisted (stepsPersisted = 1) -> drop stamp < 1.
+    registry.confirmPersistedStep(CHAT, 'run-1', 1);
+    expect(entryOf().stamps).toEqual([1]);
+    expect(entryOf().persistedFloor).toBe(1);
+  });
+
+  it('persist FAILED but the ring still fits -> attach SUCCEEDS and the tail includes step N', async () => {
+    registry.open(CHAT, 'run-1');
+    const src = makePushStream();
+    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+    src.push(textDelta('t0', 'a')); // 0
+    src.push(finishStep()); // 0
+    src.push(textDelta('t1', 'b')); // 1 (step 1's persist FAILED -> no confirm)
+    await flush();
+    // No confirmPersistedStep for step 1: the ring still holds step 1.
+
+    const c = collector();
+    // Client's last successful persist was step 0 -> stepsPersisted = 1.
+    const att = await registry.attach(CHAT, 'assist-1', 1, c.cb);
+    expect(att).not.toBeNull();
+    expect(tail(att!.replay)).toEqual([textDelta('t1', 'b')]); // includes step 1
+  });
+
+  it('persist failed AND the ring overflowed past N -> 204 (coverage gap)', async () => {
+    registry.open(CHAT, 'run-1');
+    const src = makePushStream();
+    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+    // Step 0: a fat step that blows past the cap with NO persist confirmation.
+    const big = 'x'.repeat(Math.floor(AI_CHAT_RUN_STREAM_MAX_BUFFER_BYTES / 2));
+    src.push(textDelta('t0', big)); // 0
+    src.push(textDelta('t0', big)); // 0
+    src.push(textDelta('t0', big)); // 0 -> overflow evicts stamp-0 frames
+    await flush();
+    const e = entryOf();
+    expect(e.overflowed).toBe(true);
+    expect(e.bytes).toBeLessThanOrEqual(registry.maxBufferBytes);
+
+    // A client at frontier 0 falls at/below an evicted step -> gap -> null.
+    const c = collector();
+    expect(await registry.attach(CHAT, 'assist-1', 0, c.cb)).toBeNull();
+  });
+
+  it('stale N (client seed lagged behind a rotation) -> 204; after a refetch (larger N) -> success', async () => {
+    registry.open(CHAT, 'run-1');
+    const src = makePushStream();
+    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+    src.push(textDelta('t0', 'a')); // 0
+    src.push(finishStep()); // 0
+    src.push(textDelta('t1', 'b')); // 1
+    src.push(finishStep()); // 1
+    src.push(textDelta('t2', 'c')); // 2
+    await flush();
+    // Server confirmed steps 0 and 1 -> rotate away stamp < 2.
+    registry.confirmPersistedStep(CHAT, 'run-1', 2);
+    expect(entryOf().stamps).toEqual([2]);
+
+    // A client whose seed still says stepsPersisted = 1 -> below minStamp -> 204.
+    const stale = collector();
+    expect(await registry.attach(CHAT, 'assist-1', 1, stale.cb)).toBeNull();
+
+    // It refetches (now stepsPersisted = 2) and re-attaches -> success.
+    const fresh = collector();
+    const att = await registry.attach(CHAT, 'assist-1', 2, fresh.cb);
+    expect(att).not.toBeNull();
+    expect(tail(att!.replay)).toEqual([textDelta('t2', 'c')]);
+  });
+
+  it('overflow gap CLEARS once a later persist rotates out the holey steps', async () => {
+    registry.open(CHAT, 'run-1');
+    const src = makePushStream();
+    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+    const big = 'x'.repeat(Math.floor(AI_CHAT_RUN_STREAM_MAX_BUFFER_BYTES / 2));
+    src.push(textDelta('t0', big)); // 0
+    src.push(textDelta('t0', big)); // 0
+    src.push(finishStep()); // 0 (still stamp 0)
+    src.push(textDelta('t1', 'small')); // 1
+    src.push(finishStep()); // 1
+    src.push(textDelta('t2', 'c')); // 2
+    await flush();
+    expect(entryOf().overflowed).toBe(true);
+
+    // Late persist confirms steps 0..1 -> rotates out the holey step-0 frames.
+    registry.confirmPersistedStep(CHAT, 'run-1', 2);
+    // A client at frontier 2 is now cleanly covered (the hole was below it).
+    const c = collector();
+    const att = await registry.attach(CHAT, 'assist-1', 2, c.cb);
+    expect(att).not.toBeNull();
+    expect(tail(att!.replay)).toEqual([textDelta('t2', 'c')]);
+  });
+
+  it('finished-retained + N = N_final -> empty tail plus the finish frame', async () => {
+    registry.open(CHAT, 'run-1');
+    const src = makePushStream();
+    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+    src.push(textDelta('t0', 'a')); // 0
+    src.push(finishStep()); // 0
+    src.push(finish()); // 1  (N_final = 1)
+    src.close();
+    await flush();
+    // The last step's per-step persist confirmed stepsPersisted = 1.
+    registry.confirmPersistedStep(CHAT, 'run-1', 1);
+
+    const c = collector();
+    const att = (await registry.attach(CHAT, 'assist-1', 1, c.cb))!;
+    expect(att.finished).toBe(true);
+    // Empty step tail; just the finish frame so the client's SDK closes the stream.
+    expect(tail(att.replay)).toEqual([finish()]);
+    // No subscriber registered for a finished run.
+    expect(entryOf().subscribers.size).toBe(0);
+  });
+
+  it('#491 regression (#137/#161 dup): a PARAMETERLESS attach (n=null) to a finished NON-rotated run -> 204, but n=0 still gets the tail', async () => {
+    // A finished, non-rotated run: frames present, coverageFloor 0. A missing `n`
+    // (null — a legacy/parameterless tab that never stripped its transcript) must
+    // 204 -> poll, NOT receive the whole tail it would append (duplicate). A
+    // tail-aware client (n=0 present) still resumes.
+    registry.open(CHAT, 'run-1');
+    const src = makePushStream();
+    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+    src.push(textDelta('t0', 'a')); // 0
+    src.push(finishStep()); // 0
+    src.push(finish()); // 1
+    src.close();
+    await flush();
+    // NOT rotated (no confirmPersistedStep) -> stamps[0]=0, coverageFloor=0.
+    // MUTATION-VERIFY: revert the `finished && n === null -> null` gate (default n
+    // to 0) and the parameterless attach below serves the full tail instead of 204.
+    expect(await registry.attach(CHAT, 'assist-1', null, collector().cb)).toBeNull();
+    // A tail-aware client at frontier 0 IS served (the distinction: null != 0).
+    const tailAware = await registry.attach(CHAT, 'assist-1', 0, collector().cb);
+    expect(tailAware).not.toBeNull();
+    expect(tailAware!.finished).toBe(true);
+  });
+
+  it('confirmPersistedStep is monotonic and identity-checked', async () => {
+    registry.open(CHAT, 'run-1');
+    const src = makePushStream();
+    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+    src.push(textDelta('t0', 'a'));
+    src.push(finishStep());
+    src.push(textDelta('t1', 'b'));
+    await flush();
+    registry.confirmPersistedStep(CHAT, 'run-1', 1);
+    expect(entryOf().persistedFloor).toBe(1);
+    // A stale lower count is ignored.
+    registry.confirmPersistedStep(CHAT, 'run-1', 0);
+    expect(entryOf().persistedFloor).toBe(1);
+    // A foreign runId is ignored.
+    registry.confirmPersistedStep(CHAT, 'WRONG', 5);
+    expect(entryOf().persistedFloor).toBe(1);
+  });
+
+  it('MEMORY BOUND: 5 parallel marathon runs each stream well past 32MB; each ring stays <= the cap', async () => {
+    const cap = registry.maxBufferBytes;
+    const chats = ['m0', 'm1', 'm2', 'm3', 'm4'];
+    const srcs = chats.map((chat) => {
+      registry.open(chat, `run-${chat}`);
+      const s = makePushStream();
+      registry.bind(chat, `run-${chat}`, `assist-${chat}`, s.stream);
+      return s;
+    });
+    // ~256KB frames; 160 per chat = 40MB streamed each, well past the old 32MB.
+    // Interleave a finish-step every 8 frames so steps advance realistically. No
+    // persist confirmation -> the ONLY thing keeping memory bounded is the cap.
+    const frame = 'y'.repeat(256 * 1024);
+    for (let batch = 0; batch < 20; batch++) {
+      for (let i = 0; i < 8; i++) {
+        for (const s of srcs) s.push(textDelta('t', frame));
+      }
+      for (const s of srcs) s.push(finishStep());
+      await flush(); // drain the pump so queues never hold a whole run
+    }
+    let total = 0;
+    for (const chat of chats) {
+      const e = (registry as any).entries.get(chat);
+      expect(e.bytes).toBeLessThanOrEqual(cap);
+      total += e.bytes;
+    }
+    // Total retained across all 5 runs is bounded by 5x the per-run cap — the old
+    // registry would have retained ~5x40MB = 200MB here.
+    expect(total).toBeLessThanOrEqual(cap * chats.length);
   });
 });
 
@@ -361,7 +555,7 @@ describe('AiChatStreamRegistryService retention timers', () => {
 
   it('a finished entry is removed after the retention window', () => {
     registry.open(CHAT, 'run-1');
-    registry.abortEntry(CHAT, 'run-1'); // finalize -> retention armed
+    registry.abortEntry(CHAT, 'run-1');
     expect((registry as any).entries.get(CHAT)).toBeDefined();
     jest.advanceTimersByTime(RUN_STREAM_RETAIN_FINISHED_MS + 1);
     expect((registry as any).entries.get(CHAT)).toBeUndefined();
@@ -369,20 +563,18 @@ describe('AiChatStreamRegistryService retention timers', () => {
 
   it('retention deletes ONLY its own entry (invariant 2)', () => {
     registry.open(CHAT, 'run-1');
-    registry.abortEntry(CHAT, 'run-1'); // arm retention for entry A
-    // Simulate the race where the key was replaced without clearing A's timer.
+    registry.abortEntry(CHAT, 'run-1');
     const sentinel = { marker: true };
     (registry as any).entries.set(CHAT, sentinel);
     jest.advanceTimersByTime(RUN_STREAM_RETAIN_FINISHED_MS + 1);
-    // A's timer saw entries.get(CHAT) !== A, so it did NOT delete the successor.
     expect((registry as any).entries.get(CHAT)).toBe(sentinel);
   });
 
   it('open() over a retained entry clears its timer and the successor survives', () => {
     registry.open(CHAT, 'run-1');
-    registry.abortEntry(CHAT, 'run-1'); // retained, timer armed
+    registry.abortEntry(CHAT, 'run-1');
     const clearSpy = jest.spyOn(global, 'clearTimeout');
-    registry.open(CHAT, 'run-2'); // must clear run-1's retain timer
+    registry.open(CHAT, 'run-2');
     expect(clearSpy).toHaveBeenCalled();
     jest.advanceTimersByTime(RUN_STREAM_RETAIN_FINISHED_MS + 1);
     const entry = (registry as any).entries.get(CHAT);

apps/server/src/core/ai-chat/ai-chat.controller.attach.spec.ts

@@ -8,10 +8,12 @@ import { SUBSCRIBER_MAX_BUFFERED_BYTES } from './ai-chat-stream-registry.service
 import type { User, Workspace } from '@docmost/db/types/entity.types';
 
 /**
- * Wiring spec for the #184 phase 1.5 attach endpoint
+ * Wiring spec for the #184 phase 1.5 attach endpoint (tail-only #491)
  * (`GET /ai-chat/runs/:chatId/stream`). Owner-gated via assertOwnedChat; the
- * registry is mocked so this exercises ONLY the controller's replay/live/204/
- * cleanup wiring against a fake raw socket. Constructor order is (aiChatService,
+ * registry is mocked so this exercises ONLY the controller's tail-write/live/204/
+ * cleanup wiring against a fake raw socket. The attach signature is now
+ * `(chatId, anchor, n, cb)` — the client hands its persisted step frontier `n`
+ * and its assistant row id `anchor`. Constructor order is (aiChatService,
  * aiChatRunService, aiChatRepo, aiChatMessageRepo, aiTranscription, pageRepo,
  * streamRegistry, environment).
  */
@@ -86,8 +88,8 @@ describe('AiChatController attach endpoint (#184 phase 1.5)', () => {
       attach: jest.fn(
         (
           _chatId: string,
-          _live: boolean,
           _anchor: string | undefined,
+          _n: number,
           cb: RunStreamCallbacks,
         ) => {
           capturedCb = cb;
@@ -156,7 +158,7 @@ describe('AiChatController attach endpoint (#184 phase 1.5)', () => {
     expect(res.hijack).not.toHaveBeenCalled();
   });
 
-  it('threads expect=live and anchor through to the registry', async () => {
+  it('threads anchor and the numeric frontier n through to the registry', async () => {
     const { controller, streamRegistry } = makeController({
       chat: owned,
       attachment: null,
@@ -165,8 +167,8 @@ describe('AiChatController attach endpoint (#184 phase 1.5)', () => {
     const { req } = makeReq();
     await controller.attachRunStream(
       'c1',
-      'live',
       'anchor-1',
+      '2',
       req,
       res,
       user,
@@ -174,13 +176,44 @@ describe('AiChatController attach endpoint (#184 phase 1.5)', () => {
     );
     expect(streamRegistry.attach).toHaveBeenCalledWith(
       'c1',
-      true,
       'anchor-1',
+      2, // parsed to a number
       expect.anything(),
     );
   });
 
-  it('passes expect=false when the query is absent', async () => {
+  it('#491: an ABSENT/invalid n passes null (not 0) so a finished run 204s (not-tail-aware)', async () => {
+    // Distinguishing a MISSING `n` from `n=0` is the #137/#161 dup guard: a
+    // parameterless/legacy tab must be handed null (-> the registry 204s a finished
+    // run) rather than frontier 0 (which would serve a finished non-rotated run's
+    // whole tail). MUTATION-VERIFY: revert to `Number(n) || 0` and this asserts 0.
+    const { controller, streamRegistry } = makeController({
+      chat: owned,
+      attachment: null,
+    });
+    for (const bad of [undefined, '', 'abc']) {
+      streamRegistry.attach.mockClear();
+      const { res } = makeRawRes();
+      const { req } = makeReq();
+      await controller.attachRunStream(
+        'c1',
+        undefined,
+        bad,
+        req,
+        res,
+        user,
+        workspace,
+      );
+      expect(streamRegistry.attach).toHaveBeenCalledWith(
+        'c1',
+        undefined,
+        null,
+        expect.anything(),
+      );
+    }
+  });
+
+  it('#491: a PRESENT n=0 passes 0 (tail-aware, distinct from absent)', async () => {
     const { controller, streamRegistry } = makeController({
       chat: owned,
       attachment: null,
@@ -190,7 +223,7 @@ describe('AiChatController attach endpoint (#184 phase 1.5)', () => {
     await controller.attachRunStream(
       'c1',
       undefined,
-      undefined,
+      '0',
       req,
       res,
       user,
@@ -198,8 +231,8 @@ describe('AiChatController attach endpoint (#184 phase 1.5)', () => {
     );
     expect(streamRegistry.attach).toHaveBeenCalledWith(
       'c1',
-      false,
       undefined,
+      0,
       expect.anything(),
     );
   });
@@ -245,8 +278,8 @@ describe('AiChatController attach endpoint (#184 phase 1.5)', () => {
     const { req } = makeReq();
     await controller.attachRunStream(
       'c1',
-      'live',
       'a1',
+      '1',
       req,
       res,
       user,

apps/server/src/core/ai-chat/ai-chat.controller.delta.spec.ts

@@ -0,0 +1,108 @@
+import { ForbiddenException } from '@nestjs/common';
+import { AiChatController } from './ai-chat.controller';
+import type { User, Workspace } from '@docmost/db/types/entity.types';
+
+/**
+ * Wiring spec for the #491 delta-poll endpoint (`POST /ai-chat/messages/delta`).
+ * Owner-gated via assertOwnedChat (same gate as the other reads), NOT flag-gated.
+ * The run fact rides IN the delta response (no separate /run poll). Hand-rolled
+ * mocks — no Nest graph, no DB. Constructor order: (aiChatService,
+ * aiChatRunService, aiChatRepo, aiChatMessageRepo, aiTranscription, pageRepo).
+ */
+describe('AiChatController POST /ai-chat/messages/delta (#491)', () => {
+  const user = { id: 'u1' } as User;
+  const workspace = { id: 'ws1' } as Workspace;
+
+  function makeController(opts: {
+    chat?: unknown;
+    delta?: { rows: unknown[]; cursor: string };
+    run?: unknown;
+  }) {
+    const aiChatRunService = {
+      getLatestForChat: jest.fn().mockResolvedValue(opts.run),
+    };
+    const aiChatRepo = {
+      findById: jest.fn().mockResolvedValue(opts.chat),
+    };
+    const aiChatMessageRepo = {
+      findByChatUpdatedAfter: jest
+        .fn()
+        .mockResolvedValue(opts.delta ?? { rows: [], cursor: 'C1' }),
+    };
+    const controller = new AiChatController(
+      {} as never,
+      aiChatRunService as never,
+      aiChatRepo as never,
+      aiChatMessageRepo as never,
+      {} as never,
+      {} as never,
+    );
+    return { controller, aiChatRunService, aiChatRepo, aiChatMessageRepo };
+  }
+
+  it('owner-gates: a chat the user does not own throws, never reaching the repo', async () => {
+    const { controller, aiChatMessageRepo, aiChatRunService } = makeController({
+      chat: { id: 'c1', creatorId: 'someone-else' },
+    });
+    await expect(
+      controller.getMessagesDelta({ chatId: 'c1' }, user, workspace),
+    ).rejects.toBeInstanceOf(ForbiddenException);
+    expect(aiChatMessageRepo.findByChatUpdatedAfter).not.toHaveBeenCalled();
+    expect(aiChatRunService.getLatestForChat).not.toHaveBeenCalled();
+  });
+
+  it('returns { rows, cursor, run:{id,status} } with the run fact inlined', async () => {
+    const rows = [{ id: 'm1' }];
+    const { controller } = makeController({
+      chat: { id: 'c1', creatorId: 'u1' },
+      delta: { rows, cursor: 'C2' },
+      run: { id: 'r1', status: 'running', error: 'ignored', stepCount: 3 },
+    });
+    const res = await controller.getMessagesDelta(
+      { chatId: 'c1', cursor: 'C1' },
+      user,
+      workspace,
+    );
+    expect(res).toEqual({
+      rows,
+      cursor: 'C2',
+      // ONLY id + status — never the whole run row.
+      run: { id: 'r1', status: 'running' },
+    });
+  });
+
+  it('run is null when the chat has never had a run', async () => {
+    const { controller } = makeController({
+      chat: { id: 'c1', creatorId: 'u1' },
+      run: undefined,
+    });
+    const res = await controller.getMessagesDelta(
+      { chatId: 'c1' },
+      user,
+      workspace,
+    );
+    expect(res.run).toBeNull();
+  });
+
+  it('passes cursor through, defaulting a missing cursor to null (first poll)', async () => {
+    const { controller, aiChatMessageRepo } = makeController({
+      chat: { id: 'c1', creatorId: 'u1' },
+    });
+    await controller.getMessagesDelta({ chatId: 'c1' }, user, workspace);
+    expect(aiChatMessageRepo.findByChatUpdatedAfter).toHaveBeenCalledWith(
+      'c1',
+      'ws1',
+      null,
+    );
+    await controller.getMessagesDelta(
+      { chatId: 'c1', cursor: 'CX' },
+      user,
+      workspace,
+    );
+    expect(aiChatMessageRepo.findByChatUpdatedAfter).toHaveBeenLastCalledWith(
+      'c1',
+      'ws1',
+      'CX',
+    );
+  });
+});

apps/server/src/core/ai-chat/ai-chat.controller.ts

@@ -51,6 +51,7 @@ import {
   ChatIdDto,
   ExportChatDto,
   GeneratePageTitleDto,
+  GetChatDeltaDto,
   GetChatMessagesDto,
   GetRunDto,
   RenameChatDto,
@@ -63,6 +64,47 @@ import {
   SUBSCRIBER_MAX_BUFFERED_BYTES,
 } from './ai-chat-stream-registry.service';
 import { startSseHeartbeat } from './sse-resilience';
+
+/**
+ * Write the attach TAIL to the hijacked socket in chunks that RESPECT drain
+ * (#491): each `write()` that returns false (the kernel buffer is full) is awaited
+ * on the next 'drain' before continuing. The old code wrote the whole buffer
+ * synchronously, which — with the pre-#491 32MB ring — spiked memory (half the
+ * OOM). Bails immediately if the socket ended/errored mid-write. Frames that the
+ * paused registry subscriber buffers while this awaits are delivered by start().
+ */
+async function writeTailRespectingDrain(
+  raw: {
+    write(chunk: string): boolean;
+    writableEnded?: boolean;
+    destroyed?: boolean;
+    once(event: string, cb: () => void): unknown;
+    removeListener?(event: string, cb: () => void): unknown;
+  },
+  frames: string[],
+): Promise<void> {
+  for (const frame of frames) {
+    if (raw.writableEnded || raw.destroyed) return;
+    const ok = raw.write(frame);
+    if (!ok) {
+      // Kernel buffer full — wait for drain (or an early close/error) before the
+      // next chunk, so a slow reader never forces the whole tail into memory.
+      // Remove ALL three listeners once any fires, so a many-chunk tail with
+      // repeated backpressure never leaks (MaxListenersExceededWarning).
+      await new Promise<void>((resolve) => {
+        const finish = (): void => {
+          raw.removeListener?.('drain', finish);
+          raw.removeListener?.('close', finish);
+          raw.removeListener?.('error', finish);
+          resolve();
+        };
+        raw.once('drain', finish);
+        raw.once('close', finish);
+        raw.once('error', finish);
+      });
+    }
+  }
+}
 import { EnvironmentService } from '../../integrations/environment/environment.service';
 
 /**
@@ -149,6 +191,46 @@ export class AiChatController {
     );
   }
 
+  /**
+   * Delta poll (#491) — the degraded-poll fallback's payload. Returns the chat's
+   * message rows changed since `cursor` (a DB-clock timestamp from the previous
+   * poll), a FRESH cursor, AND the current run fact `{ id, status } | null`. This
+   * replaces the old degraded poll that refetched ALL infinite-query pages (full
+   * parts) every 2.5s: the client seeds once and thereafter merges only the
+   * deltas by id (the overlap window guarantees repeats — the merge is idempotent,
+   * see mergeById). The run fact rides IN the delta (a separate /run poll would
+   * double the poll QPS), so the client FSM gets the run's status on the same tick.
+   * Owner-gated via assertOwnedChat (same gate as the other read endpoints).
+   */
+  @HttpCode(HttpStatus.OK)
+  @Post('messages/delta')
+  async getMessagesDelta(
+    @Body() dto: GetChatDeltaDto,
+    @AuthUser() user: User,
+    @AuthWorkspace() workspace: Workspace,
+  ): Promise<{
+    rows: AiChatMessage[];
+    cursor: string;
+    run: { id: string; status: string } | null;
+  }> {
+    await this.assertOwnedChat(dto.chatId, user, workspace);
+    const { rows, cursor } =
+      await this.aiChatMessageRepo.findByChatUpdatedAfter(
+        dto.chatId,
+        workspace.id,
+        dto.cursor ?? null,
+      );
+    const run = await this.aiChatRunService.getLatestForChat(
+      dto.chatId,
+      workspace.id,
+    );
+    return {
+      rows,
+      cursor,
+      run: run ? { id: run.id, status: run.status } : null,
+    };
+  }
+
   /**
    * Export a chat to Markdown (#183). The DB is the single source of truth: the
    * whole transcript is loaded (oldest -> newest) and rendered server-side. Now
@@ -249,19 +331,25 @@ export class AiChatController {
   }
 
   /**
-   * Attach to a chat's live run stream (#184 phase 1.5). A late/reloaded tab
-   * replays the frames buffered so far and then follows the live tail as a normal
-   * streamer. Owner-gated via assertOwnedChat (same gate as getRun). When there is
-   * nothing to resume — no entry, a finished run without expect=live, an
-   * overflowed buffer, or an anchor that pins a DIFFERENT run — the endpoint
-   * answers 204, the ONLY "nothing to resume" signal the AI SDK's reconnect
-   * accepts (it maps 204 to a silent no-op). With AI_CHAT_RESUMABLE_STREAM off the
-   * registry is never populated, so attach always 204s.
+   * Attach to a chat's live run stream from the client's step frontier (#184 phase
+   * 1.5, tail-only #491). A late/reloaded tab hands the server the step count it
+   * has PERSISTED (`n` = the seeded row's `metadata.stepsPersisted`) and its
+   * assistant row id (`anchor`); the registry answers with the TAIL past step `n`
+   * (a synthetic `start` frame + the buffered frames stamped >= n) and then the
+   * live tail. Owner-gated via assertOwnedChat (same gate as getRun). When there
+   * is nothing to resume — no entry, a ring that does not cover the client's
+   * frontier (overflow gap, or the client's seed lagged a rotation), or an anchor
+   * that pins a DIFFERENT run (invariant 6) — the endpoint answers 204, the ONLY
+   * "nothing to resume" signal the AI SDK's reconnect accepts (it maps 204 to a
+   * silent no-op); the client then refetches (a larger n) and re-attaches. With
+   * AI_CHAT_RESUMABLE_STREAM off the registry is never populated, so attach always
+   * 204s.
    *
-   * `expect=live` opts into replaying a finished-but-retained run (safe only when
-   * the client stripped the streaming tail); `anchor` is the client's assistant
-   * row id, which must match this run's (invariant 6) or a foreign run's
-   * transcript would be replayed into the store.
+   * The step marker `n` comes ONLY from the client — the server never reads the
+   * row to derive it, because a server-side n from a stale seed would open a
+   * silent one-step hole. The tail is written to the socket in CHUNKS respecting
+   * drain (writeTailRespectingDrain): the old code synchronously blasted the whole
+   * buffer, which — with the old 32MB cap — was half the OOM.
    */
   @SkipTransform()
   @UseGuards(JwtAuthGuard, UserThrottlerGuard)
@@ -269,39 +357,49 @@ export class AiChatController {
   @Get('runs/:chatId/stream')
   async attachRunStream(
     @Param('chatId', new ParseUUIDPipe()) chatId: string,
-    @Query('expect') expect: string | undefined,
     @Query('anchor') anchor: string | undefined,
+    @Query('n') n: string | undefined,
     @Req() req: FastifyRequest,
     @Res() res: FastifyReply,
     @AuthUser() user: User,
     @AuthWorkspace() workspace: Workspace,
   ): Promise<void> {
     await this.assertOwnedChat(chatId, user, workspace); // same gate as getRun
+    // The client's persisted step frontier. #491: distinguish a MISSING/invalid `n`
+    // (null — a NOT-tail-aware, legacy/parameterless tab expecting the old
+    // "finished -> 204 -> poll" contract) from `n=0` (a tail-aware client with
+    // nothing persisted yet). Passing 0 for a missing `n` would serve a finished,
+    // non-rotated run's WHOLE tail and a parameterless client would append it onto
+    // the steps it already shows -> #137/#161 duplicate. null makes the registry
+    // 204 such a finished run (see attach); a tail-aware n=0 still resumes.
+    const frontier: number | null =
+      n === undefined || n === '' || !Number.isFinite(Number(n))
+        ? null
+        : Math.max(0, Number(n));
+    // The per-subscriber backpressure cap tracks the (env-tunable) ring cap.
+    const subscriberCap =
+      this.streamRegistry?.subscriberMaxBufferedBytes ??
+      SUBSCRIBER_MAX_BUFFERED_BYTES;
     let stopHeartbeat: () => void = () => undefined;
-    const attachment = await this.streamRegistry?.attach(
-      chatId,
-      expect === 'live',
-      anchor,
-      {
-        onFrame: (frame) => {
-          // Backpressure guard: 2x the replay cap, so the initial replay burst
-          // alone can never trip it; only a genuinely stalled socket can.
-          try {
-            if (res.raw.writableLength > SUBSCRIBER_MAX_BUFFERED_BYTES) {
-              res.raw.destroy(); // 'close' fires -> unsubscribe below
-              return;
-            }
-            if (!res.raw.writableEnded) res.raw.write(frame);
-          } catch {
-            res.raw.destroy();
+    const attachment = await this.streamRegistry?.attach(chatId, anchor, frontier, {
+      onFrame: (frame) => {
+        // Backpressure guard: 2x the ring cap, so the initial tail burst alone
+        // can never trip it; only a genuinely stalled socket can.
+        try {
+          if (res.raw.writableLength > subscriberCap) {
+            res.raw.destroy(); // 'close' fires -> unsubscribe below
+            return;
           }
-        },
-        onEnd: () => {
-          stopHeartbeat();
-          if (!res.raw.writableEnded) res.raw.end();
-        },
+          if (!res.raw.writableEnded) res.raw.write(frame);
+        } catch {
+          res.raw.destroy();
+        }
       },
-    );
+      onEnd: () => {
+        stopHeartbeat();
+        if (!res.raw.writableEnded) res.raw.end();
+      },
+    });
     if (!attachment) {
       res.status(204).send(); // the ONLY "nothing to resume" signal the SDK accepts
       return;
@@ -330,13 +428,16 @@ export class AiChatController {
         // deliberately NO Connection/Keep-Alive (hop-by-hop; Safari/HTTP2)
       });
       res.raw.flushHeaders?.();
-      for (const frame of attachment.replay) res.raw.write(frame);
+      // Write the tail in chunks respecting drain (not a synchronous blast, which
+      // was half the OOM). Frames the paused subscriber buffers meanwhile are
+      // drained by start() below; its cap is the backstop for a stalled socket.
+      await writeTailRespectingDrain(res.raw, attachment.replay);
       if (attachment.finished) {
-        res.raw.end();
+        if (!res.raw.writableEnded) res.raw.end();
         return;
       }
       stopHeartbeat = startSseHeartbeat(res.raw, 15_000);
-      attachment.start(); // drain pending accumulated during replay, go live
+      attachment.start(); // drain pending accumulated during the tail write, go live
     } catch {
       attachment.unsubscribe();
       stopHeartbeat();

apps/server/src/core/ai-chat/ai-chat.service.run-race.spec.ts

@@ -181,7 +181,7 @@ describe('AiChatService.stream — abortSignal wiring (#184 F3)', () => {
       {} as never, // pageAccess
       { isAiChatDeferredToolsEnabled: () => false, isAiChatFinalStepLockdownEnabled: () => false } as never, // environment
     );
-    return { svc };
+    return { svc, aiChatMessageRepo };
   }
 
   const body = {
@@ -287,7 +287,7 @@ describe('AiChatService.stream — abortSignal wiring (#184 F3)', () => {
   // Drive stream() to the point streamText is called, capturing the options object
   // (which carries onStepFinish/onFinish/onError/onAbort) and the run hooks.
   async function captureStreamCallbacks() {
-    const { svc } = makeService();
+    const { svc, aiChatMessageRepo } = makeService();
     let capturedOpts: any;
     streamTextMock.mockImplementation((opts: any) => {
       capturedOpts = opts;
@@ -314,7 +314,7 @@ describe('AiChatService.stream — abortSignal wiring (#184 F3)', () => {
       runHooks: runHooks as never,
     });
     expect(capturedOpts).toBeDefined();
-    return { capturedOpts, runHooks };
+    return { capturedOpts, runHooks, aiChatMessageRepo };
   }
 
   it('F9: onStepFinish bumps the run step count, onFinish settles the run "completed" (the dominant autonomous-run path)', async () => {
@@ -369,6 +369,51 @@ describe('AiChatService.stream — abortSignal wiring (#184 F3)', () => {
       expect.stringContaining('provider exploded'),
     );
   });
+
+  // #490 reactive branch: a provider CONTEXT-OVERFLOW 400 in onError is classified,
+  // records a distinguishable cause, and stamps metadata.replayOverflow so the NEXT
+  // turn's budgeter trims aggressively (the recovery that un-bricks the chat).
+  it('#490: a context-overflow 400 stamps replayOverflow on the finalized row', async () => {
+    jest
+      .spyOn(Logger.prototype, 'error')
+      .mockImplementation(() => undefined as never);
+    jest
+      .spyOn(Logger.prototype, 'warn')
+      .mockImplementation(() => undefined as never);
+    const { capturedOpts, aiChatMessageRepo } = await captureStreamCallbacks();
+
+    const overflow = Object.assign(new Error('too large'), {
+      statusCode: 400,
+      message:
+        "This model's maximum context length is 128000 tokens. However, your messages resulted in 214000 tokens. Please reduce the length.",
+    });
+    await capturedOpts.onError({ error: overflow });
+
+    // The seed row exists (finalizeOwner is the owner-write path).
+    expect(aiChatMessageRepo.finalizeOwner).toHaveBeenCalled();
+    const calls = aiChatMessageRepo.finalizeOwner.mock.calls as any[][];
+    const patch = calls[calls.length - 1][2] as {
+      status: string;
+      metadata: Record<string, unknown>;
+    };
+    expect(patch.status).toBe('error');
+    expect(patch.metadata.replayOverflow).toBe(true);
+    expect(patch.metadata.error).toContain('контекстное окно');
+  });
+
+  it('#490: a non-overflow error does NOT stamp replayOverflow', async () => {
+    jest
+      .spyOn(Logger.prototype, 'error')
+      .mockImplementation(() => undefined as never);
+    const { capturedOpts, aiChatMessageRepo } = await captureStreamCallbacks();
+    await capturedOpts.onError({ error: new Error('network reset') });
+    const calls = aiChatMessageRepo.finalizeOwner.mock.calls as any[][];
+    const patch = calls[calls.length - 1][2] as {
+      status: string;
+      metadata: Record<string, unknown>;
+    };
+    expect('replayOverflow' in patch.metadata).toBe(false);
+  });
 });
 
 /**

apps/server/src/core/ai-chat/ai-chat.service.spec.ts

@@ -13,6 +13,7 @@ import {
   compactToolOutput,
   assistantParts,
   serializeSteps,
+  type StepPartsCache,
   rowToUiMessage,
   prepareAgentStep,
   stepBudgetWarning,
@@ -28,10 +29,14 @@ import {
   FINAL_STEP_NUDGE,
   STEP_LIMIT_NO_ANSWER_MARKER,
   OUTPUT_DEGENERATION_ERROR,
+  lastAssistantContextTokens,
+  lastAssistantReplayOverflow,
+  seedActivatedTools,
 } from './ai-chat.service';
 import type { AiChatMessage, Workspace } from '@docmost/db/types/entity.types';
 import { buildSystemPrompt } from './ai-chat.prompt';
 import type { McpClientsService } from './external-mcp/mcp-clients.service';
+import { resolveEffectiveReplayThreshold } from './history-budget';
 
 /**
  * Unit tests for compactToolOutput: the pure helper that shrinks tool outputs
@@ -114,6 +119,54 @@ describe('compactToolOutput', () => {
 describe('assistantParts', () => {
   type AnyPart = Record<string, unknown>;
 
+  // #490 memoization: assistantParts builds each step's parts once and caches
+  // them by the step OBJECT's identity, so a mid-stream flush does not
+  // re-stringify every prior step's (large) output. Observable property: with a
+  // shared cache, the second call over the SAME step object returns the cached
+  // (identical) part array even if the step's underlying output was swapped —
+  // proving the work was memoized, not redone.
+  it('memoizes a step by identity (shared cache => one build per step)', () => {
+    const cache: StepPartsCache = new WeakMap();
+    const step = {
+      text: 'x',
+      toolCalls: [{ toolCallId: 'c1', toolName: 'getPage', input: {} }],
+      toolResults: [{ toolCallId: 'c1', toolName: 'getPage', output: { v: 1 } }],
+    };
+    const first = assistantParts([step], '', cache) as AnyPart[];
+    expect((first.find((p) => p.type === 'tool-getPage')!.output as any).v).toBe(
+      1,
+    );
+    // Swap the output for a NEW value; a re-build would pick it up, a cache hit
+    // keeps the first result.
+    step.toolResults[0] = {
+      toolCallId: 'c1',
+      toolName: 'getPage',
+      output: { v: 2 },
+    };
+    const second = assistantParts([step], '', cache) as AnyPart[];
+    expect((second.find((p) => p.type === 'tool-getPage')!.output as any).v).toBe(
+      1,
+    );
+    // Same cached part objects are reused.
+    expect(second.find((p) => p.type === 'tool-getPage')).toBe(
+      first.find((p) => p.type === 'tool-getPage'),
+    );
+  });
+
+  it('without a cache, each call rebuilds (no stale memo)', () => {
+    const step = {
+      text: 'x',
+      toolCalls: [{ toolCallId: 'c1', toolName: 'getPage', input: {} }],
+      toolResults: [{ toolCallId: 'c1', toolName: 'getPage', output: { v: 1 } }],
+    };
+    const first = assistantParts([step], '') as AnyPart[];
+    step.toolResults[0].output = { v: 2 };
+    const second = assistantParts([step], '') as AnyPart[];
+    expect((second.find((p) => p.type === 'tool-getPage')!.output as any).v).toBe(
+      2,
+    );
+  });
+
   it('emits output-available for a tool-call WITH a paired result', () => {
     const steps = [
       {
@@ -231,61 +284,320 @@ describe('assistantParts', () => {
   });
 });
 
-describe('serializeSteps', () => {
+// #490 trace format v2: per call the trace stores { input } for the call and an
+// OUTCOME element — { ok: true } on success, { error, kind: 'thrown' } on a
+// thrown tool-error, { error, kind: 'interrupted' } on a mid-step abort. The tool
+// OUTPUT is no longer duplicated here (it lives once in metadata.parts).
+describe('serializeSteps (trace v2)', () => {
   it('returns null when there are no calls or results', () => {
     expect(serializeSteps([])).toBeNull();
   });
 
-  it('flattens calls and results into a compact trace', () => {
+  it('pairs a successful call with an { ok: true } outcome and NO output', () => {
     const trace = serializeSteps([
       {
-        toolCalls: [{ toolName: 'getPage', input: { id: 'p1' } }],
-        toolResults: [{ toolName: 'getPage', output: { title: 'T' } }],
+        toolCalls: [{ toolCallId: 'c1', toolName: 'getPage', input: { id: 'p1' } }],
+        toolResults: [{ toolCallId: 'c1', toolName: 'getPage' }],
       },
     ]) as Array<Record<string, unknown>>;
     expect(trace).toHaveLength(2);
     expect(trace[0]).toEqual({ toolName: 'getPage', input: { id: 'p1' } });
-    expect(trace[1]).toEqual({ toolName: 'getPage', output: { title: 'T' } });
+    expect(trace[1]).toEqual({ toolName: 'getPage', ok: true });
+    // The output is NOT stored in the trace any more (dedup: it lives in parts).
+    expect(trace.some((e) => 'output' in e)).toBe(false);
   });
 
-  it('records a THROWN tool failure (tool-error part) with its error message', () => {
+  it('records a THROWN failure with { error, kind: "thrown" }', () => {
     const trace = serializeSteps([
       {
-        toolCalls: [{ toolName: 'editPageText', input: { id: 'p1' } }],
+        toolCalls: [
+          { toolCallId: 'c1', toolName: 'editPageText', input: { id: 'p1' } },
+        ],
         toolResults: [],
         content: [
           {
             type: 'tool-error',
+            toolCallId: 'c1',
             toolName: 'editPageText',
             error: new Error('page is locked'),
           },
         ],
       },
     ]) as Array<Record<string, unknown>>;
-    // The call element is followed by a paired error element (mirroring how a
-    // successful result is appended), so the failure survives in the trace.
     expect(trace).toHaveLength(2);
     expect(trace[0]).toEqual({ toolName: 'editPageText', input: { id: 'p1' } });
     expect(trace[1]).toEqual({
       toolName: 'editPageText',
       error: 'page is locked',
+      kind: 'thrown',
     });
   });
 
-  it('truncates a very long tool-error message to the tool-output limit', () => {
+  it('marks an interrupted call (no result, no throw) with kind "interrupted"', () => {
+    const trace = serializeSteps([
+      {
+        toolCalls: [
+          { toolCallId: 'c1', toolName: 'createComment', input: { x: 1 } },
+        ],
+        toolResults: [],
+        content: [],
+      },
+    ]) as Array<Record<string, unknown>>;
+    expect(trace).toHaveLength(2);
+    expect(trace[1]).toEqual({
+      toolName: 'createComment',
+      error: 'Tool call did not complete.',
+      kind: 'interrupted',
+    });
+    // Structurally distinct from a thrown hard-fail so it never inflates an
+    // error-rate scan.
+    expect((trace[1] as { kind: string }).kind).not.toBe('thrown');
+  });
+
+  it('truncates a very long thrown-error message to the tool-output limit', () => {
     const long = 'x'.repeat(5000);
     const trace = serializeSteps([
       {
-        toolCalls: [{ toolName: 'editPageText', input: {} }],
+        toolCalls: [{ toolCallId: 'c1', toolName: 'editPageText', input: {} }],
         toolResults: [],
-        content: [{ type: 'tool-error', toolName: 'editPageText', error: long }],
+        content: [
+          {
+            type: 'tool-error',
+            toolCallId: 'c1',
+            toolName: 'editPageText',
+            error: long,
+          },
+        ],
       },
     ]) as Array<Record<string, unknown>>;
     const errorText = trace[1].error as string;
-    // Truncated (not the full 5000 chars) and carries the omission marker.
     expect(errorText.length).toBeLessThan(long.length);
     expect(errorText).toContain('chars omitted');
   });
+
+  it('pairs parallel calls in one step with their outcomes by id', () => {
+    const trace = serializeSteps([
+      {
+        toolCalls: [
+          { toolCallId: 'a', toolName: 'getPage', input: {} },
+          { toolCallId: 'b', toolName: 'searchPages', input: {} },
+        ],
+        toolResults: [{ toolCallId: 'b', toolName: 'searchPages' }],
+        content: [
+          { type: 'tool-error', toolCallId: 'a', toolName: 'getPage', error: 'nope' },
+        ],
+      },
+    ]) as Array<Record<string, unknown>>;
+    // call a, outcome a (thrown), call b, outcome b (ok)
+    expect(trace).toHaveLength(4);
+    expect(trace[1]).toEqual({ toolName: 'getPage', error: 'nope', kind: 'thrown' });
+    expect(trace[3]).toEqual({ toolName: 'searchPages', ok: true });
+  });
+});
+
+// #490: every assistant row flushAssistant writes carries the v2 era marker so a
+// dual-shape diagnostic query can branch on the trace shape without inspecting it.
+describe('toolTraceVersion era marker (#490)', () => {
+  it('stamps metadata.toolTraceVersion = 2 on every flushed row', () => {
+    const seed = flushAssistant([], '', 'streaming');
+    expect(seed.metadata.toolTraceVersion).toBe(2);
+    const done = flushAssistant(
+      [
+        {
+          text: 'ok',
+          toolCalls: [{ toolCallId: 'c1', toolName: 'getPage', input: {} }],
+          toolResults: [{ toolCallId: 'c1', toolName: 'getPage' }],
+        },
+      ],
+      '',
+      'completed',
+      { finishReason: 'stop' },
+    );
+    expect(done.metadata.toolTraceVersion).toBe(2);
+  });
+});
+
+// #490 replay-budget signal helpers over persisted history.
+describe('lastAssistantContextTokens', () => {
+  const row = (
+    role: string,
+    metadata: Record<string, unknown> | null,
+  ): AiChatMessage => ({ role, metadata }) as unknown as AiChatMessage;
+
+  it('reads the most recent assistant turn contextTokens (provider fact)', () => {
+    const hist = [
+      row('user', null),
+      row('assistant', { contextTokens: 12000 }),
+      row('user', null),
+      row('assistant', { contextTokens: 41000 }),
+    ];
+    expect(lastAssistantContextTokens(hist)).toBe(41000);
+  });
+
+  it('returns undefined when the last assistant turn recorded no usage', () => {
+    const hist = [row('assistant', { error: 'boom' }), row('user', null)];
+    expect(lastAssistantContextTokens(hist)).toBeUndefined();
+    expect(lastAssistantContextTokens([])).toBeUndefined();
+  });
+});
+
+// #490 snapshotOpenPage fast-path: skip the full Markdown export + upsert when a
+// snapshot already exists at the page's CURRENT version (same updated_at instant).
+describe('snapshotOpenPage fast-path (#490)', () => {
+  function makeSvc(existingSnapshot: unknown, pageUpdatedAt: Date) {
+    const exportPageMarkdown = jest.fn(async () => '# md');
+    const upsert = jest.fn(async () => undefined);
+    const findByChatPage = jest.fn(async () => existingSnapshot);
+    const pageRepo = {
+      findById: jest.fn(async () => ({
+        id: 'p1',
+        workspaceId: 'ws1',
+        updatedAt: pageUpdatedAt,
+      })),
+    };
+    const svc = new AiChatService(
+      {} as never, // ai
+      {} as never, // aiChatRepo
+      {} as never, // aiChatMessageRepo
+      { findByChatPage, upsert } as never, // aiChatPageSnapshotRepo
+      {} as never, // aiSettings
+      { exportPageMarkdown } as never, // tools
+      {} as never, // mcpClients
+      {} as never, // aiAgentRoleRepo
+      pageRepo as never, // pageRepo
+      {} as never, // pageAccess
+      {} as never, // environment
+    );
+    return { svc, exportPageMarkdown, upsert, findByChatPage };
+  }
+
+  const args = () =>
+    [
+      'chat1',
+      'p1',
+      { id: 'ws1' } as never,
+      { id: 'u1' } as never,
+      'sess',
+    ] as const;
+
+  it('skips export + upsert when the snapshot is already at this page version', async () => {
+    const t = new Date('2026-07-07T10:00:00Z');
+    const { svc, exportPageMarkdown, upsert } = makeSvc(
+      { pageUpdatedAt: t, contentMd: '# md' },
+      t,
+    );
+    await (svc as unknown as { snapshotOpenPage: (...a: unknown[]) => Promise<void> })
+      .snapshotOpenPage(...args());
+    expect(exportPageMarkdown).not.toHaveBeenCalled();
+    expect(upsert).not.toHaveBeenCalled();
+  });
+
+  it('exports + upserts when the page advanced since the snapshot', async () => {
+    const { svc, exportPageMarkdown, upsert } = makeSvc(
+      { pageUpdatedAt: new Date('2026-07-07T10:00:00Z'), contentMd: 'old' },
+      new Date('2026-07-07T11:00:00Z'),
+    );
+    await (svc as unknown as { snapshotOpenPage: (...a: unknown[]) => Promise<void> })
+      .snapshotOpenPage(...args());
+    expect(exportPageMarkdown).toHaveBeenCalledTimes(1);
+    expect(upsert).toHaveBeenCalledTimes(1);
+  });
+
+  it('seeds (exports + upserts) on the first turn (no snapshot yet)', async () => {
+    const { svc, exportPageMarkdown, upsert } = makeSvc(
+      undefined,
+      new Date('2026-07-07T10:00:00Z'),
+    );
+    await (svc as unknown as { snapshotOpenPage: (...a: unknown[]) => Promise<void> })
+      .snapshotOpenPage(...args());
+    expect(exportPageMarkdown).toHaveBeenCalledTimes(1);
+    expect(upsert).toHaveBeenCalledTimes(1);
+  });
+});
+
+// #490 deferred-tool activation persisted across turns.
+describe('seedActivatedTools', () => {
+  const valid = new Set(['Search_web', 'getPageJson', 'diffPageVersions']);
+
+  it('seeds from persisted metadata, intersected with current valid names', () => {
+    expect(
+      seedActivatedTools(
+        { activatedTools: ['Search_web', 'getPageJson'] },
+        valid,
+      ),
+    ).toEqual(['Search_web', 'getPageJson']);
+  });
+
+  it('drops a stored tool that is no longer valid (allowlist/role changed)', () => {
+    // 'Habr_publish' was activated before but is not in the current allowlist.
+    expect(
+      seedActivatedTools({ activatedTools: ['Search_web', 'Habr_publish'] }, valid),
+    ).toEqual(['Search_web']);
+  });
+
+  it('is empty/robust for missing, non-array, or unknown-shaped metadata', () => {
+    expect(seedActivatedTools(undefined, valid)).toEqual([]);
+    expect(seedActivatedTools({}, valid)).toEqual([]);
+    expect(seedActivatedTools({ activatedTools: 'nope' }, valid)).toEqual([]);
+    expect(
+      seedActivatedTools({ activatedTools: [1, 'getPageJson', null] }, valid),
+    ).toEqual(['getPageJson']);
+  });
+
+  it('de-duplicates stored names', () => {
+    expect(
+      seedActivatedTools(
+        { activatedTools: ['getPageJson', 'getPageJson'] },
+        valid,
+      ),
+    ).toEqual(['getPageJson']);
+  });
+});
+
+describe('lastAssistantReplayOverflow', () => {
+  const row = (
+    role: string,
+    metadata: Record<string, unknown> | null,
+  ): AiChatMessage => ({ role, metadata }) as unknown as AiChatMessage;
+
+  it('is true only when the LAST assistant turn overflowed', () => {
+    expect(
+      lastAssistantReplayOverflow([
+        row('assistant', { replayOverflow: true }),
+        row('user', null),
+      ]),
+    ).toBe(true);
+    // A recovered (later, non-overflow) assistant turn clears it.
+    expect(
+      lastAssistantReplayOverflow([
+        row('assistant', { replayOverflow: true }),
+        row('user', null),
+        row('assistant', { contextTokens: 5 }),
+      ]),
+    ).toBe(false);
+    expect(lastAssistantReplayOverflow([])).toBe(false);
+  });
+
+  // #490 reactive recovery: a prior turn stamped `replayOverflow` must make the
+  // NEXT turn's effective budget the AGGRESSIVE 0.5x cut — that harder trim is
+  // what un-bricks a chat that just 400'd on the context window. This exercises
+  // the exact wiring the service uses: read the stamp, then scale the threshold.
+  it('#490: a prior replayOverflow drives the next turn to the 0.5x aggressive budget', () => {
+    const history = [
+      row('assistant', { replayOverflow: true }),
+      row('user', null),
+    ];
+    const priorOverflowed = lastAssistantReplayOverflow(history);
+    expect(priorOverflowed).toBe(true);
+    // Base budget 100k -> aggressive recovery halves it to 50k this turn.
+    expect(resolveEffectiveReplayThreshold(100_000, priorOverflowed)).toBe(50_000);
+    // Odd base floors, not rounds.
+    expect(resolveEffectiveReplayThreshold(99_999, true)).toBe(49_999);
+    // No prior overflow -> the base budget is used verbatim (no aggressive cut).
+    expect(resolveEffectiveReplayThreshold(100_000, false)).toBe(100_000);
+    // An explicit off-switch (null) is never overridden, even on recovery.
+    expect(resolveEffectiveReplayThreshold(null, true)).toBeNull();
+  });
 });
 
 describe('rowToUiMessage', () => {
@@ -618,6 +930,23 @@ describe('flushAssistant', () => {
     expect(flushed.metadata.error).toBe('boom');
   });
 
+  // #490 observability: the replay budgeter's decision is stamped on the turn.
+  it('records replayTrimmedToTokens + replayOverflow when provided', () => {
+    const f = flushAssistant([], '', 'error', {
+      error: 'ctx',
+      replayTrimmedToTokens: 42_000,
+      replayOverflow: true,
+    });
+    expect(f.metadata.replayTrimmedToTokens).toBe(42_000);
+    expect(f.metadata.replayOverflow).toBe(true);
+  });
+
+  it('omits the replay metadata when not provided', () => {
+    const f = flushAssistant([], '', 'completed', { finishReason: 'stop' });
+    expect('replayTrimmedToTokens' in f.metadata).toBe(false);
+    expect('replayOverflow' in f.metadata).toBe(false);
+  });
+
   // #274 observability: the page-change diff the agent saw this turn is persisted
   // to metadata.pageChanged when a non-empty diff was injected, and omitted when
   // the diff is empty/whitespace or the arg is not supplied.

apps/server/src/core/ai-chat/ai-chat.service.ts

@@ -55,6 +55,12 @@ import {
   type SelectionContext,
 } from './tools/current-page.util';
 import { roleModelOverride } from './roles/role-model-config';
+import {
+  resolveReplayBudget,
+  resolveEffectiveReplayThreshold,
+  isContextOverflowError,
+  trimHistoryForReplay,
+} from './history-budget';
 import {
   startSseHeartbeat,
   stripStreamingHopByHopHeaders,
@@ -117,9 +123,14 @@ const FINAL_STEP_NUDGE =
 // NO text at all (#444, mitigates the "empty turn" the lockdown used to prevent
 // when the toggle is OFF). Makes the exhausted-without-answer state explicit to
 // the user and, on replay, to the model on the next turn.
+// The persisted content is the app's base locale (en-US) — which is ALSO the
+// i18n key the client localizes through `t()` — instead of a hardcoded Russian
+// string (it used to render Russian for every locale, and fed Russian back to
+// the model on replay). Keep it a plain, model-readable English sentence so the
+// next turn's replay reads cleanly; the client resolves the locale.
 const STEP_LIMIT_NO_ANSWER_MARKER =
-  '(Достигнут лимит шагов — итоговый ответ не сформулирован; работа могла ' +
-  'остаться незавершённой. Напишите «продолжай», чтобы агент продолжил.)';
+  '(Step limit reached — no final answer was produced; the work may be ' +
+  'unfinished. Reply "continue" to let the agent carry on.)';
 
 // Reason recorded in ai_chat_runs.error / the assistant row when the token-
 // degeneration detector (#444) aborts a run. Distinct from a user Stop (no error)
@@ -127,6 +138,15 @@ const STEP_LIMIT_NO_ANSWER_MARKER =
 const OUTPUT_DEGENERATION_ERROR =
   'Output degeneration detected (repeated token loop)';
 
+// Prefix recorded on the assistant row when the provider rejected the turn for
+// CONTEXT OVERFLOW (#490): the replayed history exceeded the model's window. The
+// row is ALSO stamped `metadata.replayOverflow` so the NEXT turn's budgeter trims
+// aggressively (the reactive recovery — the overflowing turn had no usage signal
+// to trigger preventive trimming, so the classified 400 is what un-bricks it).
+export const CONTEXT_OVERFLOW_ERROR_PREFIX =
+  'Диалог превысил контекстное окно модели; история будет агрессивно ' +
+  'сокращена на следующем ходу.';
+
 /**
  * Compute the step-budget warning text (#444), or '' when this step is outside
  * the warning band. The warning fires on steps
@@ -882,6 +902,21 @@ export class AiChatService implements OnModuleInit, OnModuleDestroy {
       const freshPage = await this.pageRepo.findById(pageId);
       // Page deleted during the turn (or somehow foreign) => don't write.
       if (!freshPage || freshPage.workspaceId !== workspace.id) return;
+      // Fast-path (#490): if a snapshot already exists at THIS page version
+      // (same updated_at instant), its content is already current — skip the full
+      // Markdown export + upsert entirely. A turn that did NOT touch the open page
+      // (the common case) thus does no snapshot work. This mirrors the read-side
+      // fast path in detectPageChange (sameInstant): both trust that a page edit
+      // bumps updated_at. When the agent (or a human) DID edit the page this turn,
+      // updated_at advanced, so this does not match and we re-export as before.
+      const existing = await this.aiChatPageSnapshotRepo.findByChatPage(
+        chatId,
+        pageId,
+        workspace.id,
+      );
+      if (existing && sameInstant(existing.pageUpdatedAt, freshPage.updatedAt)) {
+        return;
+      }
       const currentMd = await this.tools.exportPageMarkdown(
         user,
         sessionId,
@@ -921,10 +956,17 @@ export class AiChatService implements OnModuleInit, OnModuleDestroy {
     // supplied or the supplied one does not belong to this workspace.
     let isNewChat = false;
     let chatId = body.chatId;
+    // Persisted chat-level metadata bag (#490): read once here so the deferred-tool
+    // activation set can be seeded from the previous turn. Undefined for a new chat.
+    let chatMetadata: Record<string, unknown> | undefined;
     if (chatId) {
       const existing = await this.aiChatRepo.findById(chatId, workspace.id);
       if (!existing) {
         chatId = undefined;
+      } else {
+        chatMetadata = (existing.metadata ?? undefined) as
+          | Record<string, unknown>
+          | undefined;
       }
     }
     // The open page the client sent is attacker-controllable — BOTH its id and
@@ -1086,7 +1128,7 @@ export class AiChatService implements OnModuleInit, OnModuleDestroy {
       // per-row conversion and degraded to plain text with a "[tool context
       // omitted]" marker rather than 500-ing the whole turn (silent loss of tool
       // context is not acceptable — the model must see the truncation).
-      const messages = await convertHistoryResilient(uiMessages, (index, err) =>
+      let messages = await convertHistoryResilient(uiMessages, (index, err) =>
         this.logger.warn(
           `Degraded unconvertible history row ${index} on chat ${chatId} to text: ${
             err instanceof Error ? err.message : 'unknown error'
@@ -1135,6 +1177,56 @@ export class AiChatService implements OnModuleInit, OnModuleDestroy {
       // Here we only need the admin-configured system prompt.
       const resolved = await this.aiSettings.resolve(workspace.id);
 
+      // History-replay token budget (#490). The full conversation is replayed to
+      // the provider every turn, so a long chat eventually 400s on the context
+      // window — forever. Bound the REPLAYED history (never the persisted rows).
+      // PRIMARY signal is the provider's own fact: the last turn's contextTokens.
+      const replayBudget = resolveReplayBudget(resolved?.chatContextWindowRaw);
+      if (replayBudget.usedDefault) {
+        // The default fires precisely for installs with NO configured window —
+        // the ones that hit terminal overflow. Warn so it is observable.
+        this.logger.warn(
+          `AI chat (chat ${chatId}): no chatContextWindow configured; ` +
+            `applying the default replay budget (${replayBudget.thresholdTokens} tokens).`,
+        );
+      }
+      // Last turn's provider-reported context size (authoritative when present).
+      const priorContextTokens = lastAssistantContextTokens(oldHistory);
+      // Reactive recovery (#490): if the LAST turn was rejected for context
+      // overflow (stamped by onError), trim AGGRESSIVELY this turn — the
+      // overflowing turn produced no usage signal, so a normal-threshold trim may
+      // not shrink enough to fit. This is what un-bricks a chat that just 400'd.
+      const priorOverflowed = lastAssistantReplayOverflow(oldHistory);
+      const effectiveThreshold = resolveEffectiveReplayThreshold(
+        replayBudget.thresholdTokens,
+        priorOverflowed,
+      );
+      if (priorOverflowed) {
+        this.logger.warn(
+          `AI chat (chat ${chatId}): previous turn hit context overflow; ` +
+            `applying aggressive replay budget (${effectiveThreshold} tokens).`,
+        );
+      }
+      const preTrim = trimHistoryForReplay(
+        messages,
+        effectiveThreshold,
+        // A prior OVERFLOW means the provider count is stale/absent — force the
+        // char-estimate path by ignoring priorContextTokens on recovery.
+        priorOverflowed ? undefined : priorContextTokens,
+      );
+      messages = preTrim.messages;
+      // Observability (#490): record the budgeter's decision on the turn so the UI
+      // can surface "replay truncated at N tokens". Threaded into flushAssistant.
+      let replayTrimmedToTokens: number | undefined = preTrim.trimmed
+        ? preTrim.estimatedTokens
+        : undefined;
+      if (preTrim.trimmed) {
+        this.logger.log(
+          `AI chat (chat ${chatId}): replay history trimmed to ~${preTrim.estimatedTokens} ` +
+            `tokens (budget ${replayBudget.thresholdTokens}).`,
+        );
+      }
+
       // Build the external MCP toolset FIRST so the system prompt can carry each
       // connected server's admin-authored guidance (#180). Merge in admin-
       // configured external MCP tools (web search, etc.; §6.8). A down/slow
@@ -1326,10 +1418,19 @@ export class AiChatService implements OnModuleInit, OnModuleDestroy {
       //    tools + ALL external MCP tools), computed from the ACTUAL toolset so an
       //    external tool is loadable by its namespaced name. loadTools rejects any
       //    name outside this set.
-      const activatedTools = new Set<string>();
       const validDeferredNames = new Set<string>(
         Object.keys(baseTools).filter((k) => !CORE_TOOL_SET.has(k)),
       );
+      // #490: seed the activation set from the chat's PERSISTED set so the model
+      // does not re-run loadTools every turn to re-activate the same tools. Only
+      // when deferred loading is enabled, and ALWAYS intersected with the CURRENT
+      // valid deferred names — an allowlist/role change must never resurrect a tool
+      // that no longer exists (prepareAgentStep would get a phantom active name).
+      const activatedTools = new Set<string>(
+        deferredEnabled
+          ? seedActivatedTools(chatMetadata, validDeferredNames)
+          : [],
+      );
       // Add the loadTools meta-tool ONLY when the feature is enabled; when off the
       // toolset and behavior are exactly as before.
       const tools = deferredEnabled
@@ -1339,6 +1440,39 @@ export class AiChatService implements OnModuleInit, OnModuleDestroy {
           }
         : baseTools;
 
+      // #490: persist the (deterministically ordered) activation set back onto the
+      // chat metadata at turn end, so the NEXT turn seeds from it. Once-guarded and
+      // skipped when nothing new was activated (the set equals its seed) so an
+      // ordinary turn adds no extra write. Preserves other metadata keys.
+      let activatedToolsPersisted = false;
+      const persistActivatedTools = async (): Promise<void> => {
+        if (!deferredEnabled || activatedToolsPersisted || !chatId) return;
+        activatedToolsPersisted = true;
+        const current = [...activatedTools].sort();
+        const seeded = seedActivatedTools(chatMetadata, validDeferredNames).sort();
+        if (current.length === 0 || current.join('') === seeded.join('')) {
+          return; // nothing new activated -> no write
+        }
+        try {
+          await this.aiChatRepo.update(
+            chatId,
+            {
+              metadata: {
+                ...(chatMetadata ?? {}),
+                activatedTools: current,
+              },
+            } as never,
+            workspace.id,
+          );
+        } catch (err) {
+          this.logger.warn(
+            `Failed to persist activated tools (chat ${chatId}): ${
+              err instanceof Error ? err.message : 'unknown error'
+            }`,
+          );
+        }
+      };
+
       // Accumulate the turn's streamed output so a provider error / disconnect can
       // persist the PARTIAL answer the user already saw — the SDK's onError/onAbort
       // callbacks don't hand us the in-progress text. `capturedSteps` holds finished
@@ -1347,6 +1481,11 @@ export class AiChatService implements OnModuleInit, OnModuleDestroy {
       const capturedSteps: StepLike[] = [];
       let inProgressText = '';
 
+      // Per-turn step->parts memo (#490): shared across every flushAssistant call
+      // this turn so each finished step's (large) output is JSON-stringified ONCE,
+      // not re-stringified on every subsequent onStepFinish flush (was O(N²)).
+      const partsCache: StepPartsCache = new WeakMap();
+
       // Token-degeneration guard (#444). When the final-step lockdown is OFF, a
       // runaway repetition loop (the 255KB "loadTools." incident) is aborted via
       // this internal controller, unioned with the run/socket signal below. The
@@ -1404,27 +1543,39 @@ export class AiChatService implements OnModuleInit, OnModuleDestroy {
       // Per-step (non-terminal) update: persist the finished steps the moment a
       // step ends. Tolerant — a failed update is logged and swallowed so it never
       // throws into the stream. Keeps status 'streaming'.
-      const updateStreaming = async (): Promise<void> => {
-        if (!assistantId) return;
+      //
+      // #491: it now SIGNALS its outcome — the persisted `stepsPersisted` count on
+      // a CONFIRMED write, or null when it was skipped/failed. The caller rotates
+      // the run-stream registry ring ONLY on a non-null return (a confirmed
+      // persist), so a failed persist never rotates away a step nobody has (the
+      // classic inversion bug); a failure just makes the ring cover more.
+      const updateStreaming = async (): Promise<number | null> => {
+        if (!assistantId) return null;
         // Cheap short-circuit once the turn is finalized (see `finalized` below).
         // The AUTHORITATIVE guard is `onlyIfStreaming` on the UPDATE: a late
         // fire-and-forget step update could still be in flight on another pool
         // connection when finalize runs, so the SQL `WHERE status='streaming'`
         // (not this flag) is what prevents it clobbering the terminal row.
-        if (finalized) return;
+        if (finalized) return null;
+        // Build the flush ONCE so the returned count is EXACTLY the persisted
+        // `stepsPersisted` (both derive from capturedSteps.length at this instant).
+        const flushed = flushAssistant(capturedSteps, '', 'streaming', {
+          pageChanged,
+          partsCache,
+        });
+        const stepsPersisted = flushed.metadata.stepsPersisted as number;
         try {
-          await this.aiChatMessageRepo.update(
-            assistantId,
-            workspace.id,
-            flushAssistant(capturedSteps, '', 'streaming', { pageChanged }),
-            { onlyIfStreaming: true },
-          );
+          await this.aiChatMessageRepo.update(assistantId, workspace.id, flushed, {
+            onlyIfStreaming: true,
+          });
+          return stepsPersisted;
         } catch (err) {
           this.logger.warn(
             `Failed to update streaming assistant row: ${
               err instanceof Error ? err.message : 'unknown error'
             }`,
           );
+          return null;
         }
       };
 
@@ -1514,6 +1665,13 @@ export class AiChatService implements OnModuleInit, OnModuleDestroy {
           system,
           messages,
           tools,
+          // Pin the AI SDK per-request retry budget explicitly instead of relying
+          // on its default (which is also 2). Connection arithmetic per turn:
+          // (1 + maxRetries=2) × (1 + AI_STREAM_PRE_RESPONSE_RETRIES) network
+          // connects worst-case — the two retry layers compose, so making the SDK
+          // side explicit keeps that ceiling visible and pinned against SDK-default
+          // drift.
+          maxRetries: 2,
           // No maxOutputTokens cap on the agent: tool-call arguments (e.g. a full
           // page body for the write tools) are emitted as OUTPUT tokens, so a fixed
           // cap would truncate complex tool calls mid-argument. Let the model use its
@@ -1593,7 +1751,24 @@ export class AiChatService implements OnModuleInit, OnModuleDestroy {
             // this point still recovers the step. Not awaited here (never block the
             // stream), but SERIALIZED via stepUpdateChain so the writes commit in
             // step order; updateStreaming is error-tolerant (logs + swallows).
-            stepUpdateChain = stepUpdateChain.then(() => updateStreaming());
+            // #491: on a CONFIRMED persist, rotate the run-stream registry ring to
+            // drop the now-on-disk steps (stamp < stepsPersisted). Gated on the
+            // resumable flag (same as open/bind) and identity-checked in the
+            // registry; a null return (skipped/failed) rotates NOTHING (auto-safe).
+            stepUpdateChain = stepUpdateChain.then(async () => {
+              const persisted = await updateStreaming();
+              if (
+                persisted != null &&
+                runId &&
+                this.environment?.isAiChatResumableStreamEnabled?.()
+              ) {
+                this.streamRegistry?.confirmPersistedStep(
+                  chatId,
+                  runId,
+                  persisted,
+                );
+              }
+            });
             // #184: persist the run's progress (finished-step count). Fire-and-
             // forget; the hook swallows its own errors.
             if (runId) runHooks?.onStep?.(runId, capturedSteps.length);
@@ -1649,6 +1824,8 @@ export class AiChatService implements OnModuleInit, OnModuleDestroy {
                 // closure scope here). Omitted/0 = no limit.
                 maxContextTokens: resolved?.chatContextWindow,
                 pageChanged,
+                partsCache,
+                replayTrimmedToTokens,
               }),
             );
             // #184/#487: the RUN is finalized ALWAYS (never gated on the message).
@@ -1673,6 +1850,8 @@ export class AiChatService implements OnModuleInit, OnModuleDestroy {
             // own edits are baked in — and this also SEEDS the snapshot on the first
             // turn. Runs once across every terminal path (see snapshotTurnEnd).
             await snapshotTurnEnd();
+            // #490: persist the deferred-tool activation set for the next turn.
+            await persistActivatedTools();
 
             // Generate the chat title for a freshly created chat AFTER the stream's
             // provider call has completed — NOT concurrently with it. The z.ai coding
@@ -1696,7 +1875,16 @@ export class AiChatService implements OnModuleInit, OnModuleDestroy {
             // object, so the actual provider cause is clearly logged. Reuse the
             // shared formatter so provider error formatting stays unified.
             const e = error as { stack?: string };
-            const errorText = describeProviderError(error, String(error));
+            // #490 reactive branch: classify a CONTEXT-OVERFLOW rejection (the
+            // replayed history exceeded the model window). The overflowing turn had
+            // no prior usage to trigger preventive trimming, so we record a clear,
+            // distinguishable cause AND stamp the row so the NEXT turn's budgeter
+            // trims aggressively — the reactive recovery that un-bricks the chat.
+            const overflow = isContextOverflowError(error);
+            const providerError = describeProviderError(error, String(error));
+            const errorText = overflow
+              ? `${CONTEXT_OVERFLOW_ERROR_PREFIX} (${providerError})`
+              : providerError;
             this.logger.error(`AI chat stream error: ${errorText}`, e?.stack);
             // DIAGNOSTIC (Safari stream-drop investigation) — temporary: timing of
             // an error-terminated stream.
@@ -1714,6 +1902,9 @@ export class AiChatService implements OnModuleInit, OnModuleDestroy {
               flushAssistant(capturedSteps, inProgressText, 'error', {
                 error: errorText,
                 pageChanged,
+                partsCache,
+                replayTrimmedToTokens,
+                replayOverflow: overflow || undefined,
               }),
             );
             // #184: settle the RUN as failed, carrying the provider/transport cause.
@@ -1723,6 +1914,8 @@ export class AiChatService implements OnModuleInit, OnModuleDestroy {
             // committed before the error must be baked into the snapshot, or the
             // next turn would mis-report it as a user edit.
             await snapshotTurnEnd();
+            // #490: persist the deferred-tool activation set for the next turn.
+            await persistActivatedTools();
           },
           onAbort: async ({ steps }) => {
             // #444: distinguish a degeneration abort (our internal controller) from
@@ -1737,6 +1930,7 @@ export class AiChatService implements OnModuleInit, OnModuleDestroy {
                 flushAssistant(capturedSteps, truncated, 'error', {
                   error: OUTPUT_DEGENERATION_ERROR,
                   pageChanged,
+                  partsCache,
                 }),
               );
               if (runId)
@@ -1747,6 +1941,8 @@ export class AiChatService implements OnModuleInit, OnModuleDestroy {
                 );
               await closeExternalClients();
               await snapshotTurnEnd();
+              // #490: persist the deferred-tool activation set for the next turn.
+              await persistActivatedTools();
               return;
             }
             const partialChars =
@@ -1771,6 +1967,7 @@ export class AiChatService implements OnModuleInit, OnModuleDestroy {
             await finalizeAssistant(
               flushAssistant(capturedSteps, inProgressText, 'aborted', {
                 pageChanged,
+                partsCache,
               }),
             );
             // #184: settle the RUN as aborted (an explicit user stop reached the
@@ -1781,6 +1978,8 @@ export class AiChatService implements OnModuleInit, OnModuleDestroy {
             // committed before the client disconnect / stop() must be baked into the
             // snapshot, or the next turn would mis-report it as a user edit.
             await snapshotTurnEnd();
+            // #490: persist the deferred-tool activation set for the next turn.
+            await persistActivatedTools();
           },
         });
 
@@ -2091,6 +2290,70 @@ export function chatStreamMetadata(
   return undefined;
 }
 
+/**
+ * The provider-reported context size of the most recent assistant turn, read from
+ * its persisted `metadata.contextTokens` (#490 replay budgeter's PRIMARY signal —
+ * the provider's own fact, not an estimate). Returns undefined for a chat with no
+ * assistant turn yet, or one whose last turn recorded no usage (e.g. it errored),
+ * in which case the budgeter falls back to the char-estimate.
+ */
+export function lastAssistantContextTokens(
+  history: ReadonlyArray<AiChatMessage>,
+): number | undefined {
+  for (let i = history.length - 1; i >= 0; i--) {
+    const row = history[i];
+    if (row.role !== 'assistant') continue;
+    const meta = (row.metadata ?? {}) as { contextTokens?: unknown };
+    const n = meta.contextTokens;
+    return typeof n === 'number' && Number.isFinite(n) && n > 0 ? n : undefined;
+  }
+  return undefined;
+}
+
+/**
+ * Seed the per-turn deferred-tool activation set from a chat's persisted metadata
+ * (#490), INTERSECTED with the current valid deferred names. Persisting the set
+ * across turns saves the model re-running loadTools every turn to re-activate the
+ * same tools; intersecting on load means a changed allowlist / role can never
+ * resurrect a tool that no longer exists (which would hand prepareAgentStep a
+ * phantom active name). Tolerant of any stored shape — a non-array is ignored.
+ */
+export function seedActivatedTools(
+  metadata: Record<string, unknown> | undefined,
+  validDeferredNames: ReadonlySet<string>,
+): string[] {
+  const stored = metadata?.activatedTools;
+  if (!Array.isArray(stored)) return [];
+  const seen = new Set<string>();
+  const out: string[] = [];
+  for (const name of stored) {
+    if (typeof name === 'string' && validDeferredNames.has(name) && !seen.has(name)) {
+      seen.add(name);
+      out.push(name);
+    }
+  }
+  return out;
+}
+
+/**
+ * Whether the most recent assistant turn was rejected for CONTEXT OVERFLOW
+ * (#490): its row carries `metadata.replayOverflow` (stamped by the stream's
+ * onError). The next turn's budgeter reads this to trim aggressively — the
+ * reactive recovery. Only the LAST assistant turn matters (an older overflow was
+ * already recovered), so we stop at the first assistant row scanning backwards.
+ */
+export function lastAssistantReplayOverflow(
+  history: ReadonlyArray<AiChatMessage>,
+): boolean {
+  for (let i = history.length - 1; i >= 0; i--) {
+    const row = history[i];
+    if (row.role !== 'assistant') continue;
+    const meta = (row.metadata ?? {}) as { replayOverflow?: unknown };
+    return meta.replayOverflow === true;
+  }
+  return false;
+}
+
 /** The last message with role 'user' from a useChat payload, if any. */
 function lastUserMessage(
   messages: UIMessage[] | undefined,
@@ -2152,6 +2415,15 @@ export function sanitizeUserParts(
 /** Marker for a history row whose tool parts could not be replayed (#489). */
 export const TOOL_CONTEXT_OMITTED_MARKER = '[tool context omitted]';
 
+/**
+ * Synthetic error text for a tool call that neither returned a result nor threw
+ * a `tool-error` — i.e. it was interrupted mid-step (an abort / server restart).
+ * Shared by `assistantParts` (the replayed `output-error` part) and
+ * `serializeSteps` (the `{ kind: 'interrupted' }` trace element) so the replay
+ * text and the trace stay in lockstep (#490).
+ */
+export const TOOL_CALL_INCOMPLETE_TEXT = 'Tool call did not complete.';
+
 /**
  * Convert persisted UI history to model messages, tolerating a single poisoned
  * row (#489). `convertToModelMessages` over the WHOLE array throws if ANY row is
@@ -2359,71 +2631,97 @@ function normalizeToolError(error: unknown): string {
  */
 // Exported only so the unit tests can import these pure helpers; exporting
 // them does not change runtime behavior.
+/**
+ * Per-turn memo for {@link assistantParts}: a step's rebuilt parts keyed by the
+ * step OBJECT's identity (#490). A finished step in `capturedSteps` keeps a stable
+ * reference across every mid-stream flush, and `compactToolOutput` inside it does a
+ * `JSON.stringify` of the whole (often 50–200 KB) output — so without a memo each
+ * `onStepFinish` re-stringifies EVERY prior step's output (O(N²) stringify over a
+ * turn). Keyed by step identity => one stringify per step per turn. WeakMap so a
+ * turn's steps are GC'd with the turn.
+ */
+export type StepPartsCache = WeakMap<object, Array<Record<string, unknown>>>;
+
+/** Build the parts for ONE step (text + a part per tool call). Pure. */
+function buildStepParts(step: StepLike): Array<Record<string, unknown>> {
+  const parts: Array<Record<string, unknown>> = [];
+  if (step.text) {
+    parts.push({ type: 'text', text: step.text });
+  }
+  // Index this step's results by tool call id to pair calls with outputs.
+  const resultsById = new Map<string, unknown>();
+  for (const r of step.toolResults ?? []) {
+    if (r.toolCallId) resultsById.set(r.toolCallId, r.output);
+  }
+  // Index this step's THROWN tool failures (ai@6 `tool-error` content parts)
+  // by tool call id, so a call that failed replays with its real error text.
+  const errorsById = new Map<string, unknown>();
+  for (const part of step.content ?? []) {
+    if (part.type === 'tool-error' && part.toolCallId) {
+      errorsById.set(part.toolCallId, part.error);
+    }
+  }
+  for (const call of step.toolCalls ?? []) {
+    if (!call.toolName || !call.toolCallId) continue;
+    const hasResult = resultsById.has(call.toolCallId);
+    if (hasResult) {
+      // output-available: the tool returned; the next turn replays its result.
+      parts.push({
+        type: `tool-${call.toolName}`,
+        toolCallId: call.toolCallId,
+        state: 'output-available',
+        input: call.input,
+        output: compactToolOutput(resultsById.get(call.toolCallId)),
+      });
+    } else if (errorsById.has(call.toolCallId)) {
+      // The tool THREW: replay the REAL error so the model on the next turn
+      // knows WHY the call failed (and does not blindly repeat it). An
+      // output-error round-trips through convertToModelMessages as a balanced
+      // tool-call + tool-result, keeping the rebuilt history valid.
+      parts.push({
+        type: `tool-${call.toolName}`,
+        toolCallId: call.toolCallId,
+        state: 'output-error',
+        input: call.input,
+        errorText: normalizeToolError(errorsById.get(call.toolCallId)),
+      });
+    } else {
+      // No paired result AND no tool-error (e.g. aborted mid-step). Persisting
+      // a bare tool-call (input-available) would replay as an unpaired call and
+      // throw MissingToolResultsError on the next turn (convertToModelMessages
+      // emits no tool-result for it). Emit a SYNTHETIC paired result instead:
+      // an output-error round-trips through convertToModelMessages as a
+      // balanced tool-call + tool-result, keeping the rebuilt history valid.
+      parts.push({
+        type: `tool-${call.toolName}`,
+        toolCallId: call.toolCallId,
+        state: 'output-error',
+        input: call.input,
+        errorText: TOOL_CALL_INCOMPLETE_TEXT,
+      });
+    }
+  }
+  return parts;
+}
+
 export function assistantParts(
   steps: ReadonlyArray<StepLike> | undefined,
   fallbackText: string,
+  cache?: StepPartsCache,
 ): UIMessage['parts'] {
   const parts: Array<Record<string, unknown>> = [];
-  let sawText = false;
   for (const step of steps ?? []) {
-    if (step.text) {
-      parts.push({ type: 'text', text: step.text });
-      sawText = true;
-    }
-    // Index this step's results by tool call id to pair calls with outputs.
-    const resultsById = new Map<string, unknown>();
-    for (const r of step.toolResults ?? []) {
-      if (r.toolCallId) resultsById.set(r.toolCallId, r.output);
-    }
-    // Index this step's THROWN tool failures (ai@6 `tool-error` content parts)
-    // by tool call id, so a call that failed replays with its real error text.
-    const errorsById = new Map<string, unknown>();
-    for (const part of step.content ?? []) {
-      if (part.type === 'tool-error' && part.toolCallId) {
-        errorsById.set(part.toolCallId, part.error);
-      }
-    }
-    for (const call of step.toolCalls ?? []) {
-      if (!call.toolName || !call.toolCallId) continue;
-      const hasResult = resultsById.has(call.toolCallId);
-      if (hasResult) {
-        // output-available: the tool returned; the next turn replays its result.
-        parts.push({
-          type: `tool-${call.toolName}`,
-          toolCallId: call.toolCallId,
-          state: 'output-available',
-          input: call.input,
-          output: compactToolOutput(resultsById.get(call.toolCallId)),
-        });
-      } else if (errorsById.has(call.toolCallId)) {
-        // The tool THREW: replay the REAL error so the model on the next turn
-        // knows WHY the call failed (and does not blindly repeat it). An
-        // output-error round-trips through convertToModelMessages as a balanced
-        // tool-call + tool-result, keeping the rebuilt history valid.
-        parts.push({
-          type: `tool-${call.toolName}`,
-          toolCallId: call.toolCallId,
-          state: 'output-error',
-          input: call.input,
-          errorText: normalizeToolError(errorsById.get(call.toolCallId)),
-        });
-      } else {
-        // No paired result AND no tool-error (e.g. aborted mid-step). Persisting
-        // a bare tool-call (input-available) would replay as an unpaired call and
-        // throw MissingToolResultsError on the next turn (convertToModelMessages
-        // emits no tool-result for it). Emit a SYNTHETIC paired result instead:
-        // an output-error round-trips through convertToModelMessages as a
-        // balanced tool-call + tool-result, keeping the rebuilt history valid.
-        parts.push({
-          type: `tool-${call.toolName}`,
-          toolCallId: call.toolCallId,
-          state: 'output-error',
-          input: call.input,
-          errorText: 'Tool call did not complete.',
-        });
-      }
+    // Memoize per step object (#490): a finished step is immutable and keeps its
+    // reference across flushes, so its parts (and the costly output stringify) are
+    // built exactly once per turn. A cache miss (or no cache) just rebuilds.
+    let stepParts = cache?.get(step as object);
+    if (!stepParts) {
+      stepParts = buildStepParts(step);
+      cache?.set(step as object, stepParts);
     }
+    parts.push(...stepParts);
   }
+  const sawText = parts.some((p) => p.type === 'text');
   if (!sawText && fallbackText) {
     // No per-step text (e.g. a single final block): append the final text after
     // any tool parts so the natural call -> result -> answer order is preserved.
@@ -2586,6 +2884,16 @@ export function flushAssistant(
     maxContextTokens?: number;
     error?: string;
     pageChanged?: { title: string; diff: string } | null;
+    // Per-turn step->parts memo (#490): pass the SAME cache on every flush of a
+    // turn so each finished step's output is stringified once, not once per flush.
+    partsCache?: StepPartsCache;
+    // #490 observability: when the replay budgeter trimmed this turn's history,
+    // the (estimated) token size it trimmed to — the UI can show "replay truncated
+    // at N tokens". Omitted when nothing was trimmed.
+    replayTrimmedToTokens?: number;
+    // #490 reactive branch: set when the provider rejected this turn for context
+    // overflow. Stamped into metadata so the NEXT turn's budgeter trims aggressively.
+    replayOverflow?: boolean;
   },
 ): AssistantFlush {
   const finished = capturedSteps ?? [];
@@ -2595,13 +2903,32 @@ export function flushAssistant(
   // in-progress step's text (the partial answer cut off by an error/abort, or
   // simply not yet flushed mid-stream) as the last text part so the persisted
   // parts match what streamed to the client.
-  const parts = assistantParts(finished, '') as unknown as Array<
-    Record<string, unknown>
-  >;
+  const parts = assistantParts(
+    finished,
+    '',
+    extra?.partsCache,
+  ) as unknown as Array<Record<string, unknown>>;
   if (trailing) parts.push({ type: 'text', text: trailing });
 
   const metadata: Record<string, unknown> = {
     parts: parts as unknown as UIMessage['parts'],
+    // Era marker for the `tool_calls` trace shape (#490): v2 stores outcome flags
+    // ({ ok } / { error, kind }) and NO tool output (the output lives once in
+    // `parts`). Old rows have no marker and the legacy { output } shape; a
+    // dual-shape query branches on this. Old rows are deliberately NOT migrated.
+    toolTraceVersion: 2,
+    // #491 STEP MARKER: the number of FINISHED steps whose parts are in THIS row,
+    // written by the SAME flush that builds `parts` (atomically — they are both
+    // derived from `finished`, so the marker can NEVER disagree with the persisted
+    // parts). This is the step-alignment anchor the resume stack builds on:
+    // - the registry rotates its retention ring only on a CONFIRMED persist of
+    //   step N (commit 3);
+    // - attach slices the tail at "step > N" from the client's persisted seed.
+    // It is NOT `run.stepCount`: recordStep is fire-and-forget and NOT atomic with
+    // the parts write, so stepCount could race ahead of the persisted parts
+    // (seed↔marker drift). The in-progress trailing text (an error/abort partial,
+    // or a mid-stream flush) is NOT a finished step and is excluded from the count.
+    stepsPersisted: finished.length,
   };
   // finishReason: prefer an explicit one; else derive a sensible value from the
   // terminal status (so onError/onAbort records keep their historical reason).
@@ -2617,6 +2944,9 @@ export function flushAssistant(
   if (extra?.contextTokens) metadata.contextTokens = extra.contextTokens;
   if (extra?.maxContextTokens)
     metadata.maxContextTokens = extra.maxContextTokens;
+  if (extra?.replayTrimmedToTokens)
+    metadata.replayTrimmedToTokens = extra.replayTrimmedToTokens;
+  if (extra?.replayOverflow) metadata.replayOverflow = true;
   if (extra?.error) metadata.error = extra.error;
   // Persist the page-change diff the agent saw this turn (#274 observability),
   // so history / the Markdown export can show what the user changed. Only when
@@ -2642,42 +2972,85 @@ export function flushAssistant(
 
 /**
  * Reduce SDK step objects to a compact, JSON-serializable trace for the
- * `tool_calls` column. Stores only what the UI action-log and history need —
- * never raw provider payloads or keys.
+ * `tool_calls` column — trace format **v2** (#490).
+ *
+ * v2 stores, per call, ONLY the metadata a queryable trace needs — never the
+ * tool OUTPUT. Before #490 each output was persisted TWICE: once here (compacted)
+ * and once in `metadata.parts` (via `assistantParts`), so a 50-step run with
+ * 50–200 KB outputs wrote hundreds of MB per turn (each `onStepFinish` rewrote
+ * the whole row). The parts copy is the one the model replays and the UI/Markdown
+ * export render, so the trace copy of the output was pure duplication. v2 keeps
+ * the output ONLY in parts and reduces the trace to outcome flags.
+ *
+ * Element shapes (paired per call, in order):
+ *  - `{ toolName, input }`                       — the call
+ *  - `{ toolName, ok: true }`                     — it returned a result (success)
+ *  - `{ toolName, error, kind: 'thrown' }`        — it threw a `tool-error`
+ *  - `{ toolName, error, kind: 'interrupted' }`   — no result and no throw (an
+ *      abort / server restart mid-step). `kind` is MANDATORY: without it a
+ *      synthetic "Tool call did not complete." is indistinguishable from a real
+ *      hard-fail and pollutes any error-rate scan. The distinction is STRUCTURAL
+ *      (an `errorsById` hit vs the synthetic fallback branch), NOT a per-tool
+ *      classifier — soft failures stay OUT of the trace (they live in
+ *      `metadata.parts` outputs; a per-tool mirror would persist its own bugs).
+ *
+ * Rows carry `metadata.toolTraceVersion: 2` (set by {@link flushAssistant}) so a
+ * dual-shape query can branch on the era. Old rows are NOT migrated (rewriting
+ * giant jsonb is the very WAL churn this removes); see docs/reading-ai-logs.md.
  */
 export function serializeSteps(
   steps: ReadonlyArray<{
-    toolCalls?: ReadonlyArray<{ toolName?: string; input?: unknown }>;
-    toolResults?: ReadonlyArray<{ toolName?: string; output?: unknown }>;
+    toolCalls?: ReadonlyArray<{
+      toolCallId?: string;
+      toolName?: string;
+      input?: unknown;
+    }>;
+    toolResults?: ReadonlyArray<{ toolCallId?: string; toolName?: string }>;
     content?: ReadonlyArray<{
       type?: string;
+      toolCallId?: string;
       toolName?: string;
       error?: unknown;
     }>;
   }>,
 ): unknown {
-  const calls: Array<{
-    toolName?: string;
-    input?: unknown;
-    output?: unknown;
-    error?: string;
-  }> = [];
+  const calls: Array<
+    | { toolName?: string; input?: unknown }
+    | { toolName?: string; ok: true }
+    | { toolName?: string; error: string; kind: 'thrown' | 'interrupted' }
+  > = [];
   for (const step of steps ?? []) {
+    // Index this step's results + thrown errors by tool call id, so each call is
+    // paired with its outcome (mirrors assistantParts' pairing exactly).
+    const resultIds = new Set<string>();
+    for (const r of step.toolResults ?? []) {
+      if (r.toolCallId) resultIds.add(r.toolCallId);
+    }
+    const errorsById = new Map<string, unknown>();
+    for (const part of step.content ?? []) {
+      if (part.type === 'tool-error' && part.toolCallId) {
+        errorsById.set(part.toolCallId, part.error);
+      }
+    }
     for (const call of step.toolCalls ?? []) {
       calls.push({ toolName: call.toolName, input: call.input });
-    }
-    for (const r of step.toolResults ?? []) {
-      calls.push({ toolName: r.toolName, output: compactToolOutput(r.output) });
-    }
-    // ai@6 surfaces a THROWN tool failure as a `tool-error` content part, NOT as
-    // a `toolResults` entry. Record it as its own paired element (mirroring how a
-    // successful result is appended) so the failure and its reason survive in the
-    // trace instead of leaving an orphaned call with no result.
-    for (const part of step.content ?? []) {
-      if (part.type === 'tool-error') {
+      if (call.toolCallId && resultIds.has(call.toolCallId)) {
+        // Success: the output itself lives in metadata.parts, not here.
+        calls.push({ toolName: call.toolName, ok: true });
+      } else if (call.toolCallId && errorsById.has(call.toolCallId)) {
+        // Hard fail: the tool threw. Persist the real (bounded) reason.
         calls.push({
-          toolName: part.toolName,
-          error: normalizeToolError(part.error),
+          toolName: call.toolName,
+          error: normalizeToolError(errorsById.get(call.toolCallId)),
+          kind: 'thrown',
+        });
+      } else {
+        // Neither a result nor a throw: interrupted mid-step (abort/restart).
+        // Marked structurally so it never inflates a thrown-error count.
+        calls.push({
+          toolName: call.toolName,
+          error: TOOL_CALL_INCOMPLETE_TEXT,
+          kind: 'interrupted',
         });
       }
     }

apps/server/src/core/ai-chat/ai-chat.step-marker.spec.ts

@@ -0,0 +1,65 @@
+import { flushAssistant } from './ai-chat.service';
+
+/**
+ * #491 STEP MARKER — `metadata.stepsPersisted` is written by the SAME flush that
+ * builds `metadata.parts`, so the marker can never disagree with the persisted
+ * parts (the step-alignment anchor the resume stack builds on). These are
+ * PROPERTY tests: they assert the marker tracks the number of FINISHED steps for
+ * every flush shape.
+ */
+
+// A finished step carrying one line of text and one tool call/result.
+function step(i: number) {
+  return {
+    text: `step ${i}`,
+    toolCalls: [
+      { toolCallId: `c${i}`, toolName: 'getPage', input: { id: `p${i}` } },
+    ],
+    toolResults: [
+      { toolCallId: `c${i}`, toolName: 'getPage', output: { title: `T${i}` } },
+    ],
+  };
+}
+
+describe('flushAssistant step marker (#491)', () => {
+  it('seed (no steps) → stepsPersisted 0', () => {
+    const f = flushAssistant([], '', 'streaming');
+    expect(f.metadata.stepsPersisted).toBe(0);
+  });
+
+  it('PROPERTY: stepsPersisted equals the number of FINISHED steps, for any N', () => {
+    for (let n = 0; n <= 6; n++) {
+      const steps = Array.from({ length: n }, (_, i) => step(i));
+      const f = flushAssistant(steps, '', 'streaming');
+      expect(f.metadata.stepsPersisted).toBe(n);
+      // ...and the parts actually contain those N steps' text (marker agrees with
+      // the persisted parts — the atomicity the whole design relies on).
+      const parts = f.metadata.parts as Array<Record<string, unknown>>;
+      const textParts = parts.filter((p) => p.type === 'text');
+      expect(textParts).toHaveLength(n);
+    }
+  });
+
+  it('an in-progress trailing partial does NOT increment the marker', () => {
+    // 2 finished steps + a partial (not-yet-finished) trailing text: the marker
+    // counts only the CONFIRMED step boundaries, not the partial.
+    const f = flushAssistant([step(0), step(1)], 'partial third step', 'error', {
+      error: 'boom',
+    });
+    expect(f.metadata.stepsPersisted).toBe(2);
+    // The partial text IS persisted in parts (so the user sees it), but it is not a
+    // counted step.
+    const parts = f.metadata.parts as Array<Record<string, unknown>>;
+    expect(parts[parts.length - 1]).toEqual({
+      type: 'text',
+      text: 'partial third step',
+    });
+  });
+
+  it('terminal completed flush counts all finished steps', () => {
+    const f = flushAssistant([step(0), step(1), step(2)], '', 'completed', {
+      finishReason: 'stop',
+    });
+    expect(f.metadata.stepsPersisted).toBe(3);
+  });
+});

apps/server/src/core/ai-chat/ai-chat.write-volume.int-spec.ts

@@ -0,0 +1,209 @@
+import { randomBytes } from 'crypto';
+import { Client } from 'pg';
+import { flushAssistant, serializeSteps } from './ai-chat.service';
+
+/**
+ * #490 write-volume regression — an OBSERVABLE-PROPERTY test on a LIVE Postgres,
+ * not "bytes through a mock repo" (a mock measures exactly the thing that does not
+ * hurt). It drives a realistic 50-step run where each step returns a ~100 KB tool
+ * output and, at every `onStepFinish`, UPDATEs the assistant row the way the
+ * service does — then reads the REAL write volume via the `pg_current_wal_lsn()`
+ * delta around the run.
+ *
+ * The property proven: v2 stores each tool OUTPUT only in `metadata.parts`, no
+ * longer ALSO in the `tool_calls` trace. So:
+ *   1. the trace (`tool_calls`) column's write volume is now O(Σ steps) — tiny,
+ *      linear outcome flags — vs the pre-#490 O(N²) that re-persisted every prior
+ *      output on every step; and
+ *   2. the FULL-row write volume drops sharply (the duplicated output copy is gone).
+ *
+ * Connects to the local gitmost test Postgres (docker `gitmost-test-pg` on :5432);
+ * SKIPS cleanly when that DB is not reachable so it never breaks a DB-less CI.
+ */
+const CONN =
+  process.env.WAL_TEST_DATABASE_URL ??
+  'postgresql://docmost:docmost_dev_pw@localhost:5432/docmost';
+
+// A step whose tool output is ~100 KB (a page read), in the SDK StepLike shape.
+// The body is INCOMPRESSIBLE random text — a `'x'.repeat()` filler would TOAST-
+// compress to nothing and hide the real write volume (a page body does not).
+function makeStep(i: number, outputBytes = 100_000) {
+  const body = randomBytes(Math.ceil(outputBytes * 0.75)).toString('base64');
+  return {
+    text: `step ${i} reasoning`,
+    toolCalls: [{ toolCallId: `c${i}`, toolName: 'getPage', input: { id: `p${i}` } }],
+    toolResults: [
+      {
+        toolCallId: `c${i}`,
+        toolName: 'getPage',
+        output: { id: `p${i}`, title: `Page ${i}`, body },
+      },
+    ],
+  };
+}
+
+// The pre-#490 (v1) trace: outputs stored a SECOND time in `tool_calls`
+// (the duplication #490 removed). Mirrors the OLD serializeSteps shape.
+function v1Trace(steps: ReturnType<typeof makeStep>[]): unknown {
+  const calls: unknown[] = [];
+  for (const s of steps) {
+    for (const c of s.toolCalls) calls.push({ toolName: c.toolName, input: c.input });
+    for (const r of s.toolResults)
+      calls.push({ toolName: r.toolName, output: r.output });
+  }
+  return calls;
+}
+
+async function walDelta(
+  client: Client,
+  fn: () => Promise<void>,
+): Promise<number> {
+  const before = (await client.query('SELECT pg_current_wal_lsn() AS l')).rows[0]
+    .l as string;
+  await fn();
+  // NOTE: do NOT pg_switch_wal() here — a segment switch pads the LSN to the next
+  // 16 MB boundary and would swamp the actual write delta. The raw LSN advances by
+  // the bytes of WAL emitted, which is exactly what we want to measure.
+  const after = (await client.query('SELECT pg_current_wal_lsn() AS l')).rows[0]
+    .l as string;
+  return Number(
+    (await client.query('SELECT pg_wal_lsn_diff($1,$2) AS d', [after, before]))
+      .rows[0].d,
+  );
+}
+
+describe('#490 write-volume on a live Postgres (pg_current_wal_lsn delta)', () => {
+  let client: Client | undefined;
+  let available = false;
+
+  beforeAll(async () => {
+    try {
+      client = new Client(CONN);
+      await client.connect();
+      await client.query('SELECT pg_current_wal_lsn()');
+      available = true;
+    } catch {
+      available = false;
+      client = undefined;
+    }
+  });
+
+  afterAll(async () => {
+    await client?.end().catch(() => undefined);
+  });
+
+  const STEPS = 50;
+
+  it('v2 trace write volume is O(Σ steps) — a tiny fraction of the v1 duplicate', async () => {
+    if (!available || !client) {
+      console.warn('SKIP: gitmost-test-pg not reachable; skipping WAL test.');
+      return;
+    }
+    const c = client;
+    // Isolated table so we measure only the tool_calls (trace) column's writes.
+    await c.query('DROP TABLE IF EXISTS _wal_trace');
+    await c.query('CREATE TABLE _wal_trace(id int primary key, tool_calls jsonb)');
+    await c.query("INSERT INTO _wal_trace VALUES (1, '[]'::jsonb)");
+
+    const steps: ReturnType<typeof makeStep>[] = [];
+
+    // v1: each step re-persists ALL prior outputs into the trace (the O(N²) churn).
+    const v1 = await walDelta(c, async () => {
+      const acc: ReturnType<typeof makeStep>[] = [];
+      for (let i = 0; i < STEPS; i++) {
+        acc.push(makeStep(i));
+        await c.query('UPDATE _wal_trace SET tool_calls=$1 WHERE id=1', [
+          JSON.stringify(v1Trace(acc)),
+        ]);
+      }
+      steps.push(...acc);
+    });
+
+    await c.query("UPDATE _wal_trace SET tool_calls='[]'::jsonb WHERE id=1");
+
+    // v2: the REAL serializeSteps — outcome flags only, NO outputs.
+    const v2 = await walDelta(c, async () => {
+      const acc: ReturnType<typeof makeStep>[] = [];
+      for (let i = 0; i < STEPS; i++) {
+        acc.push(makeStep(i));
+        await c.query('UPDATE _wal_trace SET tool_calls=$1 WHERE id=1', [
+          JSON.stringify(serializeSteps(acc)),
+        ]);
+      }
+    });
+
+    await c.query('DROP TABLE IF EXISTS _wal_trace');
+
+    // eslint-disable-next-line no-console
+    console.log(
+      `[#490 WAL] trace column over ${STEPS} steps: v1=${(v1 / 1e6).toFixed(1)}MB ` +
+        `v2=${(v2 / 1e6).toFixed(2)}MB (${(v1 / v2).toFixed(0)}x smaller)`,
+    );
+
+    // The trace no longer carries outputs: v2 is a tiny fraction of v1's WAL.
+    expect(v2).toBeLessThan(v1 * 0.1);
+    // And v2's trace WAL is small in absolute terms — O(Σ steps) of flags, not
+    // O(N² × output). 50 steps of ~40-byte flags is well under a few MB of WAL.
+    expect(v2).toBeLessThan(5_000_000);
+    // v1's duplicate alone is huge (≈ the 100 KB output re-written N² times).
+    expect(v1).toBeGreaterThan(50_000_000);
+  }, 120_000);
+
+  it('the full assistant row write drops sharply once the duplicate is gone', async () => {
+    if (!available || !client) return;
+    const c = client;
+    await c.query('DROP TABLE IF EXISTS _wal_full');
+    await c.query(
+      'CREATE TABLE _wal_full(id int primary key, content text, tool_calls jsonb, metadata jsonb, status text)',
+    );
+    await c.query("INSERT INTO _wal_full VALUES (1, '', '[]'::jsonb, '{}'::jsonb, 'streaming')");
+
+    const writeRow = async (patch: {
+      content: string;
+      toolCalls: unknown;
+      metadata: unknown;
+      status: string;
+    }) =>
+      c.query(
+        'UPDATE _wal_full SET content=$1, tool_calls=$2, metadata=$3, status=$4 WHERE id=1',
+        [
+          patch.content,
+          JSON.stringify(patch.toolCalls ?? null),
+          JSON.stringify(patch.metadata),
+          patch.status,
+        ],
+      );
+
+    // v2 (real flushAssistant): outputs live once, in metadata.parts.
+    const v2 = await walDelta(c, async () => {
+      const acc: ReturnType<typeof makeStep>[] = [];
+      for (let i = 0; i < STEPS; i++) {
+        acc.push(makeStep(i));
+        await writeRow(flushAssistant(acc as never, '', 'streaming'));
+      }
+    });
+
+    await c.query("UPDATE _wal_full SET content='', tool_calls='[]'::jsonb, metadata='{}'::jsonb WHERE id=1");
+
+    // v1: same row PLUS the duplicated outputs in the trace column.
+    const v1 = await walDelta(c, async () => {
+      const acc: ReturnType<typeof makeStep>[] = [];
+      for (let i = 0; i < STEPS; i++) {
+        acc.push(makeStep(i));
+        const f = flushAssistant(acc as never, '', 'streaming');
+        await writeRow({ ...f, toolCalls: v1Trace(acc) });
+      }
+    });
+
+    await c.query('DROP TABLE IF EXISTS _wal_full');
+
+    // eslint-disable-next-line no-console
+    console.log(
+      `[#490 WAL] full row over ${STEPS} steps: v1=${(v1 / 1e6).toFixed(1)}MB ` +
+        `v2=${(v2 / 1e6).toFixed(1)}MB (saved ${((1 - v2 / v1) * 100).toFixed(0)}%)`,
+    );
+
+    // Removing the duplicated trace copy is a large, real write-volume reduction.
+    expect(v2).toBeLessThan(v1 * 0.75);
+  }, 120_000);
+});

apps/server/src/core/ai-chat/chat-markdown.util.spec.ts

@@ -1,5 +1,10 @@
-import { buildChatMarkdown, normalizeLang } from './chat-markdown.util';
+import {
+  buildChatMarkdown,
+  normalizeLang,
+  labelledToolNames,
+} from './chat-markdown.util';
 import type { AiChatMessage } from '@docmost/db/types/entity.types';
+import { SHARED_TOOL_SPECS } from '../../../../../packages/mcp/src/tool-specs';
 
 /**
  * normalizeLang: the client sends `i18n.language` — a FULL locale tag like
@@ -455,3 +460,43 @@ describe('buildChatMarkdown (server) — structure', () => {
     expect(md).toContain('````');
   });
 });
+
+/**
+ * #494 — REVERSE drift-guard for the export's friendly tool labels. A label keyed
+ * by a tool name that no longer exists silently degrades to the generic
+ * "Ran tool <name>" line; nothing reddened before. This asserts every labelled
+ * name is a real in-app tool and that both languages label the same set.
+ */
+describe('tool-label parity (#494)', () => {
+  // In-app tool names come from the shared registry (inAppKey, excluding
+  // mcpOnly specs) PLUS the inline in-app-only tools that carry a friendly label.
+  // The only labelled inline tool is the hybrid semantic search.
+  const INLINE_INAPP_LABELLED = new Set(['searchPages']);
+
+  function validInAppToolNames(): Set<string> {
+    const names = new Set<string>(INLINE_INAPP_LABELLED);
+    for (const spec of Object.values(SHARED_TOOL_SPECS)) {
+      if ((spec as { mcpOnly?: boolean }).mcpOnly) continue;
+      names.add((spec as { inAppKey: string }).inAppKey);
+    }
+    return names;
+  }
+
+  it('en and ru label the SAME set of tools', () => {
+    expect(labelledToolNames('en').sort()).toEqual(
+      labelledToolNames('ru').sort(),
+    );
+  });
+
+  it('every labelled tool name is a real in-app tool', () => {
+    const valid = validInAppToolNames();
+    const dead = labelledToolNames('en').filter((n) => !valid.has(n));
+    expect(dead).toEqual([]);
+  });
+
+  it('the guard REDDENS for an unknown label key (mutation check)', () => {
+    const valid = validInAppToolNames();
+    // A hypothetical renamed-away label must be caught.
+    expect(valid.has('getPageRenamedAway')).toBe(false);
+  });
+});

apps/server/src/core/ai-chat/chat-markdown.util.ts

@@ -154,6 +154,17 @@ function toolLabel(name: string, lang: ExportLang): string {
   return LABELS[lang].tools[name] ?? LABELS[lang].ranTool(name);
 }
 
+/**
+ * The tool names that carry a hand-written friendly export label, per language.
+ * Exported for the drift-guard (#494): a label keyed by a tool name that no
+ * longer exists is a DEAD entry (the tool was renamed and now silently falls back
+ * to the generic `ranTool(name)` line). The guard asserts every key here is a
+ * real in-app tool AND that the two languages label the SAME set of tools.
+ */
+export function labelledToolNames(lang: ExportLang): string[] {
+  return Object.keys(LABELS[lang].tools);
+}
+
 /**
  * Stringify an arbitrary tool input/output value for a fenced block. Strings
  * pass through as-is; everything else is pretty-printed JSON, falling back to

apps/server/src/core/ai-chat/dto/ai-chat.dto.ts

@@ -1,4 +1,10 @@
-import { IsOptional, IsString, MaxLength, MinLength } from 'class-validator';
+import {
+  IsISO8601,
+  IsOptional,
+  IsString,
+  MaxLength,
+  MinLength,
+} from 'class-validator';
 
 /** Identify a chat by id (workspace-scoped on the server). */
 export class ChatIdDto {
@@ -37,6 +43,24 @@ export class GetChatMessagesDto {
   cursor?: string;
 }
 
+/**
+ * Delta poll (#491): pull the chat's rows changed since `cursor` (a DB-clock
+ * timestamp from the previous poll) plus the current run fact — the degraded-poll
+ * fallback's payload, replacing the full infinite-query refetch. Omit `cursor` on
+ * the first poll (returns just a fresh cursor to start the chain).
+ */
+export class GetChatDeltaDto {
+  @IsString()
+  chatId: string;
+
+  // ISO-8601 timestamp echoed from the previous poll's response. Validated as
+  // ISO-8601 (not a bare string): a malformed cursor would otherwise reach the
+  // `::timestamptz` cast in findByChatUpdatedAfter and 500 instead of a clean 400.
+  @IsOptional()
+  @IsISO8601()
+  cursor?: string;
+}
+
 /** Resolve the chat bound to a document (the page's most-recent owned chat). */
 export class BoundChatDto {
   @IsString()

apps/server/src/core/ai-chat/external-mcp/dto/create-mcp-server.dto.ts

@@ -37,10 +37,13 @@ export class CreateMcpServerDto {
   @IsObject()
   headers?: Record<string, string>;
 
+  // Omit/null => no restriction; `[]` is persisted verbatim and means deny-all
+  // (zero tools) since #476. @IsOptional() skips validation for null as well,
+  // so an explicit null is accepted.
   @IsOptional()
   @IsArray()
   @IsString({ each: true })
-  toolAllowlist?: string[];
+  toolAllowlist?: string[] | null;
 
   // Admin-authored guidance ("how/when to use this server's tools") injected
   // into the agent system prompt next to the tool descriptions (#180). Trusted,

apps/server/src/core/ai-chat/external-mcp/dto/update-mcp-server.dto.ts

@@ -38,10 +38,13 @@ export class UpdateMcpServerDto {
   @IsObject()
   headers?: Record<string, string>;
 
+  // Absent => unchanged; null => no restriction; `[]` is persisted verbatim
+  // and means deny-all (zero tools) since #476. @IsOptional() skips validation
+  // for null as well, so an explicit null is accepted.
   @IsOptional()
   @IsArray()
   @IsString({ each: true })
-  toolAllowlist?: string[];
+  toolAllowlist?: string[] | null;
 
   // Admin-authored prompt guidance (#180). Absent => unchanged; blank => cleared
   // (stored as null by the repo). Capped to bound prompt/token size.

apps/server/src/core/ai-chat/external-mcp/mcp-allowlist-filter.spec.ts

@@ -0,0 +1,169 @@
+import { type Tool } from 'ai';
+import { McpClientsService } from './mcp-clients.service';
+
+/**
+ * Tool-allowlist filtering semantics on the merged external toolset (#476).
+ *
+ * COVERAGE CHOICE (documented per issue #476): the full corrupt-row chain
+ * (DB value -> repo normalizeRow -> toolsFor filter) is covered on TWO levels
+ * instead of one live-stub-MCP-server integration test:
+ *   (a) apps/server/test/integration/ai-mcp-server-repo.int-spec.ts pins the
+ *       repo read/write semantics against a real Postgres — `[]` round-trips
+ *       as jsonb `[]`, a present-but-corrupt value fails CLOSED to `[]` with
+ *       an error log;
+ *   (b) THIS spec pins what the toolset builder does with the repo's output —
+ *       null = unrestricted, `['alpha']` = only alpha, `[]` (including the
+ *       corrupt-row fallback) = ZERO tools.
+ * Together they prove the end-to-end property "corrupt/empty allowlist can
+ * never widen to all tools" without a live stub HTTP MCP server.
+ *
+ * The drive path mirrors mcp-namespacing.spec.ts: stub the repo's listEnabled,
+ * spy the private `connect` to return a fake client, inspect the merged keys.
+ */
+
+function fakeTool(): Tool {
+  return { description: 'x', inputSchema: undefined } as unknown as Tool;
+}
+
+interface FakeServer {
+  id: string;
+  name: string;
+  transport: string;
+  url: string;
+  headersEnc: string | null;
+  toolAllowlist: string[] | null;
+}
+
+function server(
+  over: Partial<FakeServer> & { id: string; name: string },
+): FakeServer {
+  return {
+    transport: 'http',
+    url: 'https://example.com/mcp',
+    headersEnc: null,
+    toolAllowlist: null,
+    ...over,
+  };
+}
+
+/**
+ * Build a service whose repo returns `servers` and whose fake clients expose
+ * `rawTools` from tools(). Returns the merged tool keys produced by toolsFor.
+ */
+async function mergedKeysFor(
+  servers: FakeServer[],
+  rawTools: Record<string, Tool>,
+): Promise<string[]> {
+  const repoStub = {
+    listEnabled: jest.fn().mockResolvedValue(servers),
+  };
+  const service = new McpClientsService(repoStub as never, {} as never);
+
+  jest
+    .spyOn(
+      service as unknown as { connect: (s: FakeServer) => unknown },
+      'connect',
+    )
+    .mockImplementation(() =>
+      Promise.resolve({
+        tools: () => Promise.resolve(rawTools),
+        close: () => Promise.resolve(),
+      }),
+    );
+
+  const toolset = await service.toolsFor('ws-1');
+  // Release the lease so the service does not hold the fake clients open.
+  await Promise.all(toolset.clients.map((c) => c.close()));
+  return Object.keys(toolset.tools);
+}
+
+describe('external MCP tool-allowlist filtering (via toolsFor, #476)', () => {
+  afterEach(() => jest.restoreAllMocks());
+
+  const RAW = () => ({
+    alpha: fakeTool(),
+    beta: fakeTool(),
+    gamma: fakeTool(),
+  });
+
+  it("['alpha'] lets ONLY alpha through", async () => {
+    const keys = await mergedKeysFor(
+      [server({ id: 'id-1', name: 'srv', toolAllowlist: ['alpha'] })],
+      RAW(),
+    );
+    expect(keys).toEqual(['srv_alpha']);
+  });
+
+  it('null (no restriction) lets every tool through', async () => {
+    const keys = await mergedKeysFor(
+      [server({ id: 'id-1', name: 'srv', toolAllowlist: null })],
+      RAW(),
+    );
+    expect(keys.sort()).toEqual(['srv_alpha', 'srv_beta', 'srv_gamma']);
+  });
+
+  it('[] (deny-all) yields ZERO tools — an empty array is authoritative, not falsy (#476)', async () => {
+    // This is the regression the #476 change guards: `[]` used to fall through
+    // the old `allow.length > 0` check and expose ALL tools. It must expose NONE.
+    const keys = await mergedKeysFor(
+      [server({ id: 'id-1', name: 'srv', toolAllowlist: [] })],
+      RAW(),
+    );
+    expect(keys).toEqual([]);
+  });
+
+  it('the corrupt-row fallback ([] from the repo) also yields ZERO tools (#476)', async () => {
+    // The repo turns a present-but-corrupt tool_allowlist into `[]` (fail-closed,
+    // see normalizeRow in ai-mcp-server.repo.ts + the int-spec); this pins that
+    // the toolset builder honours that fallback as deny-all rather than allow-all.
+    const corruptFallback: string[] = [];
+    const keys = await mergedKeysFor(
+      [server({ id: 'id-1', name: 'srv', toolAllowlist: corruptFallback })],
+      RAW(),
+    );
+    expect(keys).toEqual([]);
+  });
+
+  it('allowlisted names not exposed by the server are ignored (no phantom tools)', async () => {
+    const keys = await mergedKeysFor(
+      [
+        server({
+          id: 'id-1',
+          name: 'srv',
+          toolAllowlist: ['alpha', 'does-not-exist'],
+        }),
+      ],
+      RAW(),
+    );
+    expect(keys).toEqual(['srv_alpha']);
+  });
+
+  it('a deny-all server contributes no prompt instructions (0 tools merged)', async () => {
+    const repoStub = {
+      listEnabled: jest.fn().mockResolvedValue([
+        {
+          ...server({ id: 'id-1', name: 'srv', toolAllowlist: [] }),
+          instructions: 'use the tools wisely',
+        },
+      ]),
+    };
+    const service = new McpClientsService(repoStub as never, {} as never);
+    jest
+      .spyOn(
+        service as unknown as { connect: (s: FakeServer) => unknown },
+        'connect',
+      )
+      .mockImplementation(() =>
+        Promise.resolve({
+          tools: () => Promise.resolve(RAW()),
+          close: () => Promise.resolve(),
+        }),
+      );
+
+    const toolset = await service.toolsFor('ws-1');
+    await Promise.all(toolset.clients.map((c) => c.close()));
+    expect(Object.keys(toolset.tools)).toEqual([]);
+    // mergeNamespaced reported 0 contributed tools, so no guidance is attached.
+    expect(toolset.instructions).toEqual([]);
+  });
+});

apps/server/src/core/ai-chat/external-mcp/mcp-clients.service.ts

@@ -444,9 +444,13 @@ export class McpClientsService {
       try {
         client = await this.connectWithTimeout(server, CONNECT_TIMEOUT_MS);
         const raw = await withTimeout(client.tools(), CONNECT_TIMEOUT_MS);
+        // Allowlist semantics (#476): null/absent = no restriction (all tools);
+        // ANY array — including `[]` — is authoritative, so an EMPTY allowlist
+        // yields ZERO tools (deny-all). Do NOT add a `.length > 0` escape here:
+        // that read `[]` as falsy and silently widened deny-all to allow-all
+        // (the repo also fails corrupt rows closed to `[]` for the same reason).
         const allow = server.toolAllowlist;
-        const picked =
-          Array.isArray(allow) && allow.length > 0 ? pick(raw, allow) : raw;
+        const picked = Array.isArray(allow) ? pick(raw, allow) : raw;
         // Bound each tool's execute with a per-call total-timeout guard before
         // merging, so a single chatty-but-stuck call is aborted after the cap.
         const guarded = wrapToolsWithCallTimeout(picked, callTimeoutMs);

apps/server/src/core/ai-chat/external-mcp/mcp-servers.service.ts

@@ -100,7 +100,8 @@ export class McpServersService {
       transport: dto.transport,
       url: dto.url,
       headersEnc,
-      // undefined => unchanged; [] / value handled by repo (empty => null).
+      // undefined => unchanged; null => no restriction; `[]` is persisted
+      // verbatim and means deny-all (#476).
       toolAllowlist: dto.toolAllowlist,
       // undefined => unchanged; blank => cleared (null) by the repo.
       instructions: dto.instructions,

apps/server/src/core/ai-chat/history-budget.spec.ts

@@ -0,0 +1,266 @@
+import type { ModelMessage } from 'ai';
+import {
+  resolveReplayBudget,
+  isContextOverflowError,
+  estimateMessagesTokens,
+  trimHistoryForReplay,
+  REPLAY_BUDGET_DEFAULT_TOKENS,
+  REPLAY_TRUNCATION_MARKER,
+  REPLAY_TURN_COLLAPSED_MARKER,
+} from './history-budget';
+
+describe('resolveReplayBudget', () => {
+  it('uses floor(0.7 x window) for a configured window (no cap)', () => {
+    // 0.7 x 60k = 42k
+    expect(resolveReplayBudget(60_000)).toEqual({
+      thresholdTokens: 42_000,
+      usedDefault: false,
+    });
+    // 0.7 x 1M = 700k — NOT capped (anti-brick vs the window, not a cost limiter).
+    expect(resolveReplayBudget(1_000_000)).toEqual({
+      thresholdTokens: 700_000,
+      usedDefault: false,
+    });
+  });
+
+  it('accepts the raw ::text stored form', () => {
+    expect(resolveReplayBudget('60000').thresholdTokens).toBe(42_000);
+  });
+
+  // The crux (#490): a chat with NO context window configured must STILL be
+  // budgeted — those are exactly the installs that hit terminal overflow.
+  it('applies the flat default when the window is unset/empty', () => {
+    expect(resolveReplayBudget(undefined)).toEqual({
+      thresholdTokens: REPLAY_BUDGET_DEFAULT_TOKENS,
+      usedDefault: true,
+    });
+    expect(resolveReplayBudget('')).toEqual({
+      thresholdTokens: REPLAY_BUDGET_DEFAULT_TOKENS,
+      usedDefault: true,
+    });
+    expect(resolveReplayBudget('   ')).toEqual({
+      thresholdTokens: REPLAY_BUDGET_DEFAULT_TOKENS,
+      usedDefault: true,
+    });
+  });
+
+  it('treats an explicit 0 as the off-switch (distinct from unset)', () => {
+    expect(resolveReplayBudget(0)).toEqual({
+      thresholdTokens: null,
+      usedDefault: false,
+    });
+    expect(resolveReplayBudget('0')).toEqual({
+      thresholdTokens: null,
+      usedDefault: false,
+    });
+  });
+
+  it('falls back to the default on a negative/garbage value', () => {
+    expect(resolveReplayBudget(-5).usedDefault).toBe(true);
+    expect(resolveReplayBudget('abc').usedDefault).toBe(true);
+  });
+});
+
+describe('isContextOverflowError', () => {
+  it('classifies a real provider 400 context-overflow shape', () => {
+    // OpenAI-compatible shape.
+    expect(
+      isContextOverflowError({
+        statusCode: 400,
+        message:
+          "This model's maximum context length is 128000 tokens. However, your messages resulted in 214000 tokens. Please reduce the length of the messages.",
+      }),
+    ).toBe(true);
+    // Anthropic-style wording.
+    expect(
+      isContextOverflowError({
+        status: 400,
+        message: 'prompt is too long: 250000 tokens > 200000 maximum',
+      }),
+    ).toBe(true);
+    // Nested body + string status.
+    expect(
+      isContextOverflowError({
+        response: { status: '400' },
+        message: 'input is too long for the requested model',
+      }),
+    ).toBe(true);
+    // Error instance with the cause carrying the body.
+    const e = new Error('Bad request');
+    (e as any).statusCode = 400;
+    (e as any).cause = new Error('maximum context window exceeded');
+    expect(isContextOverflowError(e)).toBe(true);
+  });
+
+  it('does NOT classify unrelated 400s or auth/rate-limit errors', () => {
+    expect(
+      isContextOverflowError({ statusCode: 400, message: 'invalid tool schema' }),
+    ).toBe(false);
+    expect(
+      isContextOverflowError({
+        statusCode: 429,
+        message: 'context length exceeded but rate limited',
+      }),
+    ).toBe(false);
+    expect(isContextOverflowError({ statusCode: 500, message: 'server error' })).toBe(
+      false,
+    );
+    expect(isContextOverflowError(undefined)).toBe(false);
+    expect(isContextOverflowError('some random string')).toBe(false);
+  });
+});
+
+// Helpers to build ModelMessage fixtures in the ai@6 shape.
+const userMsg = (text: string): ModelMessage =>
+  ({ role: 'user', content: [{ type: 'text', text }] }) as ModelMessage;
+const assistantMsg = (
+  text: string,
+  toolCallId?: string,
+  toolName?: string,
+): ModelMessage =>
+  ({
+    role: 'assistant',
+    content: [
+      { type: 'text', text },
+      ...(toolCallId
+        ? [{ type: 'tool-call', toolCallId, toolName, input: {} }]
+        : []),
+    ],
+  }) as ModelMessage;
+const toolMsg = (
+  toolCallId: string,
+  toolName: string,
+  value: unknown,
+): ModelMessage =>
+  ({
+    role: 'tool',
+    content: [
+      { type: 'tool-result', toolCallId, toolName, output: { type: 'json', value } },
+    ],
+  }) as ModelMessage;
+
+describe('trimHistoryForReplay', () => {
+  it('null budget disables trimming (returns the same reference)', () => {
+    const msgs = [userMsg('hi'), assistantMsg('yo')];
+    const r = trimHistoryForReplay(msgs, null);
+    expect(r.trimmed).toBe(false);
+    expect(r.messages).toBe(msgs);
+  });
+
+  it('leaves history under budget untouched (same reference)', () => {
+    const msgs = [userMsg('hi'), assistantMsg('a short answer')];
+    const r = trimHistoryForReplay(msgs, 100_000);
+    expect(r.trimmed).toBe(false);
+    expect(r.messages).toBe(msgs);
+  });
+
+  it('truncates OLD tool outputs but keeps recent turns full', () => {
+    const big = 'X'.repeat(40_000); // ~16k tokens on its own
+    const msgs: ModelMessage[] = [];
+    // 6 OLD turns (indices 0..5), each with a huge tool output.
+    for (let i = 0; i < 6; i++) {
+      msgs.push(userMsg(`old q${i}`));
+      msgs.push(assistantMsg('looking', `c${i}`, 'getPage'));
+      msgs.push(toolMsg(`c${i}`, 'getPage', { body: big }));
+      msgs.push(assistantMsg(`old a${i}`));
+    }
+    // 3 small recent turns, then the CURRENT turn with its own huge tool output.
+    // With REPLAY_KEEP_RECENT_TURNS=4 the last 4 user-turns stay full, so only
+    // these small recent turns + the current big one are kept full; the 6 old
+    // turns above fall in the trim region.
+    for (let i = 0; i < 3; i++) {
+      msgs.push(userMsg(`recent q${i}`));
+      msgs.push(assistantMsg(`recent a${i}`));
+    }
+    msgs.push(userMsg('current q'));
+    msgs.push(assistantMsg('looking', 'cR', 'getPage'));
+    msgs.push(toolMsg('cR', 'getPage', { body: big }));
+    msgs.push(assistantMsg('current a'));
+
+    // Budget large enough that phase-1 tool truncation alone brings it under.
+    const r = trimHistoryForReplay(msgs, 30_000);
+    expect(r.trimmed).toBe(true);
+    const flat = JSON.stringify(r.messages);
+    // The CURRENT turn's tool output survives in full.
+    expect(flat).toContain(big);
+    // Old outputs were truncated with the marker.
+    expect(flat).toContain(REPLAY_TRUNCATION_MARKER);
+    // Phase 1 sufficed: the oldest turns were NOT collapsed.
+    expect(flat).not.toContain(REPLAY_TURN_COLLAPSED_MARKER);
+    expect(estimateMessagesTokens(r.messages)).toBeLessThan(
+      estimateMessagesTokens(msgs),
+    );
+  });
+
+  it('collapses the oldest turns when tool truncation is not enough', () => {
+    // Many turns with LARGE assistant TEXT (not tool output) so phase 1 can't help.
+    const bigText = 'слово '.repeat(8_000); // large Cyrillic text per turn
+    const msgs: ModelMessage[] = [];
+    for (let i = 0; i < 12; i++) {
+      msgs.push(userMsg(`q${i}`));
+      msgs.push(assistantMsg(bigText));
+    }
+    const r = trimHistoryForReplay(msgs, 30_000);
+    expect(r.trimmed).toBe(true);
+    // Oldest turns collapsed; result fits (best-effort) and is much smaller.
+    expect(estimateMessagesTokens(r.messages)).toBeLessThan(
+      estimateMessagesTokens(msgs),
+    );
+    // The LAST turn's text is preserved in full (recent turns stay full).
+    expect(JSON.stringify(r.messages[r.messages.length - 1])).toContain(bigText);
+  });
+
+  it('is deterministic / byte-stable for identical inputs', () => {
+    const big = 'Y'.repeat(30_000);
+    const build = (): ModelMessage[] => {
+      const m: ModelMessage[] = [];
+      for (let i = 0; i < 10; i++) {
+        m.push(userMsg(`q${i}`));
+        m.push(assistantMsg('t', `c${i}`, 'getPage'));
+        m.push(toolMsg(`c${i}`, 'getPage', { body: big }));
+      }
+      return m;
+    };
+    const a = trimHistoryForReplay(build(), 15_000);
+    const b = trimHistoryForReplay(build(), 15_000);
+    expect(JSON.stringify(a.messages)).toBe(JSON.stringify(b.messages));
+  });
+
+  it('never leaves an unpaired tool-call after collapsing (balanced history)', () => {
+    const big = 'Z'.repeat(40_000);
+    const msgs: ModelMessage[] = [];
+    for (let i = 0; i < 10; i++) {
+      msgs.push(userMsg(`q${i}`));
+      msgs.push(assistantMsg('t', `c${i}`, 'getPage'));
+      msgs.push(toolMsg(`c${i}`, 'getPage', { body: big }));
+    }
+    const r = trimHistoryForReplay(msgs, 8_000);
+    // Count tool-call vs tool-result parts in the trimmed output.
+    let calls = 0;
+    let results = 0;
+    for (const m of r.messages) {
+      if (!Array.isArray(m.content)) continue;
+      for (const p of m.content as Array<{ type?: string }>) {
+        if (p.type === 'tool-call') calls++;
+        if (p.type === 'tool-result' || p.type === 'tool-error') results++;
+      }
+    }
+    // Every surviving tool-call has a surviving result (collapsing drops BOTH).
+    expect(calls).toBe(results);
+    // Collapsed turns carry the marker.
+    expect(JSON.stringify(r.messages)).toContain(REPLAY_TURN_COLLAPSED_MARKER);
+  });
+
+  it('respects the provider fact: under-budget contextTokens skips trimming', () => {
+    const big = 'W'.repeat(60_000);
+    const msgs = [
+      userMsg('q'),
+      assistantMsg('t', 'c1', 'getPage'),
+      toolMsg('c1', 'getPage', { body: big }),
+    ];
+    // char-estimate is high, but the provider says we are well under budget.
+    const r = trimHistoryForReplay(msgs, 100_000, 5_000);
+    expect(r.trimmed).toBe(false);
+    expect(r.messages).toBe(msgs);
+  });
+});

apps/server/src/core/ai-chat/history-budget.ts

@@ -0,0 +1,375 @@
+/**
+ * History-replay token budget (#490).
+ *
+ * The whole persisted conversation is replayed to the provider on EVERY turn, so
+ * a long chat eventually exceeds the model's context window and the provider 400s
+ * on every turn — terminally (the chat "bricks"). This module bounds the replayed
+ * history at REPLAY TIME only: it never mutates what is persisted (the DB stays
+ * the full record), and its output is a deterministic, byte-stable function of its
+ * input so the trimmed prefix is identical turn to turn (provider prompt-cache
+ * friendliness — real money on long chats).
+ *
+ * The PRIMARY signal is the provider's own fact: `metadata.contextTokens` from the
+ * last turn. The chars-based {@link estimateTokens} (shared with the client) is
+ * used only for the DELTA of not-yet-sent messages, to decide WHAT to trim, and as
+ * the fallback for chats with no usage yet.
+ */
+import type { ModelMessage } from 'ai';
+import { estimateTokens } from '@docmost/token-estimate';
+
+/** Flat default budget when no context window is configured (tokens). */
+export const REPLAY_BUDGET_DEFAULT_TOKENS = 100_000;
+/** Fraction of a configured context window used as the budget. */
+export const REPLAY_BUDGET_WINDOW_FRACTION = 0.7;
+/**
+ * Fraction of the normal budget used for the REACTIVE re-trim after a provider
+ * context-overflow 400 — the preventive estimate under-counted, so cut harder.
+ */
+export const REPLAY_AGGRESSIVE_FRACTION = 0.5;
+/**
+ * Turns (a user message + its assistant/tool replies) kept FULL at the tail,
+ * including the current one — never trimmed. Older turns are compacted first.
+ */
+export const REPLAY_KEEP_RECENT_TURNS = 4;
+/** Leading chars kept from a truncated old tool output. */
+export const REPLAY_TOOL_OUTPUT_HEAD = 800;
+/** Trailing chars kept from a truncated old tool output. */
+export const REPLAY_TOOL_OUTPUT_TAIL = 300;
+/** Marker inserted where an old tool output was truncated for replay. */
+export const REPLAY_TRUNCATION_MARKER =
+  '[…truncated for replay; call the tool again to read the full output]';
+/** Marker for a whole old turn collapsed to its text. */
+export const REPLAY_TURN_COLLAPSED_MARKER =
+  '[earlier tool activity omitted for replay]';
+
+export interface ReplayBudget {
+  /** Token threshold above which replay history is trimmed; `null` = OFF. */
+  thresholdTokens: number | null;
+  /** True when the flat default was used (no context window configured). */
+  usedDefault: boolean;
+}
+
+/**
+ * Resolve the replay budget from the RAW stored `chatContextWindow` (text/number).
+ *  - a positive value  -> `floor(fraction × window)` (NO cap — the budgeter is
+ *    anti-brick protection against the window itself, not a cost/economy limiter,
+ *    exactly as the codebase already treats maxOutputTokens; the reactive branch
+ *    still guarantees anti-brick regardless of how high this budget is)
+ *  - explicit `0`       -> OFF (admin opt-out; `null` threshold)
+ *  - unset/empty/invalid-> the flat default (still protects — the installations
+ *    that hit terminal overflow are exactly the ones that never set a window)
+ *
+ * Note the raw value is needed because the parsed `chatContextWindow` collapses
+ * both `0` and unset to `undefined`, which would erase the explicit off-switch.
+ */
+export function resolveReplayBudget(rawContextWindow: unknown): ReplayBudget {
+  let n: number | undefined;
+  if (typeof rawContextWindow === 'number') {
+    n = rawContextWindow;
+  } else if (typeof rawContextWindow === 'string') {
+    const t = rawContextWindow.trim();
+    n = t === '' ? undefined : Number(t);
+  }
+  // Unset / empty / non-numeric / negative -> flat default (the protective case).
+  if (n === undefined || !Number.isFinite(n) || n < 0) {
+    return { thresholdTokens: REPLAY_BUDGET_DEFAULT_TOKENS, usedDefault: true };
+  }
+  // Explicit 0 -> off-switch.
+  if (n === 0) {
+    return { thresholdTokens: null, usedDefault: false };
+  }
+  return {
+    thresholdTokens: Math.floor(REPLAY_BUDGET_WINDOW_FRACTION * n),
+    usedDefault: false,
+  };
+}
+
+/**
+ * The effective replay threshold for THIS turn, given the base budget and whether
+ * the PREVIOUS turn hit a context-overflow 400 (the reactive-recovery signal,
+ * `metadata.replayOverflow`). On recovery the base budget is scaled down by
+ * {@link REPLAY_AGGRESSIVE_FRACTION}: the overflowing turn produced no usage
+ * signal, so the preventive estimate under-counted and a normal-threshold trim may
+ * not shrink enough to fit — this harder cut is what un-bricks the chat.
+ *
+ * A `null` base budget (trimming OFF) is passed through unchanged: an explicit
+ * off-switch is never overridden by the recovery path.
+ */
+export function resolveEffectiveReplayThreshold(
+  thresholdTokens: number | null,
+  priorOverflowed: boolean,
+): number | null {
+  if (!priorOverflowed || thresholdTokens == null) return thresholdTokens;
+  return Math.floor(thresholdTokens * REPLAY_AGGRESSIVE_FRACTION);
+}
+
+/**
+ * True when a provider error is a CONTEXT-OVERFLOW rejection (the prompt exceeds
+ * the model's window). Providers surface this as an HTTP 400 with a recognizable
+ * message; match both the status and the message patterns robustly across
+ * OpenAI-compatible / Anthropic / Gemini wordings, since the exact shape varies.
+ */
+export function isContextOverflowError(error: unknown): boolean {
+  const status = extractStatus(error);
+  const msg = extractMessage(error).toLowerCase();
+  // Message patterns seen across providers for "prompt too long".
+  const overflowPattern =
+    /context (?:length|window)|maximum context|too many tokens|too large for|reduce the length|prompt is too long|input (?:is )?too long|exceeds? the (?:maximum )?(?:context|token)|maximum.*tokens|string too long/;
+  if (!overflowPattern.test(msg)) return false;
+  // A 400/413 with an overflow-shaped message is an overflow. Some providers
+  // omit/rewrite the status, so accept the message match when the status is
+  // unknown, but reject it for auth/rate-limit statuses that never mean overflow.
+  if (status === 400 || status === 413) return true;
+  if (status === 401 || status === 403 || status === 429) return false;
+  return true;
+}
+
+function extractStatus(error: unknown): number | undefined {
+  if (!error || typeof error !== 'object') return undefined;
+  const e = error as Record<string, unknown>;
+  for (const k of ['statusCode', 'status']) {
+    const v = e[k];
+    if (typeof v === 'number') return v;
+    if (typeof v === 'string' && /^\d+$/.test(v)) return Number(v);
+  }
+  // Nested (e.g. { response: { status } } / { cause: { statusCode } }).
+  for (const k of ['response', 'cause', 'data']) {
+    const nested = e[k];
+    if (nested && typeof nested === 'object') {
+      const s = extractStatus(nested);
+      if (s !== undefined) return s;
+    }
+  }
+  return undefined;
+}
+
+function extractMessage(error: unknown): string {
+  if (error == null) return '';
+  if (typeof error === 'string') return error;
+  if (error instanceof Error) {
+    // Include nested causes (provider libs wrap the real body in `cause`).
+    const cause = (error as { cause?: unknown }).cause;
+    return `${error.message} ${cause ? extractMessage(cause) : ''}`;
+  }
+  if (typeof error === 'object') {
+    const e = error as Record<string, unknown>;
+    const parts: string[] = [];
+    for (const k of ['message', 'error', 'body', 'responseBody', 'data']) {
+      const v = e[k];
+      if (typeof v === 'string') parts.push(v);
+      else if (v && typeof v === 'object') parts.push(extractMessage(v));
+    }
+    return parts.join(' ');
+  }
+  return String(error);
+}
+
+/** Rough token size of a ModelMessage array via the shared chars estimator. */
+export function estimateMessagesTokens(
+  messages: ReadonlyArray<ModelMessage>,
+): number {
+  let total = 0;
+  for (const m of messages) {
+    total += estimateTokens(serializeContent(m.content));
+  }
+  return total;
+}
+
+function serializeContent(content: unknown): string {
+  if (typeof content === 'string') return content;
+  try {
+    return JSON.stringify(content) ?? '';
+  } catch {
+    return '';
+  }
+}
+
+/** Deep JSON string of an arbitrary value, bounded so estimation never throws. */
+function stringifyValue(value: unknown): string {
+  if (typeof value === 'string') return value;
+  try {
+    return JSON.stringify(value) ?? String(value);
+  } catch {
+    return String(value);
+  }
+}
+
+export interface TrimResult {
+  messages: ModelMessage[];
+  /** Whether any trimming was applied. */
+  trimmed: boolean;
+  /** Estimated tokens of the returned messages (chars-based). */
+  estimatedTokens: number;
+}
+
+/**
+ * Bound the replayed history to `budgetTokens`, deterministically. Returns the
+ * SAME array reference (no copy) when nothing needs trimming, so the common case
+ * is free and byte-identical. Trimming order (spec #490):
+ *   1. truncate OLD turns' tool outputs (head+tail + marker) — the bulk of the size
+ *   2. mechanically collapse the OLDEST turns to their text (concatenation, no LLM)
+ *   3. the current + last {@link REPLAY_KEEP_RECENT_TURNS} turns stay FULL
+ *
+ * `budgetTokens === null` disables trimming. `priorContextTokens` (the provider's
+ * fact from last turn) short-circuits the decision: when it is known and already
+ * under budget we skip trimming even if the char-estimate is higher (the provider
+ * count is authoritative). The char-estimate drives WHAT to cut.
+ */
+export function trimHistoryForReplay(
+  messages: ModelMessage[],
+  budgetTokens: number | null,
+  priorContextTokens?: number,
+): TrimResult {
+  if (budgetTokens == null) {
+    return { messages, trimmed: false, estimatedTokens: 0 };
+  }
+  const estimated = estimateMessagesTokens(messages);
+  // Decision signal: prefer the provider's fact (last turn's contextTokens) plus
+  // the estimated delta of the messages appended since; fall back to the pure
+  // char-estimate for a chat with no usage yet.
+  const projected =
+    priorContextTokens != null
+      ? Math.max(priorContextTokens, estimated)
+      : estimated;
+  if (projected <= budgetTokens) {
+    return { messages, trimmed: false, estimatedTokens: estimated };
+  }
+
+  // The tail we always keep full: from the Nth-from-last user message onward.
+  const boundary = recentBoundaryIndex(messages, REPLAY_KEEP_RECENT_TURNS);
+  const tail = messages.slice(boundary);
+  let head = messages.slice(0, boundary).map(cloneMessage);
+
+  // Phase 1: truncate old tool outputs.
+  for (const m of head) {
+    if (m.role === 'tool') truncateToolMessage(m);
+  }
+  let out = [...head, ...tail];
+  let est = estimateMessagesTokens(out);
+  if (est <= budgetTokens) {
+    return { messages: out, trimmed: true, estimatedTokens: est };
+  }
+
+  // Phase 2: collapse the oldest turns (in `head`) to their text, one at a time,
+  // from the oldest, until we fit or the whole head is collapsed.
+  const turns = splitTurns(head);
+  const collapsed: ModelMessage[] = [];
+  let i = 0;
+  for (; i < turns.length; i++) {
+    if (est <= budgetTokens) break;
+    collapsed.push(...collapseTurn(turns[i]));
+    // Re-estimate the whole prospective output.
+    const remaining = turns.slice(i + 1).flat();
+    out = [...collapsed, ...remaining, ...tail];
+    est = estimateMessagesTokens(out);
+  }
+  // Include any turns we didn't need to collapse.
+  const remaining = turns.slice(i).flat();
+  out = [...collapsed, ...remaining, ...tail];
+  est = estimateMessagesTokens(out);
+  return { messages: out, trimmed: true, estimatedTokens: est };
+}
+
+/** Index of the first message of the Nth-from-last user turn (0 if fewer). */
+function recentBoundaryIndex(
+  messages: ReadonlyArray<ModelMessage>,
+  keepTurns: number,
+): number {
+  const userIdx: number[] = [];
+  for (let i = 0; i < messages.length; i++) {
+    if (messages[i].role === 'user') userIdx.push(i);
+  }
+  if (userIdx.length <= keepTurns) return 0;
+  return userIdx[userIdx.length - keepTurns];
+}
+
+/** Split a message list into turns; each turn starts at a `user` message. */
+function splitTurns(messages: ModelMessage[]): ModelMessage[][] {
+  const turns: ModelMessage[][] = [];
+  for (const m of messages) {
+    if (m.role === 'user' || turns.length === 0) turns.push([m]);
+    else turns[turns.length - 1].push(m);
+  }
+  return turns;
+}
+
+/**
+ * Collapse a whole turn to its plain text (mechanical concatenation, not an LLM
+ * summary). Keeps the user message; replaces the assistant/tool messages with a
+ * single assistant text message = the assistant's concatenated text + a marker
+ * when tool activity was dropped. Dropping BOTH the tool-call and tool-result
+ * parts together keeps the rebuilt history balanced (no unpaired calls).
+ */
+function collapseTurn(turn: ModelMessage[]): ModelMessage[] {
+  const out: ModelMessage[] = [];
+  let assistantText = '';
+  let hadTools = false;
+  for (const m of turn) {
+    if (m.role === 'user') {
+      out.push(m);
+    } else if (m.role === 'assistant') {
+      const { text, tools } = extractAssistantText(m.content);
+      assistantText += text;
+      hadTools = hadTools || tools;
+    } else if (m.role === 'tool') {
+      hadTools = true;
+    } else {
+      out.push(m);
+    }
+  }
+  const note =
+    (assistantText ? assistantText.trimEnd() : '') +
+    (hadTools
+      ? `${assistantText ? '\n\n' : ''}${REPLAY_TURN_COLLAPSED_MARKER}`
+      : '');
+  if (note) out.push({ role: 'assistant', content: note } as ModelMessage);
+  return out;
+}
+
+function extractAssistantText(content: unknown): {
+  text: string;
+  tools: boolean;
+} {
+  if (typeof content === 'string') return { text: content, tools: false };
+  if (!Array.isArray(content)) return { text: '', tools: false };
+  let text = '';
+  let tools = false;
+  for (const part of content) {
+    const type = (part as { type?: string })?.type;
+    if (type === 'text') text += (part as { text?: string }).text ?? '';
+    else if (type === 'tool-call') tools = true;
+  }
+  return { text, tools };
+}
+
+/** Truncate every tool-result output in a `tool` message to head+tail+marker. */
+function truncateToolMessage(message: ModelMessage): void {
+  const content = message.content;
+  if (!Array.isArray(content)) return;
+  for (const part of content) {
+    const p = part as { type?: string; output?: { type?: string; value?: unknown } };
+    if (p.type !== 'tool-result' && p.type !== 'tool-error') continue;
+    if (!p.output) continue;
+    const raw = stringifyValue(p.output.value);
+    const budget = REPLAY_TOOL_OUTPUT_HEAD + REPLAY_TOOL_OUTPUT_TAIL;
+    if (raw.length <= budget + REPLAY_TRUNCATION_MARKER.length) continue;
+    const truncated =
+      raw.slice(0, REPLAY_TOOL_OUTPUT_HEAD) +
+      `\n${REPLAY_TRUNCATION_MARKER}\n` +
+      raw.slice(raw.length - REPLAY_TOOL_OUTPUT_TAIL);
+    // Represent the shrunk output as a text output (a valid tool-result output).
+    p.output = { type: 'text', value: truncated };
+  }
+}
+
+/** Shallow-ish clone so trimming never mutates the caller's (persisted-derived)
+ *  message objects — only the OLD region is cloned before it is edited. */
+function cloneMessage(m: ModelMessage): ModelMessage {
+  if (typeof m.content === 'string') return { ...m };
+  return {
+    ...m,
+    content: (m.content as unknown[]).map((p) =>
+      p && typeof p === 'object' ? { ...(p as object) } : p,
+    ),
+  } as ModelMessage;
+}

apps/server/src/core/ai-chat/output-degeneration.spec.ts

@@ -9,9 +9,73 @@ import {
   DEGENERATION_CHECK_STEP,
   REPEATED_LINES_THRESHOLD,
   MIN_PERIOD_REPEATS,
+  degenerationThresholds,
 } from './output-degeneration';
 import { AiChatService } from './ai-chat.service';
 
+// Part A (#495 iter10): the detector thresholds are env-tunable. These drive the
+// resolver against real repeat-count shapes and mutation-verify that the env
+// override actually changes the trigger point (not a vacuous read).
+describe('degeneration thresholds are env-configurable', () => {
+  const VARS = [
+    'AI_CHAT_DEGENERATION_REPEATED_LINES',
+    'AI_CHAT_DEGENERATION_PERIOD_MAX_LEN',
+    'AI_CHAT_DEGENERATION_PERIOD_MIN_REPEATS',
+    'AI_CHAT_DEGENERATION_CHECK_STEP',
+  ];
+  const saved: Record<string, string | undefined> = {};
+  beforeEach(() => {
+    for (const v of VARS) saved[v] = process.env[v];
+  });
+  afterEach(() => {
+    for (const v of VARS) {
+      if (saved[v] === undefined) delete process.env[v];
+      else process.env[v] = saved[v];
+    }
+  });
+
+  it('defaults to the compiled constants when unset', () => {
+    for (const v of VARS) delete process.env[v];
+    expect(degenerationThresholds()).toEqual({
+      repeatedLines: REPEATED_LINES_THRESHOLD,
+      maxPeriodLen: 150,
+      minPeriodRepeats: MIN_PERIOD_REPEATS,
+      checkStep: DEGENERATION_CHECK_STEP,
+    });
+  });
+
+  it('falls back to the default on blank / invalid / non-positive values', () => {
+    for (const bad of ['', '  ', 'abc', '0', '-3', '1.5']) {
+      process.env.AI_CHAT_DEGENERATION_REPEATED_LINES = bad;
+      // '1.5' floors to 1 (still ≥1, valid); every other bad value → default.
+      const expected = bad === '1.5' ? 1 : REPEATED_LINES_THRESHOLD;
+      expect(degenerationThresholds().repeatedLines).toBe(expected);
+    }
+  });
+
+  it('a RAISED check-step suppresses a burst the default would have flagged', () => {
+    // A ~3.3KB periodic burst is periodic-degenerate, but shouldCheckDegeneration
+    // is the throttle gate. Default checkStep=2000 arms on it; raising the step
+    // above the burst size means the throttle never re-fires for it.
+    const burstLen = 'loadTools.\n'.repeat(300).length; // ~3300
+    delete process.env.AI_CHAT_DEGENERATION_CHECK_STEP;
+    expect(shouldCheckDegeneration(burstLen, 0)).toBe(true); // default 2000
+    process.env.AI_CHAT_DEGENERATION_CHECK_STEP = String(burstLen + 1);
+    expect(shouldCheckDegeneration(burstLen, 0)).toBe(false); // raised gate
+  });
+
+  it('a LOWERED repeated-lines threshold trips on a shorter identical-line run', () => {
+    // 8 identical lines: below the default 25 (rule 1) and below the periodic
+    // rule's 20 repeats — so isDegenerateOutput is false by default.
+    const shortRun = 'x\n'.repeat(8);
+    delete process.env.AI_CHAT_DEGENERATION_REPEATED_LINES;
+    expect(isDegenerateOutput(shortRun)).toBe(false);
+    // Lower rule 1 to 5 → the 8-line run now trips.
+    process.env.AI_CHAT_DEGENERATION_REPEATED_LINES = '5';
+    expect(isDegenerateOutput(shortRun)).toBe(true);
+  });
+});
+
 // Mock ONLY streamText so we can capture the onChunk/onStepFinish callbacks the
 // service registers and drive them by hand; every other `ai` export the service
 // uses (convertToModelMessages, stepCountIs, …) stays real.

apps/server/src/core/ai-chat/output-degeneration.ts

@@ -23,6 +23,54 @@ export const MAX_PERIOD_LEN = 150;
 /** Rule 2: minimum number of consecutive block repeats to trigger. */
 export const MIN_PERIOD_REPEATS = 20;
 
+/**
+ * Read a positive-integer threshold from an env var, falling back to `fallback`
+ * on unset/blank/invalid/non-positive. Mirrors the `AI_STREAM_PRE_RESPONSE_RETRIES`
+ * resolver in `ai-streaming-fetch.ts`: read the RAW string first so a blank value
+ * is treated as "unset" (→ fallback) rather than coercing to 0. Thresholds must
+ * stay ≥ 1 — a 0/negative would make the detector fire on any text (or never), so
+ * a bad value degrades to the safe compiled default instead. Env-tunable so an
+ * operator can retune the anti-babble guard (#444) without a redeploy, following
+ * the `AI_CHAT_FINAL_STEP_LOCKDOWN` toggle convention.
+ */
+function envThreshold(name: string, fallback: number): number {
+  const rawStr = process.env[name];
+  if (rawStr === undefined || rawStr.trim() === '') return fallback;
+  const raw = Number(rawStr);
+  return Number.isFinite(raw) && raw >= 1 ? Math.floor(raw) : fallback;
+}
+
+/**
+ * Resolve the degeneration-detector thresholds from the environment, each
+ * defaulting to the compiled constant above. Read fresh per call (not cached at
+ * import) so a test — or a runtime env change — takes effect deterministically.
+ */
+export function degenerationThresholds(): {
+  repeatedLines: number;
+  maxPeriodLen: number;
+  minPeriodRepeats: number;
+  checkStep: number;
+} {
+  return {
+    repeatedLines: envThreshold(
+      'AI_CHAT_DEGENERATION_REPEATED_LINES',
+      REPEATED_LINES_THRESHOLD,
+    ),
+    maxPeriodLen: envThreshold(
+      'AI_CHAT_DEGENERATION_PERIOD_MAX_LEN',
+      MAX_PERIOD_LEN,
+    ),
+    minPeriodRepeats: envThreshold(
+      'AI_CHAT_DEGENERATION_PERIOD_MIN_REPEATS',
+      MIN_PERIOD_REPEATS,
+    ),
+    checkStep: envThreshold(
+      'AI_CHAT_DEGENERATION_CHECK_STEP',
+      DEGENERATION_CHECK_STEP,
+    ),
+  };
+}
+
 /**
  * Rule 1 — ≥`REPEATED_LINES_THRESHOLD` consecutive IDENTICAL non-empty lines at
  * the tail. Catches the classic newline-delimited loop ("loadTools.\n" ×N).
@@ -128,7 +176,11 @@ export function hasPeriodicTail(
  * Pure — the caller owns the abort side effect.
  */
 export function isDegenerateOutput(text: string): boolean {
-  return hasRepeatedLineRun(text) || hasPeriodicTail(text);
+  const cfg = degenerationThresholds();
+  return (
+    hasRepeatedLineRun(text, cfg.repeatedLines) ||
+    hasPeriodicTail(text, cfg.maxPeriodLen, cfg.minPeriodRepeats)
+  );
 }
 
 /**
@@ -154,7 +206,7 @@ export function shouldCheckDegeneration(
   textLen: number,
   lastCheckLen: number,
 ): boolean {
-  return textLen - lastCheckLen >= DEGENERATION_CHECK_STEP;
+  return textLen - lastCheckLen >= degenerationThresholds().checkStep;
 }
 
 /**

apps/server/src/core/ai-chat/public-share-chat.service.ts

@@ -307,6 +307,10 @@ export class PublicShareChatService {
         system,
         messages: modelMessages,
         tools,
+        // Pin the AI SDK per-request retry budget explicitly (matches the SDK
+        // default of 2). Connection arithmetic: (1 + maxRetries) × (1 +
+        // AI_STREAM_PRE_RESPONSE_RETRIES) worst-case connects per turn.
+        maxRetries: 2,
         // Bound the agent loop for anonymous callers.
         stopWhen: stepCountIs(5),
         // Cap per-request output so one anonymous call cannot run up the provider

apps/server/src/core/ai-chat/tools/ai-chat-tools.service.ts

@@ -426,6 +426,7 @@ export class AiChatToolsService {
     const {
       sharedToolSpecs,
       createCommentSignalTracker,
+      createListCommentsProbe,
       searchShapes,
       getGuideSection,
     } = await loadDocmostMcp();
@@ -718,7 +719,11 @@ export class AiChatToolsService {
       if (spec.mcpOnly) continue;
       if (spec.inlineBothHosts) continue;
       const run = spec.inAppExecute ?? spec.execute;
-      if (!run) continue; // defensive: a shared spec always carries one of them.
+      // Guaranteed present by assertEverySpecIsRegisterable() (#494), which runs
+      // at tool-specs module load and throws if a non-inline spec the in-app host
+      // registers carries neither inAppExecute nor execute — so this can no longer
+      // silently drop a mis-declared tool. Kept as a type-narrowing guard.
+      if (!run) continue;
       tools[spec.inAppKey] = sharedTool(
         spec,
         (async (args) =>
@@ -764,35 +769,21 @@ export class AiChatToolsService {
     // wrapper below) so the race governs the whole call. The client carries the
     // per-call composite signal via setToolAbortSignal.
     const capMs = inAppToolCallCapMs();
-    if (!createCommentSignalTracker) {
+    // The signal needs BOTH the tracker factory AND the shared count-source probe
+    // factory (#494). Either being absent (a stale @docmost/mcp build or a mocked
+    // loader) => signal disabled, tool results byte-identical.
+    if (!createCommentSignalTracker || !createListCommentsProbe) {
       return wrapInAppToolsWithCap(tools, client, capMs);
     }
 
+    // Shared probe (#494): the SAME factory the standalone MCP host uses, so the
+    // in-app probe body is no longer a hand-mirror that could drift (counting the
+    // full feed newer than the watermark, labelling a hit with the light page
+    // title). `client` supplies the loopback listComments/getPageRaw reads.
     const tracker = createCommentSignalTracker({
-      probe: async (pageId: string, sinceMs: number) => {
-        const { items } = await client.listComments(pageId, true);
-        const count = (items as Array<{ createdAt?: string }>).filter((c) => {
-          const created = c?.createdAt ? new Date(c.createdAt).getTime() : NaN;
-          return Number.isFinite(created) && created > sinceMs;
-        }).length;
-        let title: string | undefined;
-        if (count > 0) {
-          // Title labels the signal; untrusted, defanged by the shared builder.
-          // Fetched only on a hit so the no-signal path never pays for it. Uses
-          // the LIGHT raw page info (title only) — mirroring the standalone MCP
-          // probe's getPageRaw — instead of the heavy getPage (which also renders
-          // Markdown + subpages) just to read one field.
-          try {
-            const res = (await client.getPageRaw(pageId)) as {
-              title?: string;
-            } | null;
-            title = res?.title ?? undefined;
-          } catch {
-            // Title is optional — omit it when the page can't be fetched.
-          }
-        }
-        return { count, title };
-      },
+      probe: createListCommentsProbe(
+        client as unknown as Parameters<typeof createListCommentsProbe>[0],
+      ),
     });
 
     return wrapInAppToolsWithCap(

apps/server/src/core/ai-chat/tools/comment-signal-inapp.spec.ts

@@ -21,7 +21,10 @@ import { SHARED_TOOL_SPECS } from '../../../../../../packages/mcp/src/tool-specs
 // The REAL shared tracker factory, imported from source (same cross-boundary
 // approach the tool-specs spec uses) so the in-app wiring is exercised against
 // exactly the watermark/debounce/injection-safe logic the package ships.
-import { createCommentSignalTracker } from '../../../../../../packages/mcp/src/comment-signal';
+import {
+  createCommentSignalTracker,
+  createListCommentsProbe,
+} from '../../../../../../packages/mcp/src/comment-signal';
 // The REAL client-side citation extractor: proves that the passive signal does
 // NOT strip a tool's citations (the #417 in-app regression this spec guards).
 import { toolCitations } from '../../../../../../apps/client/src/features/ai-chat/utils/tool-parts';
@@ -284,9 +287,13 @@ describe('AiChatToolsService forUser + comment signal (real tracker)', () => {
         return fakeClient as DocmostClientLike;
       } as unknown as loader.DocmostClientCtor,
       sharedToolSpecs: SHARED_TOOL_SPECS as unknown as Record<string, loader.SharedToolSpec>,
-      // Wire the REAL factory so the in-app path is exercised end to end.
+      // Wire the REAL factories so the in-app path is exercised end to end —
+      // including the shared count-source probe (#494) the service now builds the
+      // tracker's `probe` from.
       createCommentSignalTracker:
         createCommentSignalTracker as unknown as loader.CommentSignalTrackerFactory,
+      createListCommentsProbe:
+        createListCommentsProbe as unknown as loader.CreateListCommentsProbeFn,
       // Pure no-network draw.io helpers (#424) — required on the loader return;
       // this comment-signal test doesn't exercise them, so no-op stubs suffice.
       searchShapes: (() => []) as unknown as loader.SearchShapesFn,

apps/server/src/core/ai-chat/tools/docmost-client.loader.ts

@@ -150,6 +150,27 @@ export type CommentSignalTrackerFactory = (options: {
   debounceMs?: number;
 }) => CommentSignalTrackerLike;
 
+/**
+ * Local mirror of `@docmost/mcp`'s `createListCommentsProbe` (#494): the SHARED
+ * count-source probe both hosts use, so the in-app probe body is no longer a
+ * hand-copy of the standalone MCP one. Given a client with the light comment feed
+ * + raw-page-title reads, it returns the tracker's `probe` (count comments newer
+ * than the watermark, label a hit with the page title). Loosely typed at this
+ * cross-package boundary, like the rest of this loader.
+ */
+export type CreateListCommentsProbeFn = (client: {
+  listComments(
+    pageId: string,
+    includeResolved: boolean,
+  ): Promise<{ items: Array<{ createdAt?: string | null }> }>;
+  getPageRaw(
+    pageId: string,
+  ): Promise<{ title?: string | null } | null | undefined>;
+}) => (
+  pageId: string,
+  sinceMs: number,
+) => Promise<CommentSignalProbeResultLike>;
+
 // Pure, no-network draw.io helpers (#424). These are plain functions on the
 // module (NOT DocmostClient methods) — the in-app AI-SDK service calls them
 // directly to wire drawioShapes / drawioGuide, mirroring the MCP server.
@@ -170,6 +191,10 @@ interface DocmostMcpModule {
   // loader in unit tests. The in-app layer treats an absent factory as "signal
   // disabled" — a pure no-op that leaves tool results byte-identical.
   createCommentSignalTracker?: CommentSignalTrackerFactory;
+  // Optional (#494): the shared count-source probe factory. Absent on a pre-#494
+  // build or a mocked loader; the in-app layer only builds a probe when the
+  // signal factory above is also present.
+  createListCommentsProbe?: CreateListCommentsProbeFn;
   // Optional (#447): a deterministic hash of the tool-specs registry content,
   // generated into build/ by the package's build. Absent on a pre-#447 build (or
   // the mocked loader in unit tests) — the stale-check below is a NO-OP when it
@@ -284,6 +309,7 @@ export async function loadDocmostMcp(): Promise<{
   DocmostClient: DocmostClientCtor;
   sharedToolSpecs: Record<string, SharedToolSpec>;
   createCommentSignalTracker?: CommentSignalTrackerFactory;
+  createListCommentsProbe?: CreateListCommentsProbeFn;
   searchShapes: SearchShapesFn;
   getGuideSection: GetGuideSectionFn;
 }> {
@@ -331,6 +357,9 @@ export async function loadDocmostMcp(): Promise<{
     // Optional: forwarded when present so the in-app layer can build the passive
     // comment signal (#417); undefined on a stale build => signal disabled.
     createCommentSignalTracker: mod.createCommentSignalTracker,
+    // Optional (#494): the shared count-source probe factory; undefined on a
+    // stale build => the in-app layer falls back to no signal.
+    createListCommentsProbe: mod.createListCommentsProbe,
     // Pure no-network draw.io helpers (#424); not client methods.
     searchShapes: mod.searchShapes,
     getGuideSection: mod.getGuideSection,

apps/server/src/core/comment/comment.controller.ts

@@ -16,6 +16,7 @@ import { UpdateCommentDto } from './dto/update-comment.dto';
 import { ResolveCommentDto } from './dto/resolve-comment.dto';
 import { ApplySuggestionDto } from './dto/apply-suggestion.dto';
 import { DismissSuggestionDto } from './dto/dismiss-suggestion.dto';
+import { ResyncSuggestionAnchorDto } from './dto/resync-suggestion-anchor.dto';
 import { PageIdDto, CommentIdDto } from './dto/comments.input';
 import { AuthUser } from '../../common/decorators/auth-user.decorator';
 import { AuthWorkspace } from '../../common/decorators/auth-workspace.decorator';
@@ -235,6 +236,39 @@ export class CommentController {
     return this.commentService.applySuggestion(comment, user, provenance);
   }
 
+  @HttpCode(HttpStatus.OK)
+  @Post('resync-suggestion-anchor')
+  async resyncSuggestionAnchor(
+    @Body() dto: ResyncSuggestionAnchorDto,
+    @AuthUser() user: User,
+    @AuthWorkspace() workspace: Workspace,
+  ) {
+    const comment = await this.commentRepo.findById(dto.commentId, {
+      includeCreator: true,
+      includeResolvedBy: true,
+    });
+    if (!comment) {
+      throw new NotFoundException('Comment not found');
+    }
+
+    const page = await this.pageRepo.findById(comment.pageId);
+    if (!page || page.deletedAt) {
+      throw new NotFoundException('Page not found');
+    }
+
+    // Authorize BEFORE revealing structural detail (mirrors apply/dismiss).
+    // Re-anchoring does NOT change the page text — it only corrects the stored
+    // selection metadata — so the page-level gate is comment access. The service
+    // further restricts it to the suggestion's own author.
+    await this.pageAccessService.validateCanComment(page, user, workspace.id);
+
+    return this.commentService.resyncSuggestionAnchor(
+      comment,
+      dto.selection,
+      user,
+    );
+  }
+
   @HttpCode(HttpStatus.OK)
   @Post('dismiss-suggestion')
   async dismissSuggestion(

apps/server/src/core/comment/comment.service.apply-suggestion.spec.ts

@@ -146,11 +146,19 @@ describe('CommentService — applySuggestion', () => {
       'page-1',
       expect.objectContaining({ operation: 'commentDeleted', commentId: 'c-1' }),
     );
+    // #496: hard-deleted row → the audit payload is the only surviving record.
     expect(auditService.log).toHaveBeenCalledWith(
       expect.objectContaining({
         event: AuditEvent.COMMENT_SUGGESTION_APPLIED,
         resourceType: AuditResource.COMMENT,
         resourceId: 'c-1',
+        metadata: expect.objectContaining({
+          pageId: 'page-1',
+          suggestedText: 'new text',
+          selection: 'old text',
+          commentAuthor: 'user-1',
+          decidedBy: 'user-1',
+        }),
       }),
     );
     expect(result.outcome).toBe('deleted');
@@ -189,17 +197,25 @@ describe('CommentService — applySuggestion', () => {
     expect(resolvePatch.resolvedAt).toBeInstanceOf(Date);
     expect(resolvePatch.resolvedById).toBe('user-1');
 
-    // NOT deleted; broadcast an update, not a deletion.
+    // NOT deleted.
     expect(commentRepo.deleteComment).not.toHaveBeenCalled();
     expect(collaborationGateway.handleYjsEvent).not.toHaveBeenCalledWith(
       'deleteCommentMark',
       expect.anything(),
       expect.anything(),
     );
+    // #496 dedup: resolveComment broadcasts `commentResolved` with the enriched
+    // row; finalize must NOT ALSO emit a redundant `commentUpdated`. So the
+    // thread receives exactly ONE resolve broadcast and no update broadcast.
     expect(wsService.emitCommentEvent).toHaveBeenCalledWith(
       'space-1',
       'page-1',
-      expect.objectContaining({ operation: 'commentUpdated', comment: UPDATED }),
+      expect.objectContaining({ operation: 'commentResolved', comment: UPDATED }),
+    );
+    expect(wsService.emitCommentEvent).not.toHaveBeenCalledWith(
+      'space-1',
+      'page-1',
+      expect.objectContaining({ operation: 'commentUpdated' }),
     );
 
     expect(auditService.log).toHaveBeenCalledWith(
@@ -211,6 +227,36 @@ describe('CommentService — applySuggestion', () => {
     expect(result.outcome).toBe('resolved');
   });
 
+  it('re-entry: already applied+resolved WITH replies → emits commentUpdated (dedup does not over-suppress)', async () => {
+    // suggestionAppliedAt set → idempotent finalize; resolvedAt set → resolveComment
+    // is skipped, so there is NO commentResolved broadcast. The applied-stamp state
+    // must still reach clients via a single commentUpdated.
+    const { service, wsService } = makeService(
+      { applied: false, currentText: 'new text' },
+      true,
+    );
+
+    await service.applySuggestion(
+      suggestionComment({
+        suggestionAppliedAt: new Date(),
+        resolvedAt: new Date(),
+      }),
+      user(),
+    );
+
+    expect(wsService.emitCommentEvent).toHaveBeenCalledWith(
+      'space-1',
+      'page-1',
+      expect.objectContaining({ operation: 'commentUpdated', comment: UPDATED }),
+    );
+    // Nothing resolved this time (already resolved) → no resolve broadcast.
+    expect(wsService.emitCommentEvent).not.toHaveBeenCalledWith(
+      'space-1',
+      'page-1',
+      expect.objectContaining({ operation: 'commentResolved' }),
+    );
+  });
+
   // --- error / rejection branches -----------------------------------------
 
   it('applied=false and currentText differs → ConflictException with currentText in payload', async () => {

apps/server/src/core/comment/comment.service.dismiss-suggestion.spec.ts

@@ -107,11 +107,21 @@ describe('CommentService — dismissSuggestion', () => {
       'page-1',
       expect.objectContaining({ operation: 'commentDeleted', commentId: 'c-1' }),
     );
+    // #496: the row is hard-deleted, so the audit payload must carry the
+    // decision's substance (what was suggested, the anchored text, who authored
+    // it, who decided) — it is the only surviving record.
     expect(auditService.log).toHaveBeenCalledWith(
       expect.objectContaining({
         event: AuditEvent.COMMENT_SUGGESTION_DISMISSED,
         resourceType: AuditResource.COMMENT,
         resourceId: 'c-1',
+        metadata: expect.objectContaining({
+          pageId: 'page-1',
+          suggestedText: 'new text',
+          selection: 'old text',
+          commentAuthor: 'user-1',
+          decidedBy: 'user-1',
+        }),
       }),
     );
     expect(result.outcome).toBe('deleted');

apps/server/src/core/comment/comment.service.resync-anchor.spec.ts

@@ -0,0 +1,126 @@
+import { BadRequestException, ForbiddenException } from '@nestjs/common';
+import { CommentService } from './comment.service';
+
+/**
+ * Coverage for CommentService.resyncSuggestionAnchor (#496): re-anchoring a
+ * suggestion's stored selection (== apply-time expectedText) to the live-doc
+ * substring. The service is built directly with jest-mocked deps (the
+ * @InjectQueue tokens can't be resolved by Test.createTestingModule — see the
+ * sibling specs).
+ */
+describe('CommentService — resyncSuggestionAnchor', () => {
+  const UPDATED = { id: 'c-1', selection: 'new anchor', __updated: true } as any;
+
+  function makeService() {
+    const commentRepo: any = {
+      updateComment: jest.fn(async () => undefined),
+      findById: jest.fn(async () => UPDATED),
+    };
+    const service = new CommentService(
+      commentRepo,
+      {} as any,
+      { emitCommentEvent: jest.fn() } as any,
+      {} as any,
+      { add: jest.fn() } as any,
+      { add: jest.fn() } as any,
+      { log: jest.fn() } as any,
+    );
+    return { service, commentRepo };
+  }
+
+  const suggestion = (over?: Partial<any>): any => ({
+    id: 'c-1',
+    creatorId: 'user-1',
+    parentCommentId: null,
+    selection: 'old anchor',
+    suggestedText: 'new text',
+    suggestionAppliedAt: null,
+    resolvedAt: null,
+    ...over,
+  });
+  const user = (over?: Partial<any>): any => ({ id: 'user-1', ...over });
+
+  it('persists the new selection and returns the enriched comment', async () => {
+    const { service, commentRepo } = makeService();
+
+    const out = await service.resyncSuggestionAnchor(
+      suggestion(),
+      'new anchor',
+      user(),
+    );
+
+    expect(commentRepo.updateComment).toHaveBeenCalledWith(
+      { selection: 'new anchor' },
+      'c-1',
+    );
+    expect(out).toBe(UPDATED);
+  });
+
+  it('is idempotent: no write when the anchor already matches', async () => {
+    const { service, commentRepo } = makeService();
+
+    const out = await service.resyncSuggestionAnchor(
+      suggestion({ selection: 'same' }),
+      'same',
+      user(),
+    );
+
+    expect(commentRepo.updateComment).not.toHaveBeenCalled();
+    expect(out).toEqual(suggestion({ selection: 'same' }));
+  });
+
+  it('rejects a non-author (only the suggestion owner may re-anchor)', async () => {
+    const { service, commentRepo } = makeService();
+    await expect(
+      service.resyncSuggestionAnchor(suggestion(), 'new anchor', user({ id: 'other' })),
+    ).rejects.toBeInstanceOf(ForbiddenException);
+    expect(commentRepo.updateComment).not.toHaveBeenCalled();
+  });
+
+  it('rejects a reply / a comment with no suggestion', async () => {
+    const { service } = makeService();
+    await expect(
+      service.resyncSuggestionAnchor(
+        suggestion({ parentCommentId: 'p-1' }),
+        'x',
+        user(),
+      ),
+    ).rejects.toBeInstanceOf(BadRequestException);
+    await expect(
+      service.resyncSuggestionAnchor(
+        suggestion({ suggestedText: null }),
+        'x',
+        user(),
+      ),
+    ).rejects.toBeInstanceOf(BadRequestException);
+  });
+
+  it('rejects re-anchoring an already applied or resolved suggestion', async () => {
+    const { service } = makeService();
+    await expect(
+      service.resyncSuggestionAnchor(
+        suggestion({ suggestionAppliedAt: new Date() }),
+        'x',
+        user(),
+      ),
+    ).rejects.toBeInstanceOf(BadRequestException);
+    await expect(
+      service.resyncSuggestionAnchor(
+        suggestion({ resolvedAt: new Date() }),
+        'x',
+        user(),
+      ),
+    ).rejects.toBeInstanceOf(BadRequestException);
+  });
+
+  it('rejects a no-op selection equal to the suggested text', async () => {
+    const { service } = makeService();
+    await expect(
+      service.resyncSuggestionAnchor(
+        suggestion({ suggestedText: 'new text' }),
+        'new text',
+        user(),
+      ),
+    ).rejects.toBeInstanceOf(BadRequestException);
+  });
+});

apps/server/src/core/comment/comment.service.ts

@@ -370,6 +370,76 @@ export class CommentService {
     return updatedComment;
   }
 
+  /**
+   * Re-sync a suggestion's stored `selection` (== apply-time expectedText) to the
+   * RAW substring the inline mark actually covers in the LIVE document (#496).
+   *
+   * The MCP client creates the comment from a DEBOUNCED REST snapshot, then
+   * anchors the mark in the live collab doc. When the two disagree (the doc moved
+   * on in the debounce window) the stored selection no longer equals the marked
+   * text, so EVERY apply 409s ("the commented text changed"). After anchoring the
+   * client re-reads the exact marked substring and calls this to store it, making
+   * apply's strict equality hold.
+   *
+   * Only meaningful for an un-settled top-level suggestion authored by the
+   * caller: applying/resolving freezes the anchor, and a reply-carrying thread is
+   * preserved rather than mutated. The new text must still differ from the
+   * suggestion (else "apply" would be a no-op), preserving create()'s invariant.
+   */
+  async resyncSuggestionAnchor(
+    comment: Comment,
+    selection: string,
+    user: User,
+  ): Promise<Comment> {
+    if (comment.creatorId !== user.id) {
+      throw new ForbiddenException(
+        'You can only re-anchor your own suggestion',
+      );
+    }
+    if (comment.parentCommentId) {
+      throw new BadRequestException(
+        'Only a top-level comment can carry a suggested edit',
+      );
+    }
+    if (!comment.suggestedText) {
+      throw new BadRequestException('This comment has no suggested edit');
+    }
+    // A settled suggestion's anchor is frozen: re-anchoring an applied/resolved
+    // thread is meaningless and could resurrect a stale expectedText.
+    if (comment.suggestionAppliedAt || comment.resolvedAt) {
+      throw new BadRequestException(
+        'Cannot re-anchor a suggestion that was already applied or resolved',
+      );
+    }
+    const trimmed = selection.trim();
+    if (trimmed.length === 0) {
+      throw new BadRequestException('The re-anchored selection cannot be empty');
+    }
+    // Same no-op guard as create(): the suggestion must differ from the text it
+    // replaces, or apply becomes indistinguishable from already-applied.
+    if (trimmed === comment.suggestedText.trim()) {
+      throw new BadRequestException(
+        'A suggested edit must differ from the selected text',
+      );
+    }
+
+    // Idempotent: nothing to persist when the anchor already matches.
+    if (comment.selection === selection) {
+      return comment;
+    }
+
+    await this.commentRepo.updateComment({ selection }, comment.id);
+
+    const updatedComment = await this.commentRepo.findById(comment.id, {
+      includeCreator: true,
+      includeResolvedBy: true,
+    });
+
+    // Re-anchoring only corrects stored metadata; it does not change the page
+    // text or the comment body, so no ws broadcast / notification is warranted.
+    return updatedComment;
+  }
+
   /**
    * Apply the suggested edit carried by a top-level inline comment: atomically
    * replace the text under the comment mark in the collaborative document with
@@ -524,7 +594,7 @@ export class CommentService {
         resourceType: AuditResource.COMMENT,
         resourceId: comment.id,
         spaceId: comment.spaceId,
-        metadata: { pageId: comment.pageId },
+        metadata: this.suggestionAuditMetadata(comment, user),
       });
       return { ...updatedComment, outcome: 'resolved' };
     }
@@ -538,7 +608,7 @@ export class CommentService {
       resourceType: AuditResource.COMMENT,
       resourceId: comment.id,
       spaceId: comment.spaceId,
-      metadata: { pageId: comment.pageId },
+      metadata: this.suggestionAuditMetadata(comment, user),
     });
     return settled;
   }
@@ -577,8 +647,10 @@ export class CommentService {
 
       // Auto-resolve the thread. resolveComment handles the resolve mark, its ws
       // broadcast and the resolve notification. Stay defensive on re-entry.
+      let didResolveBroadcast = false;
       if (!comment.resolvedAt) {
         await this.resolveComment(comment, true, user, provenance);
+        didResolveBroadcast = true;
       }
 
       const updatedComment = await this.commentRepo.findById(comment.id, {
@@ -586,18 +658,27 @@ export class CommentService {
         includeResolvedBy: true,
       });
 
-      this.wsService.emitCommentEvent(comment.spaceId, comment.pageId, {
-        operation: 'commentUpdated',
-        pageId: comment.pageId,
-        comment: updatedComment,
-      });
+      // #496 dedup: resolveComment already broadcast `commentResolved` carrying
+      // the fully-enriched row (the applied stamps were persisted above, before
+      // that call, so its re-read reflects them). Emitting `commentUpdated` here
+      // too made the client receive TWO events for one apply. Broadcast the
+      // update ONLY when we did NOT resolve — i.e. the rare re-entry on an
+      // already-resolved thread, where the applied-stamp change still needs a
+      // broadcast and resolveComment did not run.
+      if (!didResolveBroadcast) {
+        this.wsService.emitCommentEvent(comment.spaceId, comment.pageId, {
+          operation: 'commentUpdated',
+          pageId: comment.pageId,
+          comment: updatedComment,
+        });
+      }
 
       this.auditService.log({
         event: AuditEvent.COMMENT_SUGGESTION_APPLIED,
         resourceType: AuditResource.COMMENT,
         resourceId: comment.id,
         spaceId: comment.spaceId,
-        metadata: { pageId: comment.pageId },
+        metadata: this.suggestionAuditMetadata(comment, user),
       });
 
       return { ...updatedComment, outcome: 'resolved' };
@@ -616,7 +697,7 @@ export class CommentService {
       resourceType: AuditResource.COMMENT,
       resourceId: comment.id,
       spaceId: comment.spaceId,
-      metadata: { pageId: comment.pageId },
+      metadata: this.suggestionAuditMetadata(comment, user),
     });
 
     return settled;
@@ -627,14 +708,17 @@ export class CommentService {
    * inline `comment` anchor mark, then ATOMICALLY hard-delete the row only if it
    * is still childless. Shared by the apply/dismiss no-replies branches (#329).
    *
-   * ORDER MATTERS: the anchor mark is removed FIRST and FATALLY (mirrors
-   * applySuggestion, which mutates the doc before writing the DB). The row
-   * delete is irreversible, so if the mark removal fails — including the
-   * COLLAB_DISABLE_REDIS "no live instance" hard-error — we must NOT delete the
-   * row and report success, or the document is left with a permanent orphan
-   * anchor pointing at a comment that no longer exists (the exact data-integrity
-   * bug #329 targets). Let the exception propagate (→ 5xx); the operation is
-   * then repeatable with row + mark still consistent.
+   * ORDER MATTERS (updated #399 → #496): what runs FIRST and FATALLY here is the
+   * mark-removal ENQUEUE (a fast, durable Redis add), NOT the mark op itself.
+   * deleteCommentMark awaits only the enqueue, so a failed add throws BEFORE the
+   * irreversible row delete — the row + mark stay consistent and the operation is
+   * repeatable. The actual anchor strip then runs off the HTTP path in the worker
+   * (idempotent, 3 retries). Only an EXHAUSTED-retries job could leave the doc
+   * with an orphan anchor pointing at a hard-deleted comment (the data-integrity
+   * bug #329 targets); that residual divergence is now self-healed by the
+   * resolve/unresolve mark worker, which strips an orphan mark whenever its
+   * comment row is gone (#496), and it is meanwhile VISIBLE via BullMQ failed-job
+   * metrics rather than a silently-swallowed warn.
    *
    * RACE (#338 F4): the caller read `hasChildren` BEFORE the (slow) mark
    * removal, so a reply can land in that window. `comments.parent_comment_id` is
@@ -732,6 +816,27 @@ export class CommentService {
     return this.generalQueue.add(QueueJob.COMMENT_MARK_UPDATE, jobData);
   }
 
+  /**
+   * Build the audit metadata for a suggestion apply/dismiss decision (#496).
+   * The subject comment is HARD-DELETED on the childless path, so the audit row
+   * is the only surviving record — capture the decision's substance (what was
+   * suggested, the anchored text it replaced, who authored it, who decided)
+   * before the row can vanish. `decidedBy` is the acting user; `commentAuthor`
+   * is the suggestion's creator.
+   */
+  private suggestionAuditMetadata(
+    comment: Comment,
+    user: User,
+  ): Record<string, any> {
+    return {
+      pageId: comment.pageId,
+      suggestedText: comment.suggestedText ?? null,
+      selection: comment.selection ?? null,
+      commentAuthor: comment.creatorId ?? null,
+      decidedBy: user.id,
+    };
+  }
+
   private async queueCommentNotification(
     content: any,
     oldMentionIds: string[],

apps/server/src/core/comment/dto/resync-suggestion-anchor.dto.ts

@@ -0,0 +1,20 @@
+import { IsString, IsUUID, MaxLength, MinLength } from 'class-validator';
+
+/**
+ * #496: after the MCP client anchors a suggestion in the LIVE collab doc, it
+ * re-reads the exact substring under the new mark and syncs it here as the
+ * comment's stored `selection` (== apply-time expectedText). Fixes the perpetual
+ * 409 where expectedText came from a debounced REST snapshot while the mark sat
+ * in the live doc.
+ */
+export class ResyncSuggestionAnchorDto {
+  @IsUUID()
+  commentId: string;
+
+  // The raw substring the mark now covers in the live document. Bounded like the
+  // create-time selection (2000) so a legitimate anchored span is never cut.
+  @IsString()
+  @MinLength(1)
+  @MaxLength(2000)
+  selection: string;
+}

apps/server/src/core/page/dto/move-page.dto.spec.ts

@@ -17,9 +17,10 @@ import { MovePageDto } from './move-page.dto';
 //     a valid ordering key the server itself generated would be refused on move.
 //
 // The tests below assert the CORRECT contract: any key the generator can produce
-// must satisfy the DTO. The genuinely-failing case is marked `test.failing` so the
-// suite stays green while locking the bug; it flips red (alerting us) once the DTO
-// bounds are widened to cover the generator's real range.
+// must satisfy the DTO. FIXED (#495 item 9): the DTO now validates `position` by
+// CHARSET ([0-9A-Za-z], the generator's base-62 alphabet) instead of the wrong
+// @MaxLength(12) length bound, so dense between-inserts are accepted; the former
+// `test.failing` bug-lock is now a passing assertion.
 
 function constraintErrors(position: unknown) {
   const dto = plainToInstance(MovePageDto, {
@@ -47,24 +48,33 @@ describe('MovePageDto.position vs generateJitteredKeyBetween parity', () => {
     expect(hasError(errors, 'position')).toBe(false);
   });
 
-  // BUG LOCK: dense between-inserts produce keys longer than 12 chars, which
-  // MaxLength(12) rejects even though they are valid ordering keys. This SHOULD
-  // pass; it currently fails. Flips green when the DTO bound is fixed.
-  test.failing(
-    'accepts dense between-inserted keys (currently rejected by MaxLength(12))',
-    async () => {
-      let lo = generateJitteredKeyBetween(null, null);
-      let hi = generateJitteredKeyBetween(lo, null);
-      // Repeatedly insert just above `lo`, shrinking the gap so the key grows.
-      let longest = lo;
-      for (let i = 0; i < 40; i++) {
-        const mid = generateJitteredKeyBetween(lo, hi);
-        if (mid.length > longest.length) longest = mid;
-        hi = mid;
-      }
-      expect(longest.length).toBeGreaterThan(12); // sanity: we produced a long key
-      const errors = await constraintErrors(longest);
-      expect(hasError(errors, 'position')).toBe(false);
-    },
-  );
+  // FIXED: dense between-inserts produce keys longer than 12 chars, which the old
+  // MaxLength(12) rejected even though they are valid ordering keys. Now accepted.
+  it('accepts dense between-inserted keys longer than 12 chars', async () => {
+    let lo = generateJitteredKeyBetween(null, null);
+    let hi = generateJitteredKeyBetween(lo, null);
+    // Repeatedly insert just above `lo`, shrinking the gap so the key grows. The
+    // generator is JITTERED (random), so use enough iterations that the longest
+    // key reliably clears the old 12-char bound: measured min-over-50-trials is
+    // ~11 at 40 iterations (flaky) but ~36 at 200 (robust margin).
+    let longest = lo;
+    for (let i = 0; i < 200; i++) {
+      const mid = generateJitteredKeyBetween(lo, hi);
+      if (mid.length > longest.length) longest = mid;
+      hi = mid;
+    }
+    expect(longest.length).toBeGreaterThan(12); // sanity: we produced a long key
+    const errors = await constraintErrors(longest);
+    expect(hasError(errors, 'position')).toBe(false);
+  });
+
+  // The charset guard replaces the length bound: reject anything outside the
+  // generator's [0-9A-Za-z] alphabet (control chars, separators, injection) and
+  // the empty string, while still accepting every real key.
+  it('rejects a position with characters outside the fractional-index alphabet', async () => {
+    for (const bad of ['a0/b', 'a b', 'a\n0', 'a.b', '', "a';--"]) {
+      const errors = await constraintErrors(bad);
+      expect(hasError(errors, 'position')).toBe(true);
+    }
+  });
 });

apps/server/src/core/page/dto/move-page.dto.ts

@@ -1,8 +1,8 @@
 import {
   IsString,
   IsOptional,
-  MinLength,
   MaxLength,
+  Matches,
   IsNotEmpty,
 } from 'class-validator';
 
@@ -10,9 +10,19 @@ export class MovePageDto {
   @IsString()
   pageId: string;
 
+  // `position` is a fractional-indexing key from `generateJitteredKeyBetween`
+  // (the SAME generator page.service uses). Validate by CHARSET, not length: the
+  // generator's default base-62 alphabet is [0-9A-Za-z], and DENSE between-inserts
+  // legitimately grow a key well past a dozen chars (measured >40), so the old
+  // @MinLength(5)/@MaxLength(12) bounds wrongly 400'd valid ordering keys the
+  // server itself produced (Gitea #139 item 6). The charset regex rejects control
+  // chars / separators / injection, and a generous MaxLength stays only as a
+  // DoS guard — far above any realistic key, so it never rejects a real move.
   @IsString()
-  @MinLength(5)
-  @MaxLength(12)
+  @Matches(/^[0-9A-Za-z]+$/, {
+    message: 'position must be a fractional-index key ([0-9A-Za-z])',
+  })
+  @MaxLength(256)
   position: string;
 
   @IsOptional()

apps/server/src/core/page/dto/page-identity.dto.spec.ts

@@ -0,0 +1,49 @@
+import 'reflect-metadata';
+import { plainToInstance } from 'class-transformer';
+import { validate } from 'class-validator';
+import { PageIdDto } from './page.dto';
+
+// #435: PageIdDto.pageId carries a page's DOUBLE identity (internal UUID OR
+// public 10-char slugId), both as bare strings. The DTO must accept exactly
+// those two FORMATS and reject a malformed / swapped identity at the boundary.
+async function pageIdErrors(pageId: unknown) {
+  const dto = plainToInstance(PageIdDto, { pageId });
+  const errors = await validate(dto as object);
+  return errors.some((e) => e.property === 'pageId');
+}
+
+const UUID = '019f499a-9f8c-7d68-b7be-ce100d7c6c56';
+const SLUG = 'aB3xQ7kR2p';
+
+describe('PageIdDto pageId format validation', () => {
+  it('accepts a canonical page UUID', async () => {
+    expect(await pageIdErrors(UUID)).toBe(false);
+  });
+
+  it('accepts a 10-char slugId', async () => {
+    expect(await pageIdErrors(SLUG)).toBe(false);
+  });
+
+  it('rejects a truncated / wrong-length slug', async () => {
+    expect(await pageIdErrors('aB3xQ7kR2')).toBe(true); // 9 chars
+    expect(await pageIdErrors('aB3xQ7kR2pX')).toBe(true); // 11 chars
+  });
+
+  it('rejects a slug with an illegal character', async () => {
+    expect(await pageIdErrors('aB3xQ7kR2!')).toBe(true);
+  });
+
+  it('rejects a full URL / path-shaped identity (not the bare id)', async () => {
+    expect(await pageIdErrors(`my-page-title-${SLUG}`)).toBe(true);
+    expect(await pageIdErrors(`https://x/p/${SLUG}`)).toBe(true);
+  });
+
+  it('rejects a malformed UUID', async () => {
+    expect(await pageIdErrors('019f499a-9f8c-7d68-b7be')).toBe(true);
+    expect(await pageIdErrors('not-a-uuid-at-all-really')).toBe(true);
+  });
+
+  it('rejects an empty string', async () => {
+    expect(await pageIdErrors('')).toBe(true);
+  });
+});

apps/server/src/core/page/dto/page-identity.validator.ts

@@ -0,0 +1,39 @@
+import { applyDecorators } from '@nestjs/common';
+import { Matches, ValidationOptions } from 'class-validator';
+
+/**
+ * A page identity at the API boundary is EITHER the internal page UUID or the
+ * public 10-char slugId (page.repo.findById matches a non-UUID input as a
+ * slugId). Both arrive as bare strings, which is exactly how the two got swapped
+ * silently (incident family #435). This regex pins the two accepted FORMATS so a
+ * malformed / cross-wired identity (a truncated slug, a full URL, an email, an
+ * id from another entity kind that isn't even shaped like either) is rejected at
+ * the boundary instead of falling through to a confusing 404.
+ *
+ *   - UUID:    canonical 8-4-4-4-12 hex (version-agnostic — page ids are UUIDv7,
+ *              so only the shape/length is enforced, matching the MCP's UUID_RE
+ *              and the server's isValidUUID acceptance).
+ *   - slugId:  exactly 10 chars over [0-9A-Za-z] (nanoid `generateSlugId`).
+ *
+ * The two are disjoint (a UUID is 36 chars WITH dashes, a slugId 10 chars
+ * WITHOUT), so a value can only satisfy one branch.
+ */
+export const PAGE_ID_OR_SLUG_ID_REGEX =
+  /^(?:[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}|[0-9A-Za-z]{10})$/i;
+
+/**
+ * Validate that a DTO string field is a well-formed page identity (page UUID OR
+ * 10-char slugId). Composed decorator so the same format rule is applied
+ * consistently wherever a DTO accepts a `pageId` that may be either form.
+ */
+export function IsPageIdOrSlugId(
+  validationOptions?: ValidationOptions,
+): PropertyDecorator {
+  return applyDecorators(
+    Matches(PAGE_ID_OR_SLUG_ID_REGEX, {
+      message:
+        'pageId must be a page UUID or a 10-character slugId',
+      ...validationOptions,
+    }),
+  );
+}

apps/server/src/core/page/dto/page.dto.ts

@@ -9,10 +9,16 @@ import {
 import { Transform } from 'class-transformer';
 
 import { ContentFormat } from './create-page.dto';
+import { IsPageIdOrSlugId } from './page-identity.validator';
 
 export class PageIdDto {
   @IsString()
   @IsNotEmpty()
+  // Format-validate the double identity (#435): accept only a page UUID or a
+  // 10-char slugId so a malformed / swapped identity is rejected at the boundary
+  // rather than passed to the repo as a bare string. Base for PageInfoDto,
+  // DeletePageDto, BacklinksListDto, AddLabelsDto/RemoveLabelDto, etc.
+  @IsPageIdOrSlugId()
   pageId: string;
 }
 

apps/server/src/core/page/services/page.service.ts

@@ -53,8 +53,10 @@ import {
   extractPageSlugId,
 } from '../../../integrations/export/utils';
 import { canonicalizeFootnotes } from '@docmost/editor-ext';
-import { markdownToProseMirror } from '@docmost/prosemirror-markdown';
-import { normalizeForeignMarkdown } from '../../../integrations/import/utils/foreign-markdown';
+import {
+  markdownToProseMirror,
+  normalizeForeignMarkdown,
+} from '@docmost/prosemirror-markdown';
 import { WatcherService } from '../../watcher/watcher.service';
 import { sql } from 'kysely';
 import { TransclusionService } from '../transclusion/transclusion.service';

apps/server/src/core/page/services/spec/temporary-note-cleanup.service.spec.ts

@@ -1,19 +1,38 @@
 import { TemporaryNoteCleanupService } from '../temporary-note-cleanup.service';
 
 /**
- * Chainable Kysely stub that records every `.where(...)` call so the test can
- * assert the sweep only selects armed, expired, not-yet-trashed notes. The
- * terminal `.execute()` resolves the configured expired rows (the batch SELECT);
- * `.executeTakeFirst()` resolves the per-row deadline re-read done just before
- * each `removePage`. By default the re-read reports the note as still armed and
- * still expired (epoch deadline < now), so the sweep proceeds to delete it;
- * tests override `reReadFirst` to simulate a concurrent "Make permanent".
+ * Chainable Kysely stub for the temporary-note sweep.
+ *
+ * `this.db` serves the non-locking candidate SELECT (selectFrom/select/where/
+ * limit/execute -> the configured expired rows) AND `.transaction().execute(cb)`,
+ * which runs `cb` with a separate `trx` builder. The `trx` builder serves the
+ * per-row LOCKED re-check (selectFrom/select/where/forUpdate/skipLocked/
+ * executeTakeFirst). `lockedRows` drives what that locked re-check returns per
+ * candidate — an id/creator/workspace row means "still expired, delete it";
+ * `undefined` means the predicate no longer matched (made permanent / re-armed /
+ * already trashed) or the row was SKIP-LOCKED by another worker, so it is skipped.
  */
-function makeDbStub(expiredRows: any[]) {
+function makeDbStub(expiredRows: any[], lockedRows?: any[]) {
   const whereCalls: any[][] = [];
-  const reReadFirst = jest
-    .fn()
-    .mockResolvedValue({ temporaryExpiresAt: new Date(0), deletedAt: null });
+  const locked = [
+    ...(lockedRows ??
+      expiredRows.map((r) => ({
+        id: r.id,
+        creatorId: r.creatorId,
+        workspaceId: r.workspaceId,
+      }))),
+  ];
+  const lockedTakeFirst = jest.fn(() => Promise.resolve(locked.shift()));
+  const forUpdate = jest.fn(() => trxBuilder);
+  const skipLocked = jest.fn(() => trxBuilder);
+  const trxBuilder: any = {
+    selectFrom: jest.fn(() => trxBuilder),
+    select: jest.fn(() => trxBuilder),
+    where: jest.fn(() => trxBuilder),
+    forUpdate,
+    skipLocked,
+    executeTakeFirst: lockedTakeFirst,
+  };
   const builder: any = {
     selectFrom: jest.fn(() => builder),
     select: jest.fn(() => builder),
@@ -23,9 +42,11 @@ function makeDbStub(expiredRows: any[]) {
     }),
     limit: jest.fn(() => builder),
     execute: jest.fn().mockResolvedValue(expiredRows),
-    executeTakeFirst: reReadFirst,
+    transaction: jest.fn(() => ({
+      execute: (cb: (trx: any) => Promise<any>) => Promise.resolve(cb(trxBuilder)),
+    })),
   };
-  return { builder, whereCalls, reReadFirst };
+  return { builder, whereCalls, lockedTakeFirst, forUpdate, skipLocked };
 }
 
 describe('TemporaryNoteCleanupService.sweepExpiredTemporaryNotes', () => {
@@ -52,20 +73,36 @@ describe('TemporaryNoteCleanupService.sweepExpiredTemporaryNotes', () => {
     expect(builder.limit.mock.calls[0][0]).toBeGreaterThan(0);
   });
 
-  it('soft-deletes each expired note via removePage, attributed to its creator', async () => {
+  it('soft-deletes each expired note via removePage under a row lock, attributed to its creator', async () => {
     const expired = [
       { id: 'p1', creatorId: 'u1', workspaceId: 'w1' },
       { id: 'p2', creatorId: 'u2', workspaceId: 'w1' },
     ];
-    const { builder } = makeDbStub(expired);
+    const { builder, forUpdate, skipLocked } = makeDbStub(expired);
     const pageRepo = { removePage: jest.fn().mockResolvedValue(undefined) } as any;
     const service = new TemporaryNoteCleanupService(builder, pageRepo);
 
     await service.sweepExpiredTemporaryNotes();
 
     expect(pageRepo.removePage).toHaveBeenCalledTimes(2);
-    expect(pageRepo.removePage).toHaveBeenNthCalledWith(1, 'p1', 'u1', 'w1');
-    expect(pageRepo.removePage).toHaveBeenNthCalledWith(2, 'p2', 'u2', 'w1');
+    // The 4th arg is the locking transaction — the delete runs inside it.
+    expect(pageRepo.removePage).toHaveBeenNthCalledWith(
+      1,
+      'p1',
+      'u1',
+      'w1',
+      expect.anything(),
+    );
+    expect(pageRepo.removePage).toHaveBeenNthCalledWith(
+      2,
+      'p2',
+      'u2',
+      'w1',
+      expect.anything(),
+    );
+    // The re-check acquired a FOR UPDATE SKIP LOCKED lock (once per candidate).
+    expect(forUpdate).toHaveBeenCalledTimes(2);
+    expect(skipLocked).toHaveBeenCalledTimes(2);
   });
 
   it('continues past a failing note (one bad removePage does not abort the sweep)', async () => {
@@ -86,60 +123,29 @@ describe('TemporaryNoteCleanupService.sweepExpiredTemporaryNotes', () => {
       service.sweepExpiredTemporaryNotes(),
     ).resolves.toBeUndefined();
     expect(pageRepo.removePage).toHaveBeenCalledTimes(2);
-    expect(pageRepo.removePage).toHaveBeenNthCalledWith(2, 'good', 'u2', 'w1');
+    expect(pageRepo.removePage).toHaveBeenNthCalledWith(
+      2,
+      'good',
+      'u2',
+      'w1',
+      expect.anything(),
+    );
   });
 
-  it('does NOT trash a note made permanent in the race window', async () => {
-    // The batch SELECT saw the note as expired, but before its turn in the loop
-    // the user clicked "Make permanent" (temporary_expires_at -> null). The
-    // deadline re-read must catch this and skip the delete so the keep wins.
+  it('does NOT trash a note made permanent / re-armed / already trashed (locked re-check returns nothing)', async () => {
+    // The batch SELECT saw the note as expired, but by the time the LOCKED
+    // re-check runs the row no longer matches the still-armed+expired+not-trashed
+    // predicate (make-permanent, re-arm to a future deadline, or already trashed),
+    // OR another worker holds the row (SKIP LOCKED). In every case the locked
+    // SELECT returns nothing and the delete is skipped so the keep/other worker wins.
     const expired = [{ id: 'p1', creatorId: 'u1', workspaceId: 'w1' }];
-    const { builder, reReadFirst } = makeDbStub(expired);
-    reReadFirst.mockResolvedValueOnce({
-      temporaryExpiresAt: null,
-      deletedAt: null,
-    });
+    const { builder, lockedTakeFirst } = makeDbStub(expired, [undefined]);
     const pageRepo = { removePage: jest.fn() } as any;
     const service = new TemporaryNoteCleanupService(builder, pageRepo);
 
     await service.sweepExpiredTemporaryNotes();
 
-    expect(reReadFirst).toHaveBeenCalledTimes(1);
-    expect(pageRepo.removePage).not.toHaveBeenCalled();
-  });
-
-  it('skips a note already trashed since the batch SELECT', async () => {
-    const expired = [{ id: 'p1', creatorId: 'u1', workspaceId: 'w1' }];
-    const { builder, reReadFirst } = makeDbStub(expired);
-    reReadFirst.mockResolvedValueOnce({
-      temporaryExpiresAt: new Date(0),
-      deletedAt: new Date(),
-    });
-    const pageRepo = { removePage: jest.fn() } as any;
-    const service = new TemporaryNoteCleanupService(builder, pageRepo);
-
-    await service.sweepExpiredTemporaryNotes();
-
-    expect(pageRepo.removePage).not.toHaveBeenCalled();
-  });
-
-  it('does NOT trash a note re-armed to a future deadline in the race window', async () => {
-    // The batch SELECT saw the note as expired, but before its turn in the loop
-    // the user disarmed it and re-armed it to a fresh, still-future deadline
-    // (temporary_expires_at -> now + 1h). The deadline re-read must catch that
-    // the note is no longer expired and skip the delete so the keep wins.
-    const expired = [{ id: 'p1', creatorId: 'u1', workspaceId: 'w1' }];
-    const { builder, reReadFirst } = makeDbStub(expired);
-    reReadFirst.mockResolvedValueOnce({
-      temporaryExpiresAt: new Date(Date.now() + 60 * 60 * 1000),
-      deletedAt: null,
-    });
-    const pageRepo = { removePage: jest.fn() } as any;
-    const service = new TemporaryNoteCleanupService(builder, pageRepo);
-
-    await service.sweepExpiredTemporaryNotes();
-
-    expect(reReadFirst).toHaveBeenCalledTimes(1);
+    expect(lockedTakeFirst).toHaveBeenCalledTimes(1);
     expect(pageRepo.removePage).not.toHaveBeenCalled();
   });
 
@@ -151,4 +157,31 @@ describe('TemporaryNoteCleanupService.sweepExpiredTemporaryNotes', () => {
     await service.sweepExpiredTemporaryNotes();
     expect(pageRepo.removePage).not.toHaveBeenCalled();
   });
+
+  it('sweeps once on application bootstrap (catches notes expired during downtime)', async () => {
+    const expired = [{ id: 'p1', creatorId: 'u1', workspaceId: 'w1' }];
+    const { builder } = makeDbStub(expired);
+    const pageRepo = { removePage: jest.fn().mockResolvedValue(undefined) } as any;
+    const service = new TemporaryNoteCleanupService(builder, pageRepo);
+
+    await service.onApplicationBootstrap();
+
+    expect(pageRepo.removePage).toHaveBeenCalledTimes(1);
+    expect(pageRepo.removePage).toHaveBeenCalledWith(
+      'p1',
+      'u1',
+      'w1',
+      expect.anything(),
+    );
+  });
+
+  it('a startup-sweep failure never blocks application boot', async () => {
+    const { builder } = makeDbStub([]);
+    // Make the candidate SELECT throw to simulate a boot-time DB hiccup.
+    builder.execute.mockRejectedValueOnce(new Error('db not ready'));
+    const pageRepo = { removePage: jest.fn() } as any;
+    const service = new TemporaryNoteCleanupService(builder, pageRepo);
+
+    await expect(service.onApplicationBootstrap()).resolves.toBeUndefined();
+  });
 });

apps/server/src/core/page/services/temporary-note-cleanup.service.ts

@@ -1,8 +1,13 @@
-import { Injectable, Logger } from '@nestjs/common';
+import {
+  Injectable,
+  Logger,
+  OnApplicationBootstrap,
+} from '@nestjs/common';
 import { Interval } from '@nestjs/schedule';
 import { InjectKysely } from 'nestjs-kysely';
 import { KyselyDB } from '@docmost/db/types/kysely.types';
 import { PageRepo } from '@docmost/db/repos/page/page.repo';
+import { executeTx } from '@docmost/db/utils';
 
 /**
  * Background sweeper for temporary notes ("structure or die"). A note whose
@@ -11,7 +16,7 @@ import { PageRepo } from '@docmost/db/repos/page/page.repo';
  * TrashCleanupService; `@nestjs/schedule` is already enabled globally.
  */
 @Injectable()
-export class TemporaryNoteCleanupService {
+export class TemporaryNoteCleanupService implements OnApplicationBootstrap {
   private readonly logger = new Logger(TemporaryNoteCleanupService.name);
 
   // Cap a single sweep so a large backlog (e.g. many notes created during
@@ -24,6 +29,20 @@ export class TemporaryNoteCleanupService {
     private readonly pageRepo: PageRepo,
   ) {}
 
+  // Sweep once at startup so notes that expired during downtime are trashed
+  // right away instead of waiting up to an hour for the first @Interval tick.
+  // Best-effort: never let a startup-sweep failure block application boot.
+  async onApplicationBootstrap() {
+    try {
+      await this.sweepExpiredTemporaryNotes();
+    } catch (error) {
+      this.logger.error(
+        'Temporary-note startup sweep failed',
+        error instanceof Error ? error.stack : undefined,
+      );
+    }
+  }
+
   // Hourly granularity: lifetimes are configured in hours, so a sub-hour
   // overshoot past the deadline is acceptable.
   @Interval('temporary-note-cleanup', 60 * 60 * 1000)
@@ -31,9 +50,11 @@ export class TemporaryNoteCleanupService {
     try {
       const now = new Date();
 
+      // Candidate ids (non-locking). The authoritative re-check happens per row
+      // under a row lock below, so this cheap pass just bounds the batch.
       const expired = await this.db
         .selectFrom('pages')
-        .select(['id', 'creatorId', 'workspaceId'])
+        .select(['id'])
         .where('temporaryExpiresAt', 'is not', null)
         .where('temporaryExpiresAt', '<', now)
         .where('deletedAt', 'is', null) // not already in trash
@@ -41,50 +62,53 @@ export class TemporaryNoteCleanupService {
         .execute();
 
       let trashed = 0;
-      for (const page of expired) {
+      for (const candidate of expired) {
         try {
-          // Re-check the deadline at deletion time. The SELECT above is not
-          // transactional, so a user may click "Make permanent"
-          // (toggleTemporary sets temporary_expires_at = null) in the window
-          // between the SELECT and this per-row removePage. removePage deletes
-          // by id with only a `deletedAt IS NULL` filter and never re-reads the
-          // deadline, so without this guard a concurrently-kept note would
-          // still be trashed. Re-read the row and skip it unless it is still
-          // armed AND still expired, so a concurrent make-permanent wins.
-          const current = await this.db
-            .selectFrom('pages')
-            .select(['temporaryExpiresAt', 'deletedAt'])
-            .where('id', '=', page.id)
-            .executeTakeFirst();
+          const didTrash = await executeTx(this.db, async (trx) => {
+            // Re-check the row UNDER A LOCK inside the transaction. `FOR UPDATE
+            // SKIP LOCKED`:
+            //  - serialises against a concurrent "Make permanent"
+            //    (toggleTemporary UPDATE takes the same row lock): if it commits
+            //    first, the deadline predicate below no longer matches and we
+            //    skip; if we lock first, it waits until this delete commits.
+            //  - SKIP LOCKED lets a second worker/instance skip a row another
+            //    sweeper already claimed instead of blocking on it (no double
+            //    processing, no thundering herd).
+            // The predicate re-asserts still-armed AND still-expired AND
+            // not-already-trashed, so a make-permanent / prior sweep drops the row.
+            const locked = await trx
+              .selectFrom('pages')
+              .select(['id', 'creatorId', 'workspaceId'])
+              .where('id', '=', candidate.id)
+              .where('temporaryExpiresAt', 'is not', null)
+              .where('temporaryExpiresAt', '<', now)
+              .where('deletedAt', 'is', null)
+              .forUpdate()
+              .skipLocked()
+              .executeTakeFirst();
 
-          if (
-            !current ||
-            current.deletedAt !== null ||
-            current.temporaryExpiresAt === null ||
-            new Date(current.temporaryExpiresAt) >= now
-          ) {
-            // Made permanent, already trashed, or no longer expired since the
-            // SELECT — leave it alone.
-            continue;
-          }
+            if (!locked) return false;
 
-          // Reuse the exact soft-delete path: recursive over children, removes
-          // shares in a transaction, and emits PAGE_SOFT_DELETED (tree
-          // invalidation + watcher notifications). Attribute the automatic
-          // deletion to the note's creator (no schema change). Both the SELECT
-          // above and removePage filter `deletedAt IS NULL`, so a double sweep
-          // is idempotent.
-          await this.pageRepo.removePage(
-            page.id,
-            // creatorId is set on every created page; a temporary note always
-            // has one. Cast to satisfy the non-null deletedById parameter.
-            page.creatorId as string,
-            page.workspaceId,
-          );
-          trashed++;
+            // Reuse the exact soft-delete path (recursive children + share
+            // removal + PAGE_SOFT_DELETED broadcast), running IN this locked
+            // transaction so the delete is atomic with the re-check and cannot
+            // deadlock on a nested independent transaction. The broadcast is
+            // deferred by removePage to this transaction's commit. Attribute the
+            // automatic deletion to the note's creator (no schema change).
+            await this.pageRepo.removePage(
+              locked.id,
+              // creatorId is set on every created page; a temporary note always
+              // has one. Cast to satisfy the non-null deletedById parameter.
+              locked.creatorId as string,
+              locked.workspaceId,
+              trx,
+            );
+            return true;
+          });
+          if (didTrash) trashed++;
         } catch (error) {
           this.logger.error(
-            `Failed to trash expired temporary note ${page.id}`,
+            `Failed to trash expired temporary note ${candidate.id}`,
             error instanceof Error ? error.stack : undefined,
           );
         }

apps/server/src/core/share/share-alias.controller.spec.ts

@@ -133,6 +133,9 @@ describe('ShareAliasController authz gates', () => {
         creatorId: 'u-1',
         alias: 'promo',
         confirmReassign: true,
+        // The requesting user is forwarded so setAlias can gate the reassign
+        // 409 title disclosure on target-page view permission (#495).
+        user,
       });
       expect(result).toEqual({ id: 'alias-1' });
     });

apps/server/src/core/share/share-alias.controller.ts

@@ -79,6 +79,9 @@ export class ShareAliasController {
       creatorId: user.id,
       alias: dto.alias,
       confirmReassign: dto.confirmReassign,
+      // Gates whether the reassign 409 may reveal the current target's title
+      // (view-permission check on that page) — see setAlias (#495).
+      user,
     });
   }
 

apps/server/src/core/share/share-alias.service.spec.ts

@@ -1,4 +1,8 @@
-import { BadRequestException, ConflictException } from '@nestjs/common';
+import {
+  BadRequestException,
+  ConflictException,
+  ForbiddenException,
+} from '@nestjs/common';
 import { NoResultError } from 'kysely';
 import { ShareAliasService } from './share-alias.service';
 
@@ -7,6 +11,8 @@ import { ShareAliasService } from './share-alias.service';
  * 409 reassign guard, uniqueness-race handling, availability probe, and the
  * request-time readable-target resolution (which re-runs the share boundary).
  */
+const USER = { id: 'u-1' } as any;
+
 describe('ShareAliasService', () => {
   // Sentinel handed to repo calls so tests can assert they ran inside the tx.
   const trx = { __trx: true };
@@ -27,6 +33,10 @@ describe('ShareAliasService', () => {
       resolveReadableSharePage: jest.fn(),
       isSharingAllowed: jest.fn(),
     };
+    // Default: the requester CAN view the target page (validateCanView resolves),
+    // so the reassign 409 may disclose its title. Tests override to reject to
+    // assert the no-leak path.
+    const pageAccessService = { validateCanView: jest.fn().mockResolvedValue(undefined) };
     // Fake kysely db: only .transaction().execute(cb) is used by setAlias.
     const db = {
       transaction: jest.fn(() => ({
@@ -37,9 +47,10 @@ describe('ShareAliasService', () => {
       shareAliasRepo as any,
       pageRepo as any,
       shareService as any,
+      pageAccessService as any,
       db as any,
     );
-    return { service, shareAliasRepo, pageRepo, shareService, db };
+    return { service, shareAliasRepo, pageRepo, shareService, pageAccessService, db };
   }
 
   describe('setAlias', () => {
@@ -50,6 +61,7 @@ describe('ShareAliasService', () => {
           workspaceId: 'ws-1',
           pageId: 'p-1',
           creatorId: 'u-1',
+          user: USER,
           alias: 'A', // too short + uppercase
         }),
       ).rejects.toBeInstanceOf(BadRequestException);
@@ -66,6 +78,7 @@ describe('ShareAliasService', () => {
         workspaceId: 'ws-1',
         pageId: 'p-1',
         creatorId: 'u-1',
+        user: USER,
         alias: '  My Page ',
       });
 
@@ -114,6 +127,7 @@ describe('ShareAliasService', () => {
         workspaceId: 'ws-1',
         pageId: 'p-1',
         creatorId: 'u-1',
+        user: USER,
         alias: 'ted',
       });
 
@@ -144,6 +158,7 @@ describe('ShareAliasService', () => {
         workspaceId: 'ws-1',
         pageId: 'p-1',
         creatorId: 'u-1',
+        user: USER,
         alias: 'foo',
       });
 
@@ -179,6 +194,7 @@ describe('ShareAliasService', () => {
         workspaceId: 'ws-1',
         pageId: 'p-1',
         creatorId: 'u-1',
+        user: USER,
         alias: 'new',
       });
 
@@ -190,30 +206,77 @@ describe('ShareAliasService', () => {
       );
     });
 
-    it('throws 409 with current target when name is taken and not confirmed', async () => {
-      const { service, shareAliasRepo, pageRepo } = makeService();
+    it('throws 409 with the target TITLE (never its id) when the requester CAN view it', async () => {
+      const { service, shareAliasRepo, pageRepo, pageAccessService } =
+        makeService();
       shareAliasRepo.findByAliasAndWorkspace.mockResolvedValue({
         id: 'a-1',
         alias: 'foo',
         pageId: 'p-other',
       });
       pageRepo.findById.mockResolvedValue({ id: 'p-other', title: 'Other' });
+      pageAccessService.validateCanView.mockResolvedValue(undefined); // can view
 
       try {
         await service.setAlias({
           workspaceId: 'ws-1',
           pageId: 'p-1',
           creatorId: 'u-1',
+          user: USER,
           alias: 'foo',
         });
         fail('expected ConflictException');
       } catch (err) {
         expect(err).toBeInstanceOf(ConflictException);
-        expect((err as ConflictException).getResponse()).toMatchObject({
+        const body = (err as ConflictException).getResponse();
+        expect(body).toMatchObject({
           code: 'ALIAS_REASSIGN_REQUIRED',
-          currentPageId: 'p-other',
           currentPageTitle: 'Other',
         });
+        // SECURITY (#495): the page id is NEVER disclosed, even to a viewer.
+        expect(body).not.toHaveProperty('currentPageId');
+        expect(pageAccessService.validateCanView).toHaveBeenCalledWith(
+          expect.objectContaining({ id: 'p-other' }),
+          USER,
+        );
+      }
+      expect(shareAliasRepo.updatePageId).not.toHaveBeenCalled();
+    });
+
+    it('throws 409 WITHOUT the title or id when the requester CANNOT view the target (#495)', async () => {
+      const { service, shareAliasRepo, pageRepo, pageAccessService } =
+        makeService();
+      shareAliasRepo.findByAliasAndWorkspace.mockResolvedValue({
+        id: 'a-1',
+        alias: 'foo',
+        pageId: 'p-secret',
+      });
+      pageRepo.findById.mockResolvedValue({ id: 'p-secret', title: 'Secret' });
+      // No view permission on the target page -> validateCanView throws.
+      pageAccessService.validateCanView.mockRejectedValue(
+        new ForbiddenException(),
+      );
+
+      try {
+        await service.setAlias({
+          workspaceId: 'ws-1',
+          pageId: 'p-1',
+          creatorId: 'u-1',
+          user: USER,
+          alias: 'foo',
+        });
+        fail('expected ConflictException');
+      } catch (err) {
+        expect(err).toBeInstanceOf(ConflictException);
+        const body = (err as ConflictException).getResponse() as Record<
+          string,
+          unknown
+        >;
+        expect(body).toMatchObject({ code: 'ALIAS_REASSIGN_REQUIRED' });
+        // The enumeration hole: neither the id nor the title of a page the
+        // requester cannot see may leak.
+        expect(body).not.toHaveProperty('currentPageId');
+        expect(body.currentPageTitle ?? null).toBeNull();
       }
       expect(shareAliasRepo.updatePageId).not.toHaveBeenCalled();
     });
@@ -231,6 +294,7 @@ describe('ShareAliasService', () => {
         workspaceId: 'ws-1',
         pageId: 'p-1',
         creatorId: 'u-1',
+        user: USER,
         alias: 'foo',
         confirmReassign: true,
       });
@@ -269,6 +333,7 @@ describe('ShareAliasService', () => {
           workspaceId: 'ws-1',
           pageId: 'p-1',
           creatorId: 'u-1',
+          user: USER,
           alias: 'foo',
         });
         fail('expected ConflictException');
@@ -294,6 +359,7 @@ describe('ShareAliasService', () => {
           workspaceId: 'ws-1',
           pageId: 'p-1',
           creatorId: 'u-1',
+          user: USER,
           alias: 'foo',
         });
         fail('expected ConflictException');
@@ -317,6 +383,7 @@ describe('ShareAliasService', () => {
           workspaceId: 'ws-1',
           pageId: 'p-1',
           creatorId: 'u-1',
+          user: USER,
           alias: 'foo',
         });
         fail('expected ConflictException');
@@ -346,6 +413,7 @@ describe('ShareAliasService', () => {
           workspaceId: 'ws-1',
           pageId: 'p-1',
           creatorId: 'u-1',
+          user: USER,
           alias: 'foo',
         });
         fail('expected ConflictException');
@@ -375,6 +443,7 @@ describe('ShareAliasService', () => {
           workspaceId: 'ws-1',
           pageId: 'p-1',
           creatorId: 'u-1',
+          user: USER,
           alias: 'foo',
           confirmReassign: true,
         });
@@ -406,6 +475,7 @@ describe('ShareAliasService', () => {
           workspaceId: 'ws-1',
           pageId: 'p-1',
           creatorId: 'u-1',
+          user: USER,
           alias: 'ted',
         });
         fail('expected ConflictException');
@@ -428,6 +498,7 @@ describe('ShareAliasService', () => {
           workspaceId: 'ws-1',
           pageId: 'p-1',
           creatorId: 'u-1',
+          user: USER,
           alias: 'foo',
         }),
       ).rejects.toBeInstanceOf(BadRequestException);
@@ -450,18 +521,22 @@ describe('ShareAliasService', () => {
         alias: 'free-name',
         valid: true,
         available: true,
-        currentPageId: null,
       });
+      // SECURITY (#495): the availability probe must NOT leak any page id.
+      expect(res).not.toHaveProperty('currentPageId');
     });
 
-    it('reports taken with the current target page', async () => {
+    it('reports taken WITHOUT leaking the current target page id (#495)', async () => {
       const { service, shareAliasRepo } = makeService();
       shareAliasRepo.findByAliasAndWorkspace.mockResolvedValue({
         id: 'a-1',
         pageId: 'p-9',
       });
       const res = await service.checkAvailability('taken', 'ws-1');
-      expect(res).toMatchObject({ available: false, currentPageId: 'p-9' });
+      expect(res).toMatchObject({ available: false });
+      // The row exists (available:false) but its pageId is never returned — an
+      // authenticated member cannot map an alias name to a page id it can't view.
+      expect(res).not.toHaveProperty('currentPageId');
     });
   });
 

apps/server/src/core/share/share-alias.service.ts

@@ -7,7 +7,8 @@ import {
 import { ShareAliasRepo } from '@docmost/db/repos/share-alias/share-alias.repo';
 import { PageRepo } from '@docmost/db/repos/page/page.repo';
 import { ShareService } from './share.service';
-import { Page, ShareAlias } from '@docmost/db/types/entity.types';
+import { PageAccessService } from '../page/page-access/page-access.service';
+import { Page, ShareAlias, User } from '@docmost/db/types/entity.types';
 import { isValidShareAlias, normalizeShareAlias } from './share-alias.util';
 import { InjectKysely } from 'nestjs-kysely';
 import { KyselyDB } from '@docmost/db/types/kysely.types';
@@ -43,6 +44,7 @@ export class ShareAliasService {
     private readonly shareAliasRepo: ShareAliasRepo,
     private readonly pageRepo: PageRepo,
     private readonly shareService: ShareService,
+    private readonly pageAccessService: PageAccessService,
     @InjectKysely() private readonly db: KyselyDB,
   ) {}
 
@@ -55,9 +57,13 @@ export class ShareAliasService {
    *     `/l/<old>` link survives
    *   - name already points at pageId    -> no-op (idempotent)
    *   - name points at ANOTHER page      -> the "swap". Without confirmReassign
-   *     we throw 409 carrying the current target so the client can confirm;
-   *     with it we UPDATE the single row's page_id (every /l/<alias> link
-   *     follows the 302 to the new page instantly — no stale cache).
+   *     we throw 409 so the client can confirm. SECURITY (#495): the 409 reveals
+   *     the current target's title ONLY when `user` may VIEW that page, and never
+   *     its id — otherwise any member with one editable+shared page could iterate
+   *     alias names with confirmReassign=false and map them to (id, title) of
+   *     pages they cannot see. With confirmReassign we UPDATE the single row's
+   *     page_id (every /l/<alias> link follows the 302 to the new page instantly
+   *     — no stale cache).
    *
    * To keep the invariant self-healing we DELETE every other alias row still
    * pointing at this page (a legacy duplicate, or the target page's own former
@@ -77,8 +83,12 @@ export class ShareAliasService {
     creatorId: string;
     alias: string;
     confirmReassign?: boolean;
+    // The requesting user — used ONLY to gate whether the reassign 409 may reveal
+    // the current target page's title (view-permission check). Not an authz gate
+    // for the write itself (the controller already validated edit on `pageId`).
+    user: User;
   }): Promise<ShareAlias> {
-    const { workspaceId, pageId, creatorId, confirmReassign } = opts;
+    const { workspaceId, pageId, creatorId, confirmReassign, user } = opts;
     const alias = normalizeShareAlias(opts.alias);
     if (!isValidShareAlias(alias)) {
       throw new BadRequestException(
@@ -97,14 +107,30 @@ export class ShareAliasService {
         // The name is occupied by a DIFFERENT (or dangling) target page.
         if (byName && byName.pageId !== pageId) {
           if (!confirmReassign) {
+            // SECURITY (#495): only disclose the current target's TITLE, and only
+            // when the requester may VIEW that page. Never disclose its id (the
+            // client's confirm-reassign UX doesn't use it, and it is an enumerable
+            // identity). A member with one editable+shared page must NOT be able to
+            // iterate alias names and map them to (id, title) of pages they cannot
+            // see. When view is denied (or the alias is dangling) the 409 is the
+            // bare "occupied" fact — the client still shows a generic confirm modal.
             const currentPage = byName.pageId
               ? await this.pageRepo.findById(byName.pageId)
               : null;
+            let currentPageTitle: string | null = null;
+            if (currentPage) {
+              try {
+                await this.pageAccessService.validateCanView(currentPage, user);
+                currentPageTitle = currentPage.title ?? null;
+              } catch {
+                // No view permission on the target -> do not reveal its title.
+                currentPageTitle = null;
+              }
+            }
             throw new ConflictException({
               message: 'Alias already in use',
               code: 'ALIAS_REASSIGN_REQUIRED',
-              currentPageId: byName.pageId,
-              currentPageTitle: currentPage?.title ?? null,
+              currentPageTitle,
             });
           }
           // Confirmed swap. ORDER MATTERS: the partial unique index on
@@ -223,21 +249,27 @@ export class ShareAliasService {
     alias: string;
     valid: boolean;
     available: boolean;
-    currentPageId: string | null;
   }> {
     const alias = normalizeShareAlias(rawAlias);
     if (!isValidShareAlias(alias)) {
-      return { alias, valid: false, available: false, currentPageId: null };
+      return { alias, valid: false, available: false };
     }
     const existing = await this.shareAliasRepo.findByAliasAndWorkspace(
       alias,
       workspaceId,
     );
+    // SECURITY (#495): return ONLY the boolean availability. The previous shape
+    // leaked `currentPageId` — the id of whatever page the alias already targets —
+    // to ANY authenticated workspace member, with no view-permission check on that
+    // page. An attacker could enumerate alias names and map them to page ids they
+    // have no access to. The taken/free bit is all the "is this address free"
+    // probe needs. The reassign flow (setAlias 409) may surface the target's
+    // TITLE, but only behind a `validateCanView` on that page (see setAlias); it
+    // never returns the page id.
     return {
       alias,
       valid: true,
       available: !existing,
-      currentPageId: existing?.pageId ?? null,
     };
   }
 

apps/server/src/core/telemetry/client-metrics.constants.ts

@@ -24,6 +24,54 @@ export const ALLOWED_RATINGS = new Set<string>([
   'poor',
 ]);
 
+// The ONLY route labels accepted. The endpoint is anonymous, so an un-checked
+// `route` is a free-text write surface (arbitrary high-cardinality strings /
+// injected text into the metrics table). The client only ever sends a label from
+// a finite template dictionary (`templateRoute`), so we drop anything not in it.
+//
+// PARITY: this MUST mirror `KNOWN_ROUTE_TEMPLATES` in the client's
+// `apps/client/src/lib/telemetry/route-template.ts` (the canonical source). A
+// drift means legit client routes get dropped — keep the two in lockstep; the
+// client self-consistency test asserts `templateRoute` only emits these values.
+export const ALLOWED_ROUTE_TEMPLATES = new Set<string>([
+  '/',
+  'other',
+  // Static routes.
+  '/home',
+  '/spaces',
+  '/favorites',
+  '/login',
+  '/forgot-password',
+  '/password-reset',
+  '/setup/register',
+  '/settings/account/profile',
+  '/settings/account/preferences',
+  '/settings/workspace',
+  '/settings/ai',
+  '/settings/members',
+  '/settings/groups',
+  '/settings/spaces',
+  '/settings/sharing',
+  // Dynamic templates (slugs/ids are already collapsed to `:param`).
+  '/share/:shareId/p/:slug',
+  '/share/p/:slug',
+  '/share/:shareId',
+  '/p/:slug',
+  '/s/:space/p/:slug',
+  '/s/:space/trash',
+  '/s/:space',
+  '/labels/:label',
+  '/invites/:invitationId',
+  '/settings/groups/:groupId',
+]);
+
+// `attr` is a web-vitals attribution TARGET: a CSS-selector-ish string (an
+// element path like `html>body>div#app>button.cta`), never free prose. Constrain
+// it to a conservative CSS-selector charset so the anonymous endpoint cannot be
+// used to write arbitrary text / PII / markup into the metrics table. A value
+// containing anything outside this set is DROPPED (-> null); the event is kept.
+export const ATTR_ALLOWED_CHARSET = /^[A-Za-z0-9#.\-_> :()[\]="'*+~,]+$/;
+
 // Max events accepted per batch; the rest are ignored.
 export const MAX_EVENTS_PER_BATCH = 50;
 
@@ -77,14 +125,20 @@ export function sanitizeVitalEvent(
       ? e.rating
       : null;
 
+  // route: accept ONLY a known template label (dictionary check), else drop to
+  // null. The length cap stays as a cheap pre-guard before the Set lookup.
   let route: string | null = null;
   if (typeof e.route === 'string' && e.route.length > 0) {
-    route = e.route.slice(0, MAX_ROUTE_LENGTH);
+    const candidate = e.route.slice(0, MAX_ROUTE_LENGTH);
+    route = ALLOWED_ROUTE_TEMPLATES.has(candidate) ? candidate : null;
   }
 
+  // attr: truncate, then accept ONLY if it is a CSS-selector-shaped string
+  // (charset whitelist); anything with characters outside the set is dropped.
   let attr: string | null = null;
   if (typeof e.attr === 'string' && e.attr.length > 0) {
-    attr = e.attr.slice(0, MAX_ATTR_LENGTH);
+    const candidate = e.attr.slice(0, MAX_ATTR_LENGTH);
+    attr = ATTR_ALLOWED_CHARSET.test(candidate) ? candidate : null;
   }
 
   let docSize: number | null = null;

apps/server/src/core/telemetry/vitals.service.spec.ts

@@ -90,6 +90,44 @@ describe('VitalsService.buildRows', () => {
     expect(rows[0].attr).toHaveLength(MAX_ATTR_LENGTH);
   });
 
+  it('keeps a known route template but DROPS an unknown/free-text route (#495)', () => {
+    const rows = svc.buildRows(
+      {
+        events: [
+          { name: 'INP', value: 1, route: '/s/:space/p/:slug' }, // known
+          { name: 'INP', value: 2, route: '/s/acme-corp/p/secret-slug' }, // raw path (slugs) — not a template
+          { name: 'INP', value: 3, route: 'DROP TABLE client_metrics;--' }, // injected free text
+          { name: 'INP', value: 4, route: '/home' }, // known static
+        ],
+      },
+      WS,
+    );
+    expect(rows.map((r) => r.route)).toEqual([
+      '/s/:space/p/:slug',
+      null, // raw path dropped
+      null, // free text dropped
+      '/home',
+    ]);
+  });
+
+  it('DROPS an attr that is not a CSS-selector-shaped string (#495)', () => {
+    const rows = svc.buildRows(
+      {
+        events: [
+          { name: 'INP', value: 1, attr: 'div#app>button.cta' }, // valid selector
+          { name: 'INP', value: 2, attr: 'user@example.com wrote a note' }, // free text / PII
+          { name: 'INP', value: 3, attr: '<script>alert(1)</script>' }, // markup
+        ],
+      },
+      WS,
+    );
+    expect(rows.map((r) => r.attr)).toEqual([
+      'div#app>button.cta',
+      null,
+      null,
+    ]);
+  });
+
   it('caps the batch at 50 events', () => {
     const events = Array.from({ length: 200 }, () => ({ name: 'CLS', value: 1 }));
     const rows = svc.buildRows({ events }, WS);

apps/server/src/database/concurrent-indexes.spec.ts

@@ -0,0 +1,114 @@
+import * as path from 'path';
+import { readFileSync } from 'fs';
+
+// Mock ONLY kysely's `sql.raw(...).execute()` so we can observe what
+// ensureConcurrentIndexes runs and how it handles failures, without a DB.
+const execMock = jest.fn((_db: unknown) => Promise.resolve(undefined));
+const rawMock = jest.fn((_stmt: string) => ({ execute: execMock }));
+jest.mock('kysely', () => {
+  const actual = jest.requireActual('kysely');
+  return { ...actual, sql: { ...actual.sql, raw: rawMock } };
+});
+
+import { CONCURRENT_INDEXES, ensureConcurrentIndexes } from './concurrent-indexes';
+
+describe('ensureConcurrentIndexes', () => {
+  const fakeDb = { __topLevelKysely: true } as never;
+
+  beforeEach(() => {
+    execMock.mockReset().mockResolvedValue(undefined);
+    rawMock.mockClear();
+  });
+
+  it('redefines the inlinable f_unaccent BEFORE building any index, then builds each CONCURRENTLY outside a transaction', async () => {
+    const onLog = jest.fn();
+    await ensureConcurrentIndexes(fakeDb, onLog);
+
+    // One f_unaccent redefine + one statement per index.
+    expect(rawMock).toHaveBeenCalledTimes(CONCURRENT_INDEXES.length + 1);
+    const statements = rawMock.mock.calls.map((c) => c[0] as string);
+    // ORDER: the f_unaccent redefine MUST be first — otherwise an existing
+    // tenant's old 2-arg f_unaccent makes every CONCURRENTLY build fail.
+    expect(statements[0]).toContain('CREATE OR REPLACE FUNCTION f_unaccent');
+    expect(statements[0]).toContain('SELECT public.unaccent($1)');
+    expect(statements[0]).not.toContain('CREATE INDEX');
+    // The rest are the CONCURRENTLY index builds.
+    for (const stmt of statements.slice(1)) {
+      expect(stmt).toContain('CREATE INDEX CONCURRENTLY');
+      expect(stmt).toContain('IF NOT EXISTS');
+    }
+    // Executed against the top-level db (a transaction would forbid CONCURRENTLY).
+    for (const call of execMock.mock.calls) {
+      expect(call[0]).toBe(fakeDb);
+    }
+    expect(onLog).toHaveBeenCalledTimes(CONCURRENT_INDEXES.length + 1);
+  });
+
+  it('still builds the indexes when the f_unaccent redefine fails (fresh DB, best-effort)', async () => {
+    // Redefine (first execute) throws; the index loop must still run.
+    execMock
+      .mockRejectedValueOnce(new Error('extension "unaccent" does not exist'))
+      .mockResolvedValue(undefined);
+    const onLog = jest.fn();
+
+    await expect(
+      ensureConcurrentIndexes(fakeDb, onLog),
+    ).resolves.toBeUndefined();
+
+    // 1 redefine attempt + one per index.
+    expect(execMock).toHaveBeenCalledTimes(CONCURRENT_INDEXES.length + 1);
+    const errored = onLog.mock.calls.filter((c) => c[1] !== undefined);
+    expect(errored).toHaveLength(1);
+    expect(String(errored[0][1])).toContain('unaccent');
+  });
+
+  it('is best-effort: a failing index does not abort the rest and is reported', async () => {
+    // Redefine ok; fail the FIRST index; the remaining ones must still be tried.
+    execMock
+      .mockResolvedValueOnce(undefined) // f_unaccent redefine
+      .mockRejectedValueOnce(new Error('relation "pages" does not exist'))
+      .mockResolvedValue(undefined);
+    const onLog = jest.fn();
+
+    await expect(
+      ensureConcurrentIndexes(fakeDb, onLog),
+    ).resolves.toBeUndefined();
+
+    expect(execMock).toHaveBeenCalledTimes(CONCURRENT_INDEXES.length + 1);
+    const errored = onLog.mock.calls.filter((c) => c[1] !== undefined);
+    expect(errored).toHaveLength(1);
+    expect(String(errored[0][1])).toContain('does not exist');
+  });
+});
+
+// DRIFT GUARD: each CONCURRENT_INDEXES entry pre-builds an index that a plain
+// migration ALSO creates with `IF NOT EXISTS`. If the two expressions diverge,
+// Postgres would treat them as different indexes and the pre-build would NOT
+// make the migration a no-op. Assert the migration files still contain each
+// index's functional expression.
+describe('CONCURRENT_INDEXES parity with the migrations', () => {
+  const migrationsDir = path.join(__dirname, 'migrations');
+  const files = [
+    '20260705T120000-perf-indexes.ts',
+    '20260706T120000-search-lookup-trgm.ts',
+  ].map((f) => readFileSync(path.join(migrationsDir, f), 'utf8'));
+  const allMigrationSrc = files.join('\n');
+
+  it.each(CONCURRENT_INDEXES.map((i) => [i.name, i]))(
+    'migration source still creates %s with the same expression',
+    (_name, idx) => {
+      // Extract the `USING gin ((...expr...) gin_trgm_ops)` tail from the
+      // canonical create and assert the migration source contains it verbatim.
+      const m = (idx as { create: string }).create.match(
+        /ON \w+ (USING gin .+)$/,
+      );
+      expect(m).not.toBeNull();
+      const expr = (m as RegExpMatchArray)[1];
+      expect(allMigrationSrc).toContain(expr);
+      // And the migration must build it by the SAME index name.
+      expect(allMigrationSrc).toContain(
+        `CREATE INDEX IF NOT EXISTS ${(idx as { name: string }).name}`,
+      );
+    },
+  );
+});

apps/server/src/database/concurrent-indexes.ts

@@ -0,0 +1,142 @@
+import { Kysely, sql } from 'kysely';
+
+/**
+ * Indexes that MUST be built with `CREATE INDEX CONCURRENTLY` so an auto-deploy
+ * migration never takes a `SHARE` lock that blocks writes on a hot table
+ * (`pages`) for the — potentially minutes-long — GIN trigram build (#495 item 12).
+ *
+ * Kysely runs each migration INSIDE a transaction (Postgres has transactional
+ * DDL), and `CREATE INDEX CONCURRENTLY` cannot run inside a transaction block, so
+ * these cannot live in an ordinary migration. Instead {@link ensureConcurrentIndexes}
+ * builds them out-of-band (no transaction) BEFORE the migrator runs; the matching
+ * migrations keep a plain `CREATE INDEX IF NOT EXISTS` as a backstop, which then
+ * no-ops because the index already exists. So:
+ *   - existing prod DB, incremental deploy: the pre-build FIRST redefines
+ *     `f_unaccent` to its index-inlinable 1-arg form (see below), THEN builds each
+ *     index CONCURRENTLY (no write lock); the migration's IF NOT EXISTS no-ops →
+ *     the write-blocking build is gone;
+ *   - fresh DB (or a DB whose `pages` / `unaccent` extension does not exist yet):
+ *     the pre-build fails and is swallowed (best-effort), and the migration builds
+ *     the index normally on an empty/small table where the lock is irrelevant.
+ *
+ * NOT a strict "worst case = previous behaviour":
+ *   - The f_unaccent form matters. The trigram expressions use `LOWER(f_unaccent(col))`.
+ *     The INDEX-INLINABLE 1-arg `f_unaccent(text)` (`SELECT public.unaccent($1)`)
+ *     is created INSIDE migration 20260705, which runs AFTER this pre-build. On an
+ *     existing tenant the live `f_unaccent` is still the OLD 2-arg form from
+ *     20250729, which Postgres CANNOT inline into a CONCURRENTLY index expression
+ *     (`function unaccent(unknown, text) does not exist ... during inlining`) — the
+ *     build fails outright. So this pre-build redefines `f_unaccent` to the 1-arg
+ *     form FIRST (idempotent, output-identical, in lockstep with 20260705). Without
+ *     that redefine the whole feature is dead-on-arrival for the very case it
+ *     targets.
+ *   - An INTERRUPTED `CREATE INDEX CONCURRENTLY` (killed pod, cancelled query) with
+ *     a working `f_unaccent` leaves an INVALID index behind. A subsequent name-based
+ *     `IF NOT EXISTS` — in BOTH this pre-build and the migration backstop — sees the
+ *     name and skips, so the invalid index is NEVER repaired automatically and the
+ *     query keeps seq-scanning until an operator `DROP INDEX`es it. The old
+ *     in-transaction build could not leave an invalid index (a failed tx rolled the
+ *     whole index back), so this is a genuinely new failure mode, not "= previous".
+ *     (A future hardening could `DROP` an `indisvalid = false` index before
+ *     rebuilding; not done here.)
+ *
+ * The `create` text is the CANONICAL definition — it MUST match the migration's
+ * `IF NOT EXISTS` create expression exactly (same functional expression + opclass)
+ * or Postgres would treat them as two different indexes.
+ */
+export const CONCURRENT_INDEXES: ReadonlyArray<{
+  name: string;
+  create: string;
+}> = [
+  {
+    // #348 perf-indexes — pages.title trigram (coalesce-free functional expr).
+    name: 'idx_pages_title_trgm',
+    create:
+      'CREATE INDEX CONCURRENTLY IF NOT EXISTS idx_pages_title_trgm ' +
+      'ON pages USING gin ((LOWER(f_unaccent(title))) gin_trgm_ops)',
+  },
+  {
+    // #348 perf-indexes — users.name trigram (member search-suggest).
+    name: 'idx_users_name_trgm',
+    create:
+      'CREATE INDEX CONCURRENTLY IF NOT EXISTS idx_users_name_trgm ' +
+      'ON users USING gin ((LOWER(f_unaccent(name))) gin_trgm_ops)',
+  },
+  {
+    // #443 search-lookup-trgm — pages.text_content trigram (the slow, large one).
+    name: 'idx_pages_text_content_trgm',
+    create:
+      'CREATE INDEX CONCURRENTLY IF NOT EXISTS idx_pages_text_content_trgm ' +
+      'ON pages USING gin ((LOWER(f_unaccent(text_content))) gin_trgm_ops)',
+  },
+];
+
+/**
+ * Idempotent, output-identical redefinition of `f_unaccent` to the INDEX-INLINABLE
+ * 1-arg form. MUST stay byte-for-byte in lockstep with migration
+ * `20260705T120000-perf-indexes.ts` (same signature + body): the trigram indexes
+ * above only build CONCURRENTLY once `f_unaccent(text)` inlines, and on an existing
+ * tenant it is still the old 2-arg form until that migration runs — which is AFTER
+ * this pre-build.
+ */
+const F_UNACCENT_REDEF = `
+    CREATE OR REPLACE FUNCTION f_unaccent(text)
+      RETURNS text
+      LANGUAGE sql
+      IMMUTABLE PARALLEL SAFE STRICT
+    AS $func$
+      SELECT public.unaccent($1);
+    $func$
+`;
+
+/**
+ * Best-effort, non-transactional pre-build of {@link CONCURRENT_INDEXES}. Run
+ * BEFORE the migrator so the blocking `CREATE INDEX` in the corresponding
+ * migration becomes an `IF NOT EXISTS` no-op.
+ *
+ * `db` MUST be the top-level Kysely instance (NOT a transaction): each statement
+ * then executes on its own connection with no surrounding `BEGIN`, which is
+ * required for `CONCURRENTLY`. Every statement is independent and swallowed on
+ * error: the migration backstop still builds the index, so a failure here is
+ * never fatal. `onLog` reports progress/failures for the caller to route to its
+ * logger.
+ *
+ * ORDER MATTERS: redefine `f_unaccent` to the inlinable form BEFORE the index
+ * loop — otherwise an existing tenant's old 2-arg `f_unaccent` makes every
+ * CONCURRENTLY build fail and the feature is a no-op (see the module docstring).
+ */
+export async function ensureConcurrentIndexes(
+  db: Kysely<any>,
+  onLog?: (message: string, error?: unknown) => void,
+): Promise<void> {
+  // Make f_unaccent inlinable FIRST. On a fresh DB (no `unaccent` extension yet)
+  // this throws harmlessly and is swallowed — the migration builds everything.
+  try {
+    await sql.raw(F_UNACCENT_REDEF).execute(db);
+    onLog?.('f_unaccent redefined to the index-inlinable 1-arg form');
+  } catch (error) {
+    onLog?.(
+      'f_unaccent redefine skipped (fresh DB / no unaccent extension) — ' +
+        'the migration will define it and build the indexes',
+      error,
+    );
+  }
+
+  for (const idx of CONCURRENT_INDEXES) {
+    try {
+      await sql.raw(idx.create).execute(db);
+      onLog?.(`Concurrent index ensured: ${idx.name}`);
+    } catch (error) {
+      // Non-fatal by design — the migration's IF NOT EXISTS create is the
+      // backstop. Benign on a fresh DB (the table/extension does not exist yet).
+      // NOT benign, but still swallowed, on an existing tenant whose f_unaccent
+      // could not be redefined above (then the migration builds the index NON-
+      // concurrently under the SHARE lock this feature meant to avoid).
+      onLog?.(
+        `Concurrent index pre-build skipped for ${idx.name} ` +
+          `(will fall back to the in-migration build)`,
+        error,
+      );
+    }
+  }
+}

apps/server/src/database/execute-tx-after-commit.spec.ts

@@ -0,0 +1,113 @@
+import { executeTx, registerAfterCommit } from './utils';
+import { KyselyDB, KyselyTransaction } from './types/kysely.types';
+
+// Post-commit hook contract (#495 item 13): a side effect registered via
+// registerAfterCommit must run ONLY AFTER the owning transaction commits, and a
+// hook registered against a passed-through existingTrx must fire at the OUTER
+// commit boundary — never inside the inner call. We fake the Kysely transaction
+// runner so the ordering is observable without a real DB.
+
+/**
+ * A minimal db whose `.transaction().execute(cb)` records the commit ORDER: it
+ * runs `cb(trx)`, pushes 'commit' onto `log` (simulating the real commit that
+ * happens after the callback resolves), then returns the callback's result.
+ */
+function fakeDb(log: string[]): { db: KyselyDB; trx: KyselyTransaction } {
+  const trx = { __fakeTrx: true } as unknown as KyselyTransaction;
+  const db = {
+    transaction: () => ({
+      execute: async (cb: (t: KyselyTransaction) => Promise<unknown>) => {
+        const result = await cb(trx);
+        log.push('commit');
+        return result;
+      },
+    }),
+  } as unknown as KyselyDB;
+  return { db, trx };
+}
+
+describe('executeTx post-commit hooks', () => {
+  it('runs an afterCommit hook only AFTER the transaction commits', async () => {
+    const log: string[] = [];
+    const { db } = fakeDb(log);
+
+    await executeTx(db, async (trx) => {
+      log.push('body');
+      registerAfterCommit(trx, () => {
+        log.push('hook');
+      });
+      // The hook must NOT have run yet — the tx is still open.
+      expect(log).toEqual(['body']);
+    });
+
+    // Order proves post-commit: body → commit → hook (never body → hook → commit).
+    expect(log).toEqual(['body', 'commit', 'hook']);
+  });
+
+  it('drains hooks registered against a passed-through existingTrx at the OUTER commit', async () => {
+    const log: string[] = [];
+    const { db, trx: outerTrx } = fakeDb(log);
+
+    await executeTx(db, async (outer) => {
+      // Nested executeTx reuses the outer trx: it must NOT commit or drain now.
+      await executeTx(
+        db,
+        async (inner) => {
+          registerAfterCommit(inner, () => {
+            log.push('inner-hook');
+          });
+        },
+        outer,
+      );
+      // Still inside the outer tx — the inner hook has not fired.
+      expect(log).toEqual([]);
+    });
+
+    // The single (outer) commit drains the hook registered on the shared trx.
+    expect(log).toEqual(['commit', 'inner-hook']);
+    // Sanity: the trx the hooks were registered against is the outer one.
+    expect(outerTrx).toBeDefined();
+  });
+
+  it('a hook failure does not reject the already-committed executeTx', async () => {
+    const log: string[] = [];
+    const { db } = fakeDb(log);
+
+    await expect(
+      executeTx(db, async (trx) => {
+        registerAfterCommit(trx, () => {
+          throw new Error('cache del blew up');
+        });
+        registerAfterCommit(trx, () => {
+          log.push('second-hook-still-runs');
+        });
+        return 'ok';
+      }),
+    ).resolves.toBe('ok');
+
+    // The throwing hook is swallowed; a later hook still runs.
+    expect(log).toEqual(['commit', 'second-hook-still-runs']);
+  });
+
+  it('does NOT run afterCommit hooks when the transaction body throws (rollback)', async () => {
+    // The body rejects -> the fake transaction never pushes 'commit' and
+    // db.transaction().execute() rejects, mirroring a real rolled-back tx. The
+    // drain runs only AFTER the awaited (committed) transaction, so a rollback
+    // must leave every registered hook UN-run — otherwise a cache-bust / event
+    // would fire for a write that never landed.
+    const log: string[] = [];
+    const { db } = fakeDb(log);
+    const hook = jest.fn();
+
+    await expect(
+      executeTx(db, async (trx) => {
+        registerAfterCommit(trx, hook);
+        throw new Error('write failed -> rollback');
+      }),
+    ).rejects.toThrow('write failed -> rollback');
+
+    // No commit happened, and the post-commit hook never ran.
+    expect(log).toEqual([]); // no 'commit'
+    expect(hook).not.toHaveBeenCalled();
+  });
+});

apps/server/src/database/jsonb-bind.spec.ts

@@ -35,4 +35,25 @@ describe('jsonbBind', () => {
     expect(out).not.toBeNull();
     expect(out).toBeDefined();
   });
+
+  // preserveEmpty (#476): opts a column OUT of the empty-to-null collapse so an
+  // empty container is persisted verbatim (e.g. `[]` = deny-all for
+  // tool_allowlist). null stays null regardless of the flag.
+  describe('preserveEmpty', () => {
+    it('returns a (non-null) bind for an empty array', () => {
+      const out = jsonbBind([], { preserveEmpty: true });
+      expect(out).not.toBeNull();
+      expect(out).toBeDefined();
+    });
+
+    it('returns a (non-null) bind for an empty object', () => {
+      const out = jsonbBind({}, { preserveEmpty: true });
+      expect(out).not.toBeNull();
+      expect(out).toBeDefined();
+    });
+
+    it('still returns null for null (null means null, flag or not)', () => {
+      expect(jsonbBind(null, { preserveEmpty: true })).toBeNull();
+    });
+  });
 });

apps/server/src/database/migrations/20260705T120000-perf-indexes.ts

@@ -33,16 +33,18 @@ import { type Kysely, sql } from 'kysely';
  *    - comments: `findPageComments` does WHERE page_id ORDER BY id ASC, but only
  *      `(page_id)` exists → extra sort.
  *
- * DEPLOY-TIME LOCK WARNING: these are plain (non-CONCURRENT) CREATE INDEX
- * statements — CONCURRENTLY is impossible because Kysely runs each migration in a
- * transaction. They take a SHARE lock that BLOCKS writes (INSERT/UPDATE/DELETE) on
- * pages/users/groups/comments/page_history for the duration of the build. The two
- * GIN trigram builds on pages.title / users.name are the slow ones and can take
- * minutes on a large tenant → a write-outage window during the deploy migration.
- * For large installations, run this migration in a maintenance window, or build
- * the trigram indexes out-of-band with CREATE INDEX CONCURRENTLY before deploying
- * (then this migration's `IF NOT EXISTS` is a no-op). Small/typical tenants are
- * unaffected.
+ * DEPLOY-TIME LOCK: these are plain (non-CONCURRENT) CREATE INDEX statements —
+ * CONCURRENTLY is impossible HERE because Kysely runs each migration in a
+ * transaction. The two GIN trigram builds on pages.title / users.name are the
+ * slow ones and would take a SHARE lock that BLOCKS writes on pages/users for
+ * minutes on a large tenant. To avoid that, `ensureConcurrentIndexes`
+ * (database/concurrent-indexes.ts) now pre-builds BOTH trigram indexes with
+ * CREATE INDEX CONCURRENTLY (no transaction) BEFORE the migrator runs, so on an
+ * existing DB the two `IF NOT EXISTS` trigram creates below no-op and no write
+ * lock is taken. On a fresh DB the pre-build is skipped and they build on an
+ * empty table. Keep the two trigram creates in lockstep with their CANONICAL
+ * definitions in CONCURRENT_INDEXES. (The plain b-tree indexes further down are
+ * fast metadata-only builds; they are not pre-built.)
  */
 export async function up(db: Kysely<any>): Promise<void> {
   // Index-compatible, output-identical redefinition of f_unaccent (see header).

apps/server/src/database/migrations/20260706T120000-search-lookup-trgm.ts

@@ -26,13 +26,16 @@ import { type Kysely, sql } from 'kysely';
  *    to update than b-trees); on the small instances this fork targets that cost
  *    is acceptable and the read win on agent lookups is the priority.
  *
- * DEPLOY-TIME LOCK WARNING: plain (non-CONCURRENT) CREATE INDEX — Kysely runs
- * each migration in a transaction, so CONCURRENTLY is impossible. The build takes
- * a SHARE lock that BLOCKS writes on `pages` for its duration. The text_content
- * GIN build is the slow one and can take minutes on a large tenant. For large
- * installations, run this in a maintenance window or build the index out-of-band
- * with CREATE INDEX CONCURRENTLY before deploying (then `IF NOT EXISTS` no-ops
- * here). Small/typical tenants are unaffected.
+ * DEPLOY-TIME LOCK: this is a plain (non-CONCURRENT) CREATE INDEX — Kysely runs
+ * each migration in a transaction, so CONCURRENTLY is impossible HERE, and the
+ * build would take a SHARE lock that BLOCKS writes on `pages` for its duration
+ * (the text_content GIN build can take minutes on a large tenant). To avoid that,
+ * `ensureConcurrentIndexes` (database/concurrent-indexes.ts) now pre-builds this
+ * index with CREATE INDEX CONCURRENTLY (no transaction) BEFORE the migrator runs,
+ * so on an existing DB the `IF NOT EXISTS` below no-ops and no write lock is taken.
+ * On a fresh DB the pre-build is skipped and this builds it on an empty table
+ * where the lock is irrelevant. Keep this create in lockstep with the CANONICAL
+ * definition in CONCURRENT_INDEXES (same expression + opclass).
  */
 export async function up(db: Kysely<any>): Promise<void> {
   // The title predicate is served by #348's idx_pages_title_trgm — see header.

apps/server/src/database/migrations/20260707T120000-ai-chat-metadata.ts

@@ -0,0 +1,24 @@
+import { type Kysely, sql } from 'kysely';
+
+export async function up(db: Kysely<any>): Promise<void> {
+  // Chat-level metadata bag (#490). First use: the deferred-tool ACTIVATION set
+  // (`activatedTools`) is persisted here so it survives across turns — previously
+  // the set was reset every turn, forcing the model to re-run loadTools and pay a
+  // fresh round-trip to re-activate the same tools each turn. On load the stored
+  // set is intersected with the current valid deferred names, so an allowlist /
+  // role change can never inject a now-nonexistent tool.
+  //
+  // jsonb, defaulted to '{}' so every row (incl. pre-migration ones, backfilled
+  // by the default) is a readable object — the app never has to null-guard the
+  // bag itself, only individual keys.
+  await db.schema
+    .alterTable('ai_chats')
+    .addColumn('metadata', 'jsonb', (col) =>
+      col.notNull().defaultTo(sql`'{}'::jsonb`),
+    )
+    .execute();
+}
+
+export async function down(db: Kysely<any>): Promise<void> {
+  await db.schema.alterTable('ai_chats').dropColumn('metadata').execute();
+}

apps/server/src/database/repos/ai-chat/ai-chat-message.repo.delta.int-spec.ts

@@ -0,0 +1,280 @@
+import { randomUUID } from 'crypto';
+import { CamelCasePlugin, Kysely, sql } from 'kysely';
+import { PostgresJSDialect } from 'kysely-postgres-js';
+// NOT a default import: the project tsconfig is `module: commonjs` with NO
+// esModuleInterop, so `import postgres from 'postgres'` compiles to
+// `postgres_1.default(...)` and the CJS `postgres` export has no `.default` —
+// it threw in beforeAll, was swallowed as "DB unreachable", and SILENTLY voided
+// all six tests. Mirror the working integration harness (test/integration/db.ts).
+import * as postgres from 'postgres';
+import { AiChatMessageRepo } from './ai-chat-message.repo';
+import { AiChatRunRepo } from './ai-chat-run.repo';
+
+/**
+ * #491 delta-poll — OBSERVABLE-PROPERTY tests against a LIVE Postgres (the local
+ * gitmost test DB, docker `gitmost-test-pg` on :5432), not "rows through a mock"
+ * (a mock cannot observe the DB clock nor the overlap-window race — the very
+ * things that matter here). Drives the REAL repo methods (`findByChatUpdatedAfter`,
+ * the now()-stamped `update`) and asserts:
+ *   1. delta-relevant writes stamp `updatedAt` from the DB clock, not the app clock
+ *      (proven by faking the process clock far into the future and observing the
+ *      stamp stays on real DB time);
+ *   2. the poll returns only rows changed after the cursor, ordered, with a fresh
+ *      DB-clock cursor;
+ *   3. the "committed late but stamped earlier than the cursor" RACE is caught by
+ *      the overlap window (a naive `updatedAt > cursor` would MISS it);
+ *   4. the overlap GUARANTEES repeats across close polls — the contract behind the
+ *      client's idempotent merge (mergeById).
+ *
+ * INTEGRATION lane (`*.int-spec.ts`): runs under `test:int`, whose global-setup
+ * DROPS + RE-CREATES + MIGRATES `docmost_test`, so the real `ai_chat_messages` /
+ * `ai_chat_runs` tables EXIST here. (It was previously a `.spec.ts` defaulting to
+ * the UNmigrated dev `docmost`; in the CI unit lane — where `WAL_TEST_DATABASE_URL`
+ * is unset and only `test:int` migrates — that meant 5/6 ERROR
+ * `relation "ai_chat_messages" does not exist`, silently voiding coverage of the
+ * risky cursor/overlap logic. Renaming to `.int-spec.ts` + defaulting the DSN to
+ * `docmost_test` fixes the CI fidelity.)
+ *
+ * FK triggers are bypassed (`session_replication_role = replica`) so synthetic
+ * chat/workspace ids need no parent fixtures; a single pooled connection (max 1)
+ * keeps that session setting for every query. SKIPS cleanly when the DB is
+ * unreachable so a DB-less CI never breaks.
+ */
+const CONN =
+  process.env.WAL_TEST_DATABASE_URL ??
+  process.env.TEST_DATABASE_URL ??
+  'postgresql://docmost:docmost_dev_pw@localhost:5432/docmost_test';
+
+let db: Kysely<any>;
+let sqlClient: ReturnType<typeof postgres>;
+let msgRepo: AiChatMessageRepo;
+let runRepo: AiChatRunRepo;
+let reachable = false;
+
+const workspaceId = randomUUID();
+const chatId = randomUUID();
+
+beforeAll(async () => {
+  try {
+    sqlClient = postgres(CONN, { max: 1, onnotice: () => {} });
+    db = new Kysely<any>({
+      dialect: new PostgresJSDialect({ postgres: sqlClient }),
+      plugins: [new CamelCasePlugin()],
+    });
+    // Single connection keeps this session-scoped bypass for the whole suite.
+    await sql`set session_replication_role = replica`.execute(db);
+    await sql`select 1`.execute(db);
+    reachable = true;
+  } catch (err) {
+    reachable = false;
+    // A genuine connection failure (ECONNREFUSED etc.) is a legitimate skip on a
+    // DB-less CI. A PROGRAMMING error (bad import, typo, driver misuse) must NOT
+    // masquerade as "DB unreachable" and silently void the whole suite (that is
+    // exactly the bug that hid this spec's zero coverage) — rethrow it so the
+    // suite fails LOUDLY.
+    const msg = String((err as Error)?.message ?? err);
+    if (
+      !/ECONNREFUSED|ENOTFOUND|ETIMEDOUT|EHOSTUNREACH|connect|terminating|password|authentication|role .* does not exist|database .* does not exist/i.test(
+        msg,
+      )
+    ) {
+      throw err;
+    }
+  }
+  msgRepo = new AiChatMessageRepo(db as never);
+  runRepo = new AiChatRunRepo(db as never);
+});
+
+afterAll(async () => {
+  if (db) {
+    try {
+      await db
+        .deleteFrom('aiChatMessages')
+        .where('workspaceId', '=', workspaceId)
+        .execute();
+      await db
+        .deleteFrom('aiChatRuns')
+        .where('workspaceId', '=', workspaceId)
+        .execute();
+    } catch {
+      /* best-effort cleanup */
+    }
+    await db.destroy();
+  }
+});
+
+afterEach(() => {
+  jest.useRealTimers();
+});
+
+async function seedMessage(overrides: Record<string, unknown> = {}) {
+  return msgRepo.insert({
+    chatId,
+    workspaceId,
+    userId: null as never,
+    role: 'assistant',
+    content: 'x',
+    status: 'streaming',
+    ...overrides,
+  } as never);
+}
+
+async function dbNow(): Promise<string> {
+  const r = await sql<{ now: Date }>`select now() as now`.execute(db);
+  return r.rows[0].now.toISOString();
+}
+
+// Fake ONLY the Date object (so in-process `new Date()`/`Date.now()` jump), while
+// leaving every TIMER function real. Faking timers wholesale freezes postgres.js's
+// internal connection/query timers, so the awaited DB round-trip would hang the
+// test (and the afterAll cleanup) at the jest 5s cap. With Date-only faking the
+// query resolves normally, and we still prove the stamp is the DB clock (not the
+// skewed process clock).
+function fakeDateOnly(iso: string): void {
+  jest.useFakeTimers({
+    doNotFake: [
+      'hrtime',
+      'nextTick',
+      'performance',
+      'queueMicrotask',
+      'requestAnimationFrame',
+      'cancelAnimationFrame',
+      'requestIdleCallback',
+      'cancelIdleCallback',
+      'setImmediate',
+      'clearImmediate',
+      'setInterval',
+      'clearInterval',
+      'setTimeout',
+      'clearTimeout',
+    ],
+    now: new Date(iso),
+  });
+}
+
+const maybe = (name: string, fn: () => Promise<void>) =>
+  it(name, async () => {
+    if (!reachable) {
+      console.warn(`SKIP (${name}): test DB unreachable at ${CONN}`);
+      return;
+    }
+    await fn();
+  });
+
+describe('AiChatMessageRepo.findByChatUpdatedAfter (#491 delta poll)', () => {
+  maybe('null cursor returns no rows and a fresh DB-clock cursor', async () => {
+    const before = await dbNow();
+    const { rows, cursor } = await msgRepo.findByChatUpdatedAfter(
+      chatId,
+      workspaceId,
+      null,
+    );
+    expect(rows).toEqual([]);
+    expect(new Date(cursor).getTime()).toBeGreaterThanOrEqual(
+      new Date(before).getTime(),
+    );
+  });
+
+  maybe('returns only rows changed after the cursor', async () => {
+    const { cursor: c0 } = await msgRepo.findByChatUpdatedAfter(
+      chatId,
+      workspaceId,
+      null,
+    );
+    const m = await seedMessage();
+    const { rows, cursor: c1 } = await msgRepo.findByChatUpdatedAfter(
+      chatId,
+      workspaceId,
+      c0,
+    );
+    expect(rows.map((r) => r.id)).toContain(m.id);
+    // Cursor is monotonic (advances).
+    expect(new Date(c1).getTime()).toBeGreaterThanOrEqual(
+      new Date(c0).getTime(),
+    );
+  });
+
+  maybe(
+    'RACE: a row stamped BEFORE the cursor but seen after is caught by the overlap',
+    async () => {
+      // Cursor taken now; then a row appears whose updatedAt is 2s in the PAST
+      // (committed late on another connection but stamped earlier). A naive
+      // `updatedAt > cursor` would MISS it; the 5s overlap window catches it.
+      const cursor = await dbNow();
+      const m = await seedMessage();
+      await sql`update ai_chat_messages set updated_at = now() - interval '2 seconds' where id = ${m.id}`.execute(
+        db,
+      );
+      const { rows } = await msgRepo.findByChatUpdatedAfter(
+        chatId,
+        workspaceId,
+        cursor,
+      );
+      expect(rows.map((r) => r.id)).toContain(m.id);
+    },
+  );
+
+  maybe(
+    'overlap GUARANTEES repeats across close polls (idempotent-merge contract)',
+    async () => {
+      const { cursor: c0 } = await msgRepo.findByChatUpdatedAfter(
+        chatId,
+        workspaceId,
+        null,
+      );
+      const m = await seedMessage();
+      const first = await msgRepo.findByChatUpdatedAfter(
+        chatId,
+        workspaceId,
+        c0,
+      );
+      expect(first.rows.map((r) => r.id)).toContain(m.id);
+      // Immediately re-poll with the JUST-returned cursor: the row is still within
+      // the overlap window, so it is returned AGAIN — the client MUST dedupe by id.
+      const second = await msgRepo.findByChatUpdatedAfter(
+        chatId,
+        workspaceId,
+        first.cursor,
+      );
+      expect(second.rows.map((r) => r.id)).toContain(m.id);
+    },
+  );
+
+  maybe(
+    'update() stamps updatedAt from the DB clock, not the app clock',
+    async () => {
+      const m = await seedMessage();
+      // Skew the PROCESS clock ~73 years into the future (Date only). If the stamp
+      // came from `new Date()` the row would read year 2099; sql now() keeps it on
+      // DB time.
+      fakeDateOnly('2099-01-01T00:00:00Z');
+      const updated = await msgRepo.update(m.id, workspaceId, {
+        content: 'y',
+      });
+      jest.useRealTimers();
+      expect(updated).toBeDefined();
+      expect(new Date(updated!.updatedAt).getFullYear()).toBeLessThan(2099);
+    },
+  );
+
+  maybe(
+    'run update() also stamps updatedAt from the DB clock',
+    async () => {
+      const run = await runRepo.insert({
+        chatId,
+        workspaceId,
+        createdBy: null as never,
+        trigger: 'user',
+        status: 'running',
+        stepCount: 0,
+      } as never);
+      fakeDateOnly('2099-01-01T00:00:00Z');
+      const updated = await runRepo.update(run.id, workspaceId, {
+        stepCount: 1,
+      });
+      jest.useRealTimers();
+      expect(updated).toBeDefined();
+      expect(new Date(updated!.updatedAt).getFullYear()).toBeLessThan(2099);
+    },
+  );
+});

apps/server/src/database/repos/ai-chat/ai-chat-message.repo.ts

@@ -25,6 +25,20 @@ const SWEEP_STREAMING_STALE_MS = 10 * 60 * 1000; // 10 minutes
 // into memory; far above any realistic transcript length.
 const FIND_ALL_BY_CHAT_LIMIT = 5000;
 
+// Delta-poll overlap (#491): the poll query reaches this far BEHIND the client's
+// echoed cursor, so a row that committed with an `updatedAt` marginally before the
+// previous cursor was taken (on another autocommit connection) is still caught.
+// Sized well above realistic single-row commit skew; the client merge is
+// idempotent by id (mergeById), so the guaranteed repeats the overlap produces are
+// harmless.
+export const DELTA_POLL_OVERLAP_SECONDS = 5;
+
+// Hard cap on rows one delta poll returns — a safety bound (a poll should carry a
+// handful of just-changed rows, never a whole transcript). Ordered by (updatedAt,
+// id) asc, so on the pathological overflow the OLDEST changes win and the newest
+// are picked up by the next poll (its cursor did not advance past them).
+export const DELTA_POLL_MAX_ROWS = 500;
+
 @Injectable()
 export class AiChatMessageRepo {
   private readonly logger = new Logger(AiChatMessageRepo.name);
@@ -139,6 +153,72 @@ export class AiChatMessageRepo {
       .executeTakeFirst();
   }
 
+  /**
+   * Delta read (#491) for the degraded poll: the chat's messages whose row
+   * changed AFTER the client's `cursor`, plus a FRESH cursor taken from the DB
+   * clock. Replaces the old "refetch ALL infinite-query pages every 2.5s with
+   * full parts" poll — the client seeds once (findByChat) and thereafter pulls
+   * only the deltas and merges them by id (mergeById).
+   *
+   * Cursor: a DB-clock timestamp (now()) the client echoes back each poll. All
+   * delta-relevant writes stamp `updatedAt` with now() (see `update` /
+   * `finalizeOwner`), so this is a SINGLE monotonic axis. The query overlaps the
+   * cursor by DELTA_POLL_OVERLAP_SECONDS to catch a row committed with an
+   * `updatedAt` marginally BEFORE the previous cursor was taken on another
+   * connection (single-row autocommit UPDATEs; no long transactions). The overlap
+   * GUARANTEES occasional REPEATS, so the client merge MUST be idempotent by id.
+   *
+   * `cursor === null` (first poll after the full seed) returns NO rows — there is
+   * nothing "new" relative to a just-loaded seed — only the fresh cursor to start
+   * the delta chain. The fresh cursor is read AFTER the rows, so it is >= every
+   * returned row's `updatedAt` (they were read strictly earlier) — a row that
+   * commits between the rows-read and the cursor-read is at most
+   * DELTA_POLL_OVERLAP_SECONDS behind the returned cursor, so the next poll's
+   * overlap window always re-includes it (no miss).
+   */
+  async findByChatUpdatedAfter(
+    chatId: string,
+    workspaceId: string,
+    cursor: string | null,
+  ): Promise<{ rows: AiChatMessage[]; cursor: string }> {
+    if (cursor === null) {
+      const nowRow = await sql<{ now: Date }>`select now() as now`.execute(
+        this.db,
+      );
+      return { rows: [], cursor: nowRow.rows[0].now.toISOString() };
+    }
+    // Overlap the client cursor by DELTA_POLL_OVERLAP_SECONDS, computed in SQL off
+    // the echoed cursor so the whole comparison stays on the DB clock.
+    const rows = await this.db
+      .selectFrom('aiChatMessages')
+      .select(this.baseFields)
+      .where('chatId', '=', chatId)
+      .where('workspaceId', '=', workspaceId)
+      .where('deletedAt', 'is', null)
+      .where(
+        'updatedAt',
+        '>',
+        sql<Date>`${cursor}::timestamptz - make_interval(secs => ${DELTA_POLL_OVERLAP_SECONDS})`,
+      )
+      .orderBy('updatedAt', 'asc')
+      .orderBy('id', 'asc')
+      .limit(DELTA_POLL_MAX_ROWS)
+      .execute();
+    // When the page filled (pathological overflow), DO NOT advance the cursor to
+    // now(): that would skip the changed rows past the cap that this poll did not
+    // return. Resume from the last returned row's updatedAt instead (the next
+    // poll's overlap re-includes ties by id). In the normal case the fresh DB-clock
+    // now() is the cursor.
+    if (rows.length === DELTA_POLL_MAX_ROWS) {
+      return {
+        rows,
+        cursor: rows[rows.length - 1].updatedAt.toISOString(),
+      };
+    }
+    const nowRow = await sql<{ now: Date }>`select now() as now`.execute(this.db);
+    return { rows, cursor: nowRow.rows[0].now.toISOString() };
+  }
+
   async insert(
     insertable: InsertableAiChatMessage,
     trx?: KyselyTransaction,
@@ -172,7 +252,13 @@ export class AiChatMessageRepo {
     const db = dbOrTx(this.db, opts?.trx);
     let query = db
       .updateTable('aiChatMessages')
-      .set({ ...(patch as Record<string, unknown>), updatedAt: new Date() })
+      // #491: stamp `updatedAt` from the DB clock (sql now()), NOT the app clock
+      // (new Date()). The delta-poll cursor (findByChatUpdatedAfter) is a single
+      // DB-clock axis; a per-step 'streaming' UPDATE stamped with the app clock
+      // would be a SECOND, skewed clock source and could leave a row's updatedAt
+      // just under a cursor taken from now() on another connection — an
+      // independent source of delta MISSES. All delta-relevant writes use now().
+      .set({ ...(patch as Record<string, unknown>), updatedAt: sql`now()` })
       .where('id', '=', id)
       .where('workspaceId', '=', workspaceId);
     // Concurrency guard (#183 review): a per-step 'streaming' update must NEVER
@@ -214,7 +300,9 @@ export class AiChatMessageRepo {
     const db = dbOrTx(this.db, trx);
     return db
       .updateTable('aiChatMessages')
-      .set({ ...(patch as Record<string, unknown>), updatedAt: new Date() })
+      // #491: DB-clock stamp (see `update`) — this terminal write flips the row's
+      // status, which the delta poll must observe on the shared now() cursor axis.
+      .set({ ...(patch as Record<string, unknown>), updatedAt: sql`now()` })
       .where('id', '=', id)
       .where('workspaceId', '=', workspaceId)
       .where((eb) =>
@@ -249,7 +337,9 @@ export class AiChatMessageRepo {
       .set({
         status,
         metadata: sql`coalesce(metadata, '{}'::jsonb) || jsonb_build_object('finalizeFailed', true)`,
-        updatedAt: new Date(),
+        // #491: DB-clock stamp (see `update`) so a reconcile status flip lands on
+        // the same now() cursor axis the delta poll reads.
+        updatedAt: sql`now()`,
       })
       .where('id', '=', id)
       .where('workspaceId', '=', workspaceId)
@@ -307,7 +397,9 @@ export class AiChatMessageRepo {
       .set({
         status: 'aborted',
         metadata: sql`coalesce(m.metadata, '{}'::jsonb) || jsonb_build_object('finalizeFailed', true)`,
-        updatedAt: new Date(),
+        // #491: DB-clock stamp (see `update`). The staleness WHERE below stays on
+        // the app clock — a >minutes window makes the ms-scale skew irrelevant.
+        updatedAt: sql`now()`,
       })
       .where('m.status', '=', 'streaming')
       .where('m.updatedAt', '<', staleBefore)
@@ -351,7 +443,8 @@ export class AiChatMessageRepo {
       .set({
         status: 'aborted',
         metadata: sql`coalesce(metadata, '{}'::jsonb) || jsonb_build_object('finalizeFailed', true)`,
-        updatedAt: new Date(),
+        // #491: DB-clock stamp (see `update`). Staleness WHERE stays app-clock.
+        updatedAt: sql`now()`,
       })
       .where('status', '=', 'streaming')
       .where('updatedAt', '<', staleBefore)

apps/server/src/database/repos/ai-chat/ai-chat-run.repo.spec.ts

@@ -62,10 +62,17 @@ describe('AiChatRunRepo.sweepRunning', () => {
     // ...but a fresh 'running' run (updatedAt = now) must NOT be skipped: no
     // updatedAt predicate at all on the boot path.
     expect(rec.wheres.some(([col]) => col === 'updatedAt')).toBe(false);
-    // It flips to 'aborted' and stamps finishedAt.
-    expect(rec.set).toEqual(
-      expect.objectContaining({ status: 'aborted', finishedAt: expect.any(Date) }),
-    );
+    // It flips to 'aborted' and stamps finishedAt + updatedAt. #491: the stamps
+    // are now DB-clock `sql now()` expressions (raw builders), NOT app-clock
+    // `new Date()`, so the run row shares the delta poll's single now() cursor axis
+    // — assert they are present and are the sql raw-builder objects (not a Date,
+    // not undefined).
+    expect(rec.set?.status).toBe('aborted');
+    for (const stamp of ['finishedAt', 'updatedAt'] as const) {
+      expect(rec.set?.[stamp]).toBeDefined();
+      expect(rec.set?.[stamp]).not.toBeInstanceOf(Date);
+      expect(typeof rec.set?.[stamp]).toBe('object');
+    }
   });
 
   it('phase-2 path: an explicit staleMs reintroduces the updatedAt window', async () => {

apps/server/src/database/repos/ai-chat/ai-chat-run.repo.ts

@@ -136,7 +136,11 @@ export class AiChatRunRepo {
     const db = dbOrTx(this.db, trx);
     return db
       .updateTable('aiChatRuns')
-      .set({ ...(patch as Record<string, unknown>), updatedAt: new Date() })
+      // #491: DB-clock stamp (sql now()) so the run row shares the delta poll's
+      // single now() cursor axis with the assistant message rows — a run-status
+      // change (the run fact the delta carries) must never sit on a skewed app
+      // clock relative to the message updatedAt cursor.
+      .set({ ...(patch as Record<string, unknown>), updatedAt: sql`now()` })
       .where('id', '=', id)
       .where('workspaceId', '=', workspaceId)
       .returning(this.baseFields)
@@ -162,14 +166,15 @@ export class AiChatRunRepo {
     trx?: KyselyTransaction,
   ): Promise<AiChatRun | undefined> {
     const db = dbOrTx(this.db, trx);
-    const now = new Date();
     return db
       .updateTable('aiChatRuns')
       .set({
         status: patch.status,
         error: patch.error,
-        finishedAt: now,
-        updatedAt: now,
+        // #491: DB-clock stamps (finished_at + updated_at) so the terminal run
+        // fact lands on the delta poll's now() cursor axis.
+        finishedAt: sql`now()`,
+        updatedAt: sql`now()`,
       })
       .where('id', '=', id)
       .where('workspaceId', '=', workspaceId)
@@ -192,7 +197,8 @@ export class AiChatRunRepo {
     const db = dbOrTx(this.db, trx);
     return db
       .updateTable('aiChatRuns')
-      .set({ stopRequestedAt: new Date(), updatedAt: new Date() })
+      // #491: DB-clock stamps (see `update`).
+      .set({ stopRequestedAt: sql`now()`, updatedAt: sql`now()` })
       .where('id', '=', id)
       .where('workspaceId', '=', workspaceId)
       .where('status', 'in', ACTIVE_RUN_STATUSES as unknown as string[])
@@ -249,13 +255,14 @@ export class AiChatRunRepo {
     trx?: KyselyTransaction,
   ): Promise<number> {
     const db = dbOrTx(this.db, trx);
-    const now = new Date();
     let query = db
       .updateTable('aiChatRuns')
       .set({
         status: 'aborted',
-        finishedAt: now,
-        updatedAt: now,
+        // #491: DB-clock stamps (see `update`). The staleness WHERE below stays on
+        // the app clock — a >minutes window makes the ms-scale skew irrelevant.
+        finishedAt: sql`now()`,
+        updatedAt: sql`now()`,
         error: sql`coalesce(error, ${'Run interrupted by a server restart.'})`,
       })
       .where('status', 'in', ACTIVE_RUN_STATUSES as unknown as string[]);
@@ -263,7 +270,7 @@ export class AiChatRunRepo {
     // sibling replica's live run is never aborted. Omitted on the phase-1 boot
     // sweep -> unconditional.
     if (typeof opts.staleMs === 'number') {
-      const staleBefore = new Date(now.getTime() - opts.staleMs);
+      const staleBefore = new Date(Date.now() - opts.staleMs);
       query = query.where('updatedAt', '<', staleBefore);
     }
     const rows = await query.returning('id').execute();

apps/server/src/database/repos/ai-chat/ai-mcp-server.repo.ts

@@ -78,7 +78,9 @@ export class AiMcpServerRepo {
         headersEnc: values.headersEnc ?? null,
         // jsonb column: the postgres driver would otherwise encode a JS array as
         // a Postgres array literal. Bind the JSON text and cast it to jsonb.
-        toolAllowlist: jsonbBind(values.toolAllowlist),
+        // preserveEmpty (#476): `[]` is a real value here (deny-all), distinct
+        // from null ("no restriction") — it must round-trip as `[]`, not null.
+        toolAllowlist: jsonbBind(values.toolAllowlist, { preserveEmpty: true }),
         // Plain text column: blank/whitespace-only guidance is stored as null.
         instructions: blankToNull(values.instructions),
         enabled: values.enabled ?? true,
@@ -111,7 +113,10 @@ export class AiMcpServerRepo {
     if (patch.url !== undefined) set.url = patch.url;
     if (patch.headersEnc !== undefined) set.headersEnc = patch.headersEnc;
     if (patch.toolAllowlist !== undefined) {
-      set.toolAllowlist = jsonbBind(patch.toolAllowlist);
+      // preserveEmpty (#476): see insert — `[]` (deny-all) must not become null.
+      set.toolAllowlist = jsonbBind(patch.toolAllowlist, {
+        preserveEmpty: true,
+      });
     }
     if (patch.instructions !== undefined) {
       // Blank/whitespace-only guidance clears the column (stored as null).
@@ -158,7 +163,9 @@ export function blankToNull(value: string | null | undefined): string | null {
  * fix), so the driver hands back a string like `'["a","b"]'` rather than an
  * array. Be tolerant: normalize a JSON string to its value, then accept it only
  * if it is an array of strings; null / a non-array / unparseable value / an
- * array with a non-string element all become null (unrestricted).
+ * array with a non-string element all become null. NOTE: null here only means
+ * "could not parse" — the null-vs-deny-all policy decision lives in
+ * normalizeRow (#476: present-but-corrupt fails CLOSED to `[]`).
  */
 export function parseToolAllowlist(value: unknown): string[] | null {
   // Shape guard only; the legacy double-encoding self-heal lives in
@@ -173,17 +180,20 @@ export function parseToolAllowlist(value: unknown): string[] | null {
 /**
  * Normalize a DB row so `toolAllowlist` is always `string[] | null`.
  *
- * FAIL-OPEN logging: a stored value that is present but cannot be parsed into a
- * string[] (corrupt JSON, a non-array, non-string elements) degrades to `null` =
- * "no restriction", so the agent silently gets ALL of the server's tools. Log
- * one line (server id only, never the contents) so that widening is not silent.
+ * FAIL-CLOSED (#476): a stored value that is PRESENT but cannot be parsed into
+ * a string[] (corrupt JSON, a non-array, non-string elements) degrades to `[]`
+ * = deny-all, so a corrupted allowlist can never silently widen to "the agent
+ * gets ALL of the server's tools" (the old fail-open null). An error line is
+ * logged (server id only, never the contents) so the admin can repair the row.
+ * A column that is truly NULL/absent stays `null` = "no restriction".
  */
 function normalizeRow(row: AiMcpServer): AiMcpServer {
   const parsed = parseToolAllowlist(row.toolAllowlist);
   if (parsed === null && row.toolAllowlist != null) {
-    logger.warn(
-      `Corrupt tool_allowlist for MCP server ${row.id}; ignoring it (no tool restriction applied)`,
+    logger.error(
+      `Corrupt tool_allowlist for MCP server ${row.id}; failing closed (NO tools allowed) — re-save the server's allowlist to repair it`,
     );
+    return { ...row, toolAllowlist: [] };
   }
   return { ...row, toolAllowlist: parsed };
 }

apps/server/src/database/repos/page/page.repo.ts

@@ -1,7 +1,7 @@
 import { Injectable } from '@nestjs/common';
 import { InjectKysely } from 'nestjs-kysely';
 import { KyselyDB, KyselyTransaction } from '../../types/kysely.types';
-import { dbOrTx, executeTx } from '../../utils';
+import { dbOrTx, executeTx, registerAfterCommit } from '../../utils';
 import {
   InsertablePage,
   Page,
@@ -349,14 +349,23 @@ export class PageRepo {
     pageId: string,
     deletedById: string,
     workspaceId: string,
+    // Optional caller transaction. When passed, the reads + soft-delete run in
+    // THAT transaction (so a caller holding a `FOR UPDATE` lock on the row — e.g.
+    // the temporary-note sweeper — can delete under the lock without deadlocking
+    // on a nested independent transaction) and the PAGE_SOFT_DELETED broadcast is
+    // deferred to the caller's COMMIT via registerAfterCommit (so a rolled-back
+    // delete never broadcasts). With no trx the behaviour is unchanged: own
+    // transaction, broadcast right after it commits.
+    existingTrx?: KyselyTransaction,
   ): Promise<void> {
     const currentDate = new Date();
+    const readDb = dbOrTx(this.db, existingTrx);
 
     // Read the root snapshot up front so PAGE_SOFT_DELETED can carry it without
     // a post-commit DB read (variant A). Only the root of the deleted subtree is
     // needed for the tree broadcast — the client `treeModel.remove` drops all
     // descendants, so we don't snapshot/broadcast every descendant.
-    const rootSnapshot = await this.db
+    const rootSnapshot = await readDb
       .selectFrom('pages')
       .select([
         'id',
@@ -371,7 +380,7 @@ export class PageRepo {
       .where('deletedAt', 'is', null)
       .executeTakeFirst();
 
-    const descendants = await this.db
+    const descendants = await readDb
       .withRecursive('page_descendants', (db) =>
         db
           .selectFrom('pages')
@@ -393,39 +402,60 @@ export class PageRepo {
     const pageIds = descendants.map((d) => d.id);
 
     if (pageIds.length > 0) {
-      await executeTx(this.db, async (trx) => {
-        await trx
-          .updateTable('pages')
-          .set({
-            deletedById: deletedById,
-            deletedAt: currentDate,
-          })
-          .where('id', 'in', pageIds)
-          .where('deletedAt', 'is', null)
-          .execute();
+      // Reuse the caller's transaction when given (executeTx passes it straight
+      // through), else own a fresh one.
+      await executeTx(
+        this.db,
+        async (trx) => {
+          await trx
+            .updateTable('pages')
+            .set({
+              deletedById: deletedById,
+              deletedAt: currentDate,
+            })
+            .where('id', 'in', pageIds)
+            .where('deletedAt', 'is', null)
+            .execute();
 
-        await trx.deleteFrom('shares').where('pageId', 'in', pageIds).execute();
-      });
+          await trx
+            .deleteFrom('shares')
+            .where('pageId', 'in', pageIds)
+            .execute();
+        },
+        existingTrx,
+      );
 
-      this.eventEmitter.emit(EventName.PAGE_SOFT_DELETED, {
-        pageIds: pageIds,
-        workspaceId,
-        // Root-only snapshot: one `deleteTreeNode` is enough, the client removes
-        // the whole subtree. Skip if the root vanished between the two reads.
-        pages: rootSnapshot
-          ? [
-              {
-                id: rootSnapshot.id,
-                slugId: rootSnapshot.slugId,
-                title: rootSnapshot.title,
-                icon: rootSnapshot.icon,
-                position: rootSnapshot.position,
-                spaceId: rootSnapshot.spaceId,
-                parentPageId: rootSnapshot.parentPageId,
-              },
-            ]
-          : [],
-      });
+      const emitSoftDeleted = () => {
+        this.eventEmitter.emit(EventName.PAGE_SOFT_DELETED, {
+          pageIds: pageIds,
+          workspaceId,
+          // Root-only snapshot: one `deleteTreeNode` is enough, the client
+          // removes the whole subtree. Skip if the root vanished between reads.
+          pages: rootSnapshot
+            ? [
+                {
+                  id: rootSnapshot.id,
+                  slugId: rootSnapshot.slugId,
+                  title: rootSnapshot.title,
+                  icon: rootSnapshot.icon,
+                  position: rootSnapshot.position,
+                  spaceId: rootSnapshot.spaceId,
+                  parentPageId: rootSnapshot.parentPageId,
+                },
+              ]
+            : [],
+        });
+      };
+
+      if (existingTrx) {
+        // Inside a caller transaction: the delete above is NOT committed yet.
+        // Defer the tree broadcast to the caller's commit so a rolled-back delete
+        // never broadcasts a phantom removal.
+        registerAfterCommit(existingTrx, emitSoftDeleted);
+      } else {
+        // Own transaction already committed above — broadcast now.
+        emitSoftDeleted();
+      }
     }
   }
 

apps/server/src/database/repos/workspace/workspace.repo.ts

@@ -3,7 +3,7 @@ import { CACHE_MANAGER } from '@nestjs/cache-manager';
 import { Cache } from 'cache-manager';
 import { InjectKysely } from 'nestjs-kysely';
 import { KyselyDB, KyselyTransaction } from '../../types/kysely.types';
-import { dbOrTx } from '../../utils';
+import { dbOrTx, registerAfterCommit } from '../../utils';
 import {
   InsertableWorkspace,
   UpdatableWorkspace,
@@ -80,16 +80,30 @@ export class WorkspaceRepo {
    */
   private async bustWorkspaceCache(
     workspace?: Pick<Workspace, 'hostname'> | undefined,
+    trx?: KyselyTransaction,
   ): Promise<void> {
-    try {
-      await this.cacheManager.del(CacheKey.WORKSPACE_SELF_HOSTED);
-      if (workspace?.hostname) {
-        await this.cacheManager.del(
-          CacheKey.WORKSPACE_BY_HOST(workspace.hostname),
-        );
+    const del = async () => {
+      try {
+        await this.cacheManager.del(CacheKey.WORKSPACE_SELF_HOSTED);
+        if (workspace?.hostname) {
+          await this.cacheManager.del(
+            CacheKey.WORKSPACE_BY_HOST(workspace.hostname),
+          );
+        }
+      } catch {
+        // cache is best-effort; TTL is the backstop
       }
-    } catch {
-      // cache is best-effort; TTL is the backstop
+    };
+    if (trx) {
+      // Inside a caller transaction the write is NOT yet committed: busting now
+      // opens a repopulation window (a concurrent reader reloads the cache with
+      // the pre-commit / stale row, which then survives until TTL). Defer the del
+      // to the transaction's commit (drained by the owning executeTx) (#495).
+      registerAfterCommit(trx, del);
+    } else {
+      // No transaction: the mutation above already auto-committed, so this del is
+      // already post-commit.
+      await del();
     }
   }
 
@@ -180,7 +194,7 @@ export class WorkspaceRepo {
       .where('id', '=', workspaceId)
       .returning(this.baseFields)
       .executeTakeFirst();
-    await this.bustWorkspaceCache(workspace);
+    await this.bustWorkspaceCache(workspace, trx);
     return workspace;
   }
 
@@ -195,7 +209,7 @@ export class WorkspaceRepo {
       .returning(this.baseFields)
       .executeTakeFirst();
     // Bust the cached "not found" so a fresh install / new tenant is seen at once.
-    await this.bustWorkspaceCache(workspace);
+    await this.bustWorkspaceCache(workspace, trx);
     return workspace;
   }
 
@@ -249,7 +263,7 @@ export class WorkspaceRepo {
       .where('id', '=', workspaceId)
       .returning(this.baseFields)
       .executeTakeFirst();
-    await this.bustWorkspaceCache(workspace);
+    await this.bustWorkspaceCache(workspace, trx);
     return workspace;
   }
 
@@ -271,7 +285,7 @@ export class WorkspaceRepo {
       .where('id', '=', workspaceId)
       .returning(this.baseFields)
       .executeTakeFirst();
-    await this.bustWorkspaceCache(workspace);
+    await this.bustWorkspaceCache(workspace, trx);
     return workspace;
   }
 
@@ -326,7 +340,7 @@ export class WorkspaceRepo {
       .where('id', '=', workspaceId)
       .returning(this.baseFields)
       .executeTakeFirst();
-    await this.bustWorkspaceCache(workspace);
+    await this.bustWorkspaceCache(workspace, trx);
     return workspace;
   }
 
@@ -354,7 +368,7 @@ export class WorkspaceRepo {
       .where('id', '=', workspaceId)
       .returning(this.baseFields)
       .executeTakeFirst();
-    await this.bustWorkspaceCache(workspace);
+    await this.bustWorkspaceCache(workspace, trx);
     return workspace;
   }
 
@@ -376,7 +390,7 @@ export class WorkspaceRepo {
       .where('id', '=', workspaceId)
       .returning(this.baseFields)
       .executeTakeFirst();
-    await this.bustWorkspaceCache(workspace);
+    await this.bustWorkspaceCache(workspace, trx);
     return workspace;
   }
 
@@ -398,7 +412,7 @@ export class WorkspaceRepo {
       .where('id', '=', workspaceId)
       .returning(this.baseFields)
       .executeTakeFirst();
-    await this.bustWorkspaceCache(workspace);
+    await this.bustWorkspaceCache(workspace, trx);
     return workspace;
   }
 

apps/server/src/database/services/migration.service.ts

@@ -4,6 +4,7 @@ import { promises as fs } from 'fs';
 import { Migrator, FileMigrationProvider } from 'kysely';
 import { InjectKysely } from 'nestjs-kysely';
 import { KyselyDB } from '@docmost/db/types/kysely.types';
+import { ensureConcurrentIndexes } from '@docmost/db/concurrent-indexes';
 
 @Injectable()
 export class MigrationService {
@@ -12,6 +13,16 @@ export class MigrationService {
   constructor(@InjectKysely() private readonly db: KyselyDB) {}
 
   async migrateToLatest(): Promise<void> {
+    // Build write-blocking trigram indexes CONCURRENTLY (no transaction) BEFORE
+    // the migrator runs, so the corresponding in-migration `CREATE INDEX IF NOT
+    // EXISTS` no-ops instead of taking a SHARE lock on `pages` during deploy
+    // (#495). Best-effort: on a fresh DB (no `pages`/`f_unaccent` yet) this is a
+    // no-op and the migrations build the index normally.
+    await ensureConcurrentIndexes(this.db, (message, error) => {
+      if (error) this.logger.warn(`${message}: ${String(error)}`);
+      else this.logger.log(message);
+    });
+
     const migrator = new Migrator({
       db: this.db,
       provider: new FileMigrationProvider({

apps/server/src/database/types/db.d.ts

@@ -606,6 +606,9 @@ export interface AiChats {
   // The document the chat was created in (open page at first message). NULL =>
   // started outside any document. ON DELETE SET NULL on the page FK.
   pageId: string | null;
+  // Chat-level metadata bag (#490). jsonb, defaulted to '{}'. First key:
+  // `activatedTools` — the deferred-tool activation set persisted across turns.
+  metadata: Generated<Json>;
   createdAt: Generated<Timestamp>;
   updatedAt: Generated<Timestamp>;
   deletedAt: Timestamp | null;

apps/server/src/database/utils.ts

@@ -6,16 +6,76 @@ import { KyselyDB, KyselyTransaction } from './types/kysely.types';
  * If an existing transaction is provided, it directly executes the callback with it.
  * Otherwise, it starts a new transaction using the provided database instance and executes the callback within that transaction.
  */
+/**
+ * Post-commit side-effect hooks, keyed by the transaction they were registered
+ * against. A WeakMap so an abandoned/never-drained transaction's entry is GC'd
+ * with the trx object (no leak). Used by {@link registerAfterCommit} /
+ * {@link executeTx}.
+ */
+const afterCommitHooks = new WeakMap<
+  KyselyTransaction,
+  Array<() => Promise<void> | void>
+>();
+
+/**
+ * Register a side effect to run ONLY AFTER the transaction that owns `trx`
+ * commits. THE fix for "bust the cache inside the open transaction" bugs: a
+ * cache-invalidation (or any read-your-write-visible side effect) done while the
+ * writing transaction is still open opens a window where a concurrent reader
+ * repopulates the cache with the PRE-COMMIT (stale) row, so after commit the
+ * cache holds the old value until its TTL. Deferring the effect to post-commit
+ * closes that window.
+ *
+ * The hook is drained by the OUTERMOST {@link executeTx} that actually owns
+ * (created) this transaction — so registering against a passed-through
+ * `existingTrx` still fires at the real commit boundary, not at the inner call.
+ * NOTE: a hook registered against a transaction that was NOT created via
+ * `executeTx` (untracked) will never be drained — always create transactions
+ * through `executeTx` when you rely on post-commit hooks.
+ */
+export function registerAfterCommit(
+  trx: KyselyTransaction,
+  hook: () => Promise<void> | void,
+): void {
+  const existing = afterCommitHooks.get(trx);
+  if (existing) existing.push(hook);
+  else afterCommitHooks.set(trx, [hook]);
+}
+
 export async function executeTx<T>(
   db: KyselyDB,
   callback: (trx: KyselyTransaction) => Promise<T>,
   existingTrx?: KyselyTransaction,
 ): Promise<T> {
   if (existingTrx) {
-    return await callback(existingTrx); // Execute callback with existing transaction
-  } else {
-    return await db.transaction().execute((trx) => callback(trx)); // Start new transaction and execute callback
+    // Reuse the caller's transaction. Any post-commit hooks registered here are
+    // drained by the OUTER executeTx that created `existingTrx`, at the true
+    // commit boundary — so we must NOT drain them now.
+    return await callback(existingTrx);
   }
+  // We OWN this transaction: run the body, then (only once it has COMMITTED)
+  // drain the post-commit hooks registered against it during the body.
+  let ownTrx: KyselyTransaction | undefined;
+  const result = await db.transaction().execute((trx) => {
+    ownTrx = trx;
+    return callback(trx);
+  });
+  if (ownTrx) {
+    const hooks = afterCommitHooks.get(ownTrx);
+    if (hooks) {
+      afterCommitHooks.delete(ownTrx);
+      for (const hook of hooks) {
+        // Best-effort: a failed side effect (e.g. a cache del) must not fail the
+        // already-committed transaction.
+        try {
+          await hook();
+        } catch {
+          // swallow — the durable write already committed
+        }
+      }
+    }
+  }
+  return result;
 }
 
 /*
@@ -78,18 +138,30 @@ export function violatedConstraint(err: unknown): string | undefined {
  * verbatim); `::jsonb` then parses it into a real array/object. Read-side
  * parsers repair rows written the old buggy way without a migration.
  *
- * Returns `null` for null/undefined and for "empty" values (an empty array, or
- * an object with no own enumerable keys) — callers treat empty as "clear/unset",
- * so an empty allowlist/config never round-trips as `[]`/`{}`.
+ * Returns `null` for null/undefined. By default it ALSO returns `null` for
+ * "empty" values (an empty array, or an object with no own enumerable keys) —
+ * most callers treat empty as "clear/unset", so an empty config never
+ * round-trips as `[]`/`{}`.
+ *
+ * `preserveEmpty` (issue #476) opts a column OUT of that empty-to-null
+ * normalization so `[]`/`{}` are persisted as real jsonb values. Needed where
+ * empty and null mean DIFFERENT things: an empty `tool_allowlist` is
+ * deny-all ("zero tools allowed"), while null is "no restriction" — collapsing
+ * `[]` to null silently widened deny-all to allow-all. Deliberately an opt-in
+ * flag, NOT a global change: the other jsonb callers (model_config, source)
+ * keep the empty-means-unset contract.
  */
 export function jsonbBind<T>(
   value: T | null | undefined,
+  opts?: { preserveEmpty?: boolean },
 ): RawBuilder<T> | null {
   if (value === null || value === undefined) return null;
-  if (Array.isArray(value)) {
-    if (value.length === 0) return null;
-  } else if (typeof value === 'object') {
-    if (Object.keys(value as object).length === 0) return null;
+  if (!opts?.preserveEmpty) {
+    if (Array.isArray(value)) {
+      if (value.length === 0) return null;
+    } else if (typeof value === 'object') {
+      if (Object.keys(value as object).length === 0) return null;
+    }
   }
   return sql<T>`${JSON.stringify(value)}::text::jsonb`;
 }

apps/server/src/integrations/ai/ai-settings.service.spec.ts

@@ -99,10 +99,12 @@ describe('AiSettingsService.getMasked reindex progress', () => {
     // actually pins the progress.total branch rather than coincidentally
     // matching the DB fallback. With fix #1 the two sources agree in practice,
     // but getMasked must still return progress.total when a record is active.
+    const startedAt = Date.now();
     reindexProgress.get.mockResolvedValue({
       total: 500,
       done: 120,
-      startedAt: Date.now(),
+      startedAt,
+      runId: 'run-abc',
     });
 
     const masked = await service.getMasked(WORKSPACE_ID);
@@ -110,6 +112,10 @@ describe('AiSettingsService.getMasked reindex progress', () => {
     expect(masked.indexedPages).toBe(120); // progress.done, not DB 478
     expect(masked.totalPages).toBe(500); // progress.total, not DB 478
     expect(masked.reindexing).toBe(true);
+    // The status payload must carry the run identity so the client can key its
+    // poll on it (a changed runId => a NEW run).
+    expect(masked.runId).toBe('run-abc');
+    expect(masked.reindexStartedAt).toBe(startedAt);
   });
 
   it('falls back to countIndexedPages when no reindex is active', async () => {
@@ -121,6 +127,10 @@ describe('AiSettingsService.getMasked reindex progress', () => {
     expect(masked.indexedPages).toBe(478);
     expect(masked.totalPages).toBe(478);
     expect(masked.reindexing).toBe(false);
+    // No active run -> no run identity surfaced (the client keeps its prior
+    // steady-state behaviour).
+    expect(masked.runId).toBeUndefined();
+    expect(masked.reindexStartedAt).toBeUndefined();
   });
 });
 

apps/server/src/integrations/ai/ai-settings.service.ts

@@ -245,6 +245,9 @@ export class AiSettingsService {
       // Max context window for the chat header badge denominator. Stored as
       // ::text; 0/unset/invalid = no limit (undefined).
       chatContextWindow: parsePositiveInt(provider.chatContextWindow),
+      // RAW stored value (#490): the replay budgeter reads this to distinguish an
+      // explicit `0` (off-switch) from unset, which parsePositiveInt cannot.
+      chatContextWindowRaw: provider.chatContextWindow,
       // Plain passthrough; getChatModel defaults unset to 'openai-compatible'.
       chatApiStyle: provider.chatApiStyle,
       // Cheap model id for the anonymous public-share assistant; reuses the chat
@@ -371,6 +374,12 @@ export class AiSettingsService {
       totalPages,
       // Optional hint for the client: a reindex run is currently in progress.
       reindexing: progress != null,
+      // Per-run identity so the client can key its poll on a stable run id and
+      // reset its per-run state when a NEW run starts. Present only while a run
+      // is active; `runId` may be '' for a legacy/degraded record (the client
+      // treats that as "no identity").
+      runId: progress?.runId,
+      reindexStartedAt: progress?.startedAt,
     };
   }
 

apps/server/src/integrations/ai/ai.service.provider-fetch.spec.ts

@@ -0,0 +1,86 @@
+// `.provider` alone cannot prove the gemini/ollama chat factories were built
+// with the instrumented streaming fetch — a regression dropping it (which drops
+// them back to the global undici fetch: no keep-alive recycle, no reset retries,
+// unbounded silence timeout; incident classes #140/#175/#310) would still pass.
+// So mock the factories and assert the exact fetch argument. jest.mock is
+// module-scoped, hence a dedicated file.
+
+const mockGeminiModel = { provider: 'google.generative-ai', modelId: 'm' };
+const mockOllamaModel = { provider: 'ollama.chat', modelId: 'm' };
+
+// jest allows `mock`-prefixed vars inside a jest.mock factory.
+const mockCreateGoogle = jest.fn((_settings: unknown) => () => mockGeminiModel);
+const mockCreateOllama = jest.fn((_settings: unknown) => () => mockOllamaModel);
+
+jest.mock('@ai-sdk/google', () => ({
+  createGoogleGenerativeAI: (settings: unknown) => mockCreateGoogle(settings),
+}));
+jest.mock('ai-sdk-ollama', () => ({
+  createOllama: (settings: unknown) => mockCreateOllama(settings),
+}));
+
+import { AiService } from './ai.service';
+
+describe('AiService.getChatModel provider transport fetch (gemini/ollama)', () => {
+  function serviceWith(cfg: Record<string, unknown>) {
+    const aiSettings = {
+      resolve: jest.fn().mockResolvedValue(cfg),
+    };
+    return new AiService(
+      // eslint-disable-next-line @typescript-eslint/no-explicit-any
+      aiSettings as any,
+      { find: jest.fn() } as never,
+      { decryptSecret: jest.fn() } as never,
+    );
+  }
+
+  beforeEach(() => {
+    mockCreateGoogle.mockClear();
+    mockCreateOllama.mockClear();
+  });
+
+  it('builds the gemini chat model with the instrumented streaming fetch', async () => {
+    await serviceWith({
+      driver: 'gemini',
+      chatModel: 'gemini-2.5-pro',
+      apiKey: 'the-key',
+    }).getChatModel('ws-1');
+    expect(mockCreateGoogle).toHaveBeenCalledTimes(1);
+    expect(mockCreateGoogle).toHaveBeenCalledWith(
+      expect.objectContaining({
+        apiKey: 'the-key',
+        fetch: expect.any(Function),
+      }),
+    );
+  });
+
+  it('builds the ollama chat model with the instrumented streaming fetch', async () => {
+    await serviceWith({
+      driver: 'ollama',
+      chatModel: 'llama3',
+      baseUrl: 'http://localhost:11434/api',
+    }).getChatModel('ws-1');
+    expect(mockCreateOllama).toHaveBeenCalledTimes(1);
+    expect(mockCreateOllama).toHaveBeenCalledWith(
+      expect.objectContaining({
+        baseURL: 'http://localhost:11434/api',
+        fetch: expect.any(Function),
+      }),
+    );
+  });
+
+  it('reuses ONE service-lifetime fetch instance across both providers', async () => {
+    const svc = serviceWith({
+      driver: 'gemini',
+      chatModel: 'gemini-2.5-pro',
+      apiKey: 'k',
+    });
+    await svc.getChatModel('ws-1');
+    const geminiFetch = mockCreateGoogle.mock.calls[0][0] as { fetch: unknown };
+    // Same instance on a second call — the fetch is held for the service
+    // lifetime to reuse the streaming dispatcher's connection pool.
+    await svc.getChatModel('ws-1');
+    const geminiFetch2 = mockCreateGoogle.mock.calls[1][0] as { fetch: unknown };
+    expect(geminiFetch.fetch).toBe(geminiFetch2.fetch);
+  });
+});

apps/server/src/integrations/ai/ai.service.ts

@@ -190,10 +190,22 @@ export class AiService {
         }).chat(chatModel);
       }
       case 'gemini':
-        return createGoogleGenerativeAI({ apiKey })(chatModel);
+        // Route gemini through the same instrumented streaming fetch as openai
+        // (finite silence timeouts + keep-alive recycling + pre-response
+        // connection-reset retry). Without it the provider ran on the global
+        // undici fetch — no keep-alive recycle, no reset retries, default
+        // (unbounded silence) timeout — so incident classes #140/#175/#310 were
+        // reproducible for gemini too.
+        return createGoogleGenerativeAI({
+          apiKey,
+          fetch: this.aiProviderFetch,
+        })(chatModel);
       case 'ollama':
-        // Ollama needs no API key.
-        return createOllama({ baseURL: baseUrl })(chatModel);
+        // Ollama needs no API key. Same transport hardening as above (#140/#175/#310).
+        return createOllama({
+          baseURL: baseUrl,
+          fetch: this.aiProviderFetch,
+        })(chatModel);
       default:
         throw new AiNotConfiguredException();
     }

apps/server/src/integrations/ai/ai.types.ts

@@ -105,6 +105,10 @@ export interface ResolvedAiConfig extends Partial<AiProviderSettings> {
   // Max context window in tokens; surfaced to the chat header badge as the
   // "current / max" denominator. 0/unset = no limit.
   chatContextWindow?: number;
+  // RAW stored context window (::text), BEFORE parsePositiveInt collapses `0` and
+  // unset to `undefined`. The #490 replay budgeter needs the raw value to honor an
+  // explicit `0` off-switch distinctly from "unset -> flat default".
+  chatContextWindowRaw?: string | number;
   // Cheap model id for the public-share assistant; reuses the chat creds.
   publicShareChatModel?: string;
   // Agent-role id whose persona the public-share assistant adopts (empty/unset
@@ -149,4 +153,14 @@ export interface MaskedAiSettings {
   // True while a full workspace reindex is actively running (the counts above
   // then reflect the live run progress rather than the steady-state DB count).
   reindexing?: boolean;
+  // Identity of the ACTIVE reindex run (present only while `reindexing`). The
+  // client keys its poll on `runId`: a changed value means a NEW run (reset the
+  // per-run poll state it latched), the same value means the run it is already
+  // watching — removing the "same run or a fresh one?" ambiguity a stale
+  // pre-reindex snapshot otherwise causes. Absent/empty degrades gracefully.
+  runId?: string;
+  // Epoch-ms the active run started (present only while `reindexing`). Paired
+  // with `runId` so a run that restarts with the same (recycled) id is still
+  // seen as new.
+  reindexStartedAt?: number;
 }

apps/server/src/integrations/ai/embedding-reindex-progress.service.spec.ts

@@ -48,19 +48,38 @@ describe('EmbeddingReindexProgressService', () => {
   }
 
   describe('get', () => {
-    it('maps a valid hash to a ReindexProgress object', async () => {
+    it('maps a valid hash to a ReindexProgress object (incl. the run identity)', async () => {
       const { redis, hgetall } = makeRedis();
-      hgetall.mockResolvedValue({ total: '478', done: '120', startedAt: '1000' });
+      hgetall.mockResolvedValue({
+        total: '478',
+        done: '120',
+        startedAt: '1000',
+        runId: 'run-xyz',
+      });
       const service = makeService(redis);
 
       await expect(service.get(WORKSPACE_ID)).resolves.toEqual({
         total: 478,
         done: 120,
         startedAt: 1000,
+        runId: 'run-xyz',
       });
       expect(hgetall).toHaveBeenCalledWith(KEY);
     });
 
+    it('degrades a missing runId to an empty string (legacy/partial record)', async () => {
+      const { redis, hgetall } = makeRedis();
+      // A record written before runId existed: get() must still succeed and
+      // report runId='' so the client treats it as "no identity", never breaks.
+      hgetall.mockResolvedValue({ total: '10', done: '3', startedAt: '5' });
+      await expect(makeService(redis).get(WORKSPACE_ID)).resolves.toEqual({
+        total: 10,
+        done: 3,
+        startedAt: 5,
+        runId: '',
+      });
+    });
+
     it('returns null for an empty hash (no record)', async () => {
       const { redis, hgetall } = makeRedis();
       hgetall.mockResolvedValue({});
@@ -87,11 +106,17 @@ describe('EmbeddingReindexProgressService', () => {
 
     it('coerces a non-finite startedAt to 0', async () => {
       const { redis, hgetall } = makeRedis();
-      hgetall.mockResolvedValue({ total: '10', done: '2', startedAt: 'nope' });
+      hgetall.mockResolvedValue({
+        total: '10',
+        done: '2',
+        startedAt: 'nope',
+        runId: 'run-1',
+      });
       await expect(makeService(redis).get(WORKSPACE_ID)).resolves.toEqual({
         total: 10,
         done: 2,
         startedAt: 0,
+        runId: 'run-1',
       });
     });
 
@@ -115,6 +140,21 @@ describe('EmbeddingReindexProgressService', () => {
       expect(multiObj.exec).toHaveBeenCalledTimes(1);
     });
 
+    it('mints a fresh non-empty runId into the record on each start', async () => {
+      const { redis, multiObj } = makeRedis();
+      const service = makeService(redis);
+      await service.start(WORKSPACE_ID, 1);
+      await service.start(WORKSPACE_ID, 1);
+
+      const firstRunId = multiObj.hset.mock.calls[0][1].runId;
+      const secondRunId = multiObj.hset.mock.calls[1][1].runId;
+      expect(typeof firstRunId).toBe('string');
+      expect(firstRunId).not.toBe('');
+      // Each run gets its OWN identity so the client can tell a re-trigger apart
+      // from the run it is already watching.
+      expect(secondRunId).not.toBe(firstRunId);
+    });
+
     it('defaults the expire TTL to the full 1h record TTL', async () => {
       const { redis, multiObj } = makeRedis();
       await makeService(redis).start(WORKSPACE_ID, 478);

apps/server/src/integrations/ai/embedding-reindex-progress.service.ts

@@ -1,17 +1,28 @@
 import { Injectable, Logger } from '@nestjs/common';
 import { RedisService } from '@nestjs-labs/nestjs-ioredis';
+import { randomUUID } from 'node:crypto';
 import type { Redis } from 'ioredis';
 
 /**
  * Live progress of an in-flight workspace embeddings reindex run.
  * `total` is the number of pages the run will process, `done` how many it has
  * already processed (success OR handled failure), `startedAt` the epoch-ms the
- * record was created.
+ * record was created, and `runId` a per-run identity minted at `start()`.
+ *
+ * `runId` gives each reindex run a stable identity so a poller can tell "same
+ * run I've been watching" from "a NEW run started" WITHOUT guessing from the
+ * progress counters (the ambiguity that a stale pre-reindex snapshot vs a fresh
+ * run otherwise causes — the bug class fixed twice under #262). It is best-
+ * effort like the rest of this record: a record written before this field
+ * existed (or a Redis hiccup) yields an empty `runId`, which the client must
+ * treat as "no identity available" and degrade to its prior behaviour, never
+ * break.
  */
 export interface ReindexProgress {
   total: number;
   done: number;
   startedAt: number;
+  runId: string;
 }
 
 /** Redis key namespace for the per-workspace reindex-progress record. */
@@ -86,12 +97,18 @@ export class EmbeddingReindexProgressService {
   ): Promise<void> {
     const key = this.key(workspaceId);
     try {
+      // A fresh identity per run so the client poll can key on it: a changed
+      // runId means a genuinely NEW run (reset any latched per-run poll state),
+      // the same runId means the run the client is already watching. Best-effort
+      // like the counters — never surfaced to the user, only used to disambiguate.
+      const runId = randomUUID();
       await this.redis
         .multi()
         .hset(key, {
           total: String(total),
           done: '0',
           startedAt: String(Date.now()),
+          runId,
         })
         .expire(key, ttlSeconds)
         .exec();
@@ -150,7 +167,15 @@ export class EmbeddingReindexProgressService {
       const done = Number(data.done);
       const startedAt = Number(data.startedAt);
       if (!Number.isFinite(total) || !Number.isFinite(done)) return null;
-      return { total, done, startedAt: Number.isFinite(startedAt) ? startedAt : 0 };
+      // `runId` degrades gracefully: a pre-existing record (written before this
+      // field) or a stripped value reads as '' — the client treats that as "no
+      // identity" and keeps its prior behaviour rather than breaking the poll.
+      return {
+        total,
+        done,
+        startedAt: Number.isFinite(startedAt) ? startedAt : 0,
+        runId: typeof data.runId === 'string' ? data.runId : '',
+      };
     } catch (err) {
       this.logger.warn(
         `reindex-progress read failed for workspace ${workspaceId}; ` +

apps/server/src/integrations/audit/audit.module.ts

@@ -1,14 +1,19 @@
 import { Global, Module } from '@nestjs/common';
-import { AUDIT_SERVICE, NoopAuditService } from './audit.service';
+import { AUDIT_SERVICE } from './audit.service';
+import { DatabaseAuditService } from './database-audit.service';
 
+// #496: bind the audit token to a real DB-backed trail (was NoopAuditService,
+// which silently dropped every event). Kysely (@Global DatabaseModule) and
+// ClsService (@Global ClsModule) are both globally available, so this module
+// needs no extra imports.
 @Global()
 @Module({
   providers: [
     {
       provide: AUDIT_SERVICE,
-      useClass: NoopAuditService,
+      useClass: DatabaseAuditService,
     },
   ],
   exports: [AUDIT_SERVICE],
 })
-export class NoopAuditModule {}
+export class AuditModule {}

apps/server/src/integrations/audit/database-audit.service.spec.ts

@@ -0,0 +1,198 @@
+import { Logger } from '@nestjs/common';
+import { DatabaseAuditService } from './database-audit.service';
+import { AUDIT_CONTEXT_KEY } from '../../common/middlewares/audit-context.middleware';
+import { AuditEvent, AuditResource } from '../../common/events/audit-events';
+
+/**
+ * Observable-property coverage for the DB-backed audit trail (#496): every
+ * assertion pins what actually reaches the `audit` table (or that nothing does),
+ * driven through a chainable Kysely mock that captures the inserted rows.
+ */
+describe('DatabaseAuditService', () => {
+  function makeService(clsContext: any) {
+    const inserted: any[] = [];
+    const updated: any[] = [];
+    let failNextInsert = false;
+
+    const db: any = {
+      insertInto: jest.fn(() => ({
+        values: jest.fn((rows: any) => ({
+          execute: jest.fn(async () => {
+            if (failNextInsert) {
+              failNextInsert = false;
+              throw new Error('boom');
+            }
+            inserted.push(rows);
+          }),
+        })),
+      })),
+      updateTable: jest.fn(() => ({
+        set: jest.fn((patch: any) => ({
+          where: jest.fn(() => ({
+            execute: jest.fn(async () => {
+              updated.push(patch);
+            }),
+          })),
+        })),
+      })),
+    };
+
+    const store: Record<string, any> = { [AUDIT_CONTEXT_KEY]: clsContext };
+    const cls: any = {
+      get: jest.fn((key: string) => store[key]),
+      set: jest.fn((key: string, val: any) => {
+        store[key] = val;
+      }),
+    };
+
+    const service = new DatabaseAuditService(db, cls);
+    return {
+      service,
+      inserted,
+      updated,
+      cls,
+      store,
+      failInsert: () => {
+        failNextInsert = true;
+      },
+    };
+  }
+
+  const applyPayload = () => ({
+    event: AuditEvent.COMMENT_SUGGESTION_APPLIED,
+    resourceType: AuditResource.COMMENT,
+    resourceId: 'c-1',
+    spaceId: 'space-1',
+    metadata: { pageId: 'p-1', suggestedText: 'new', decidedBy: 'u-2' },
+  });
+
+  const ctx = (over?: any) => ({
+    workspaceId: 'ws-1',
+    actorId: 'u-2',
+    actorType: 'user',
+    ipAddress: '10.0.0.1',
+    userAgent: 'jest',
+    ...over,
+  });
+
+  it('log() persists a row with the CLS context merged onto the payload', async () => {
+    const { service, inserted } = makeService(ctx());
+    service.log(applyPayload());
+    // log() is fire-and-forget; flush the microtask queue.
+    await Promise.resolve();
+    await Promise.resolve();
+
+    expect(inserted).toHaveLength(1);
+    expect(inserted[0]).toMatchObject({
+      workspaceId: 'ws-1',
+      actorId: 'u-2',
+      actorType: 'user',
+      event: AuditEvent.COMMENT_SUGGESTION_APPLIED,
+      resourceType: AuditResource.COMMENT,
+      resourceId: 'c-1',
+      spaceId: 'space-1',
+      ipAddress: '10.0.0.1',
+      metadata: { pageId: 'p-1', suggestedText: 'new', decidedBy: 'u-2' },
+    });
+  });
+
+  it('log() is a no-op when there is no workspace in scope', async () => {
+    const { service, inserted } = makeService(ctx({ workspaceId: null }));
+    service.log(applyPayload());
+    await Promise.resolve();
+    expect(inserted).toHaveLength(0);
+  });
+
+  it('log() drops EXCLUDED_AUDIT_EVENTS (e.g. comment.created)', async () => {
+    const { service, inserted } = makeService(ctx());
+    service.log({
+      event: AuditEvent.COMMENT_CREATED,
+      resourceType: AuditResource.COMMENT,
+      resourceId: 'c-1',
+      spaceId: 'space-1',
+    });
+    await Promise.resolve();
+    await Promise.resolve();
+    expect(inserted).toHaveLength(0);
+  });
+
+  it('a failed insert is swallowed with a warn and floats no rejection (audit is a side-record)', async () => {
+    // This pins the load-bearing swallow inside persist(). Because log() is
+    // fire-and-forget (`void this.persist(...)`), it always returns synchronously
+    // without throwing — so `not.toThrow()` alone would stay green even if the
+    // try/catch were removed. We instead observe the two effects the catch is
+    // responsible for: a warn IS emitted, and NO unhandled rejection floats.
+    // Removing persist()'s try/catch reddens both assertions (warn count 0 + a
+    // captured rejection).
+    const warnSpy = jest
+      .spyOn(Logger.prototype, 'warn')
+      .mockImplementation(() => undefined as any);
+    const rejections: unknown[] = [];
+    const onRejection = (err: unknown) => rejections.push(err);
+    process.on('unhandledRejection', onRejection);
+    try {
+      const { service, failInsert } = makeService(ctx());
+      failInsert();
+      expect(() => service.log(applyPayload())).not.toThrow();
+      // Flush microtasks so the rejected insert settles, then give any floated
+      // rejection a macrotask tick to be reported by the runtime.
+      await Promise.resolve();
+      await Promise.resolve();
+      await new Promise((r) => setImmediate(r));
+
+      expect(warnSpy).toHaveBeenCalledTimes(1);
+      expect(String(warnSpy.mock.calls[0][0])).toContain(
+        'Failed to persist audit event',
+      );
+      expect(rejections).toHaveLength(0);
+    } finally {
+      process.off('unhandledRejection', onRejection);
+      warnSpy.mockRestore();
+    }
+  });
+
+  it('logWithContext() persists with an explicit (non-request) context', async () => {
+    const { service, inserted } = makeService(undefined);
+    service.logWithContext(applyPayload(), ctx({ actorType: 'system' }) as any);
+    await Promise.resolve();
+    await Promise.resolve();
+    expect(inserted).toHaveLength(1);
+    expect(inserted[0].actorType).toBe('system');
+    expect(inserted[0].workspaceId).toBe('ws-1');
+  });
+
+  it('logBatchWithContext() inserts only non-excluded events', async () => {
+    const { service, inserted } = makeService(undefined);
+    service.logBatchWithContext(
+      [
+        applyPayload(),
+        {
+          event: AuditEvent.COMMENT_CREATED,
+          resourceType: AuditResource.COMMENT,
+          resourceId: 'c-2',
+        },
+      ],
+      ctx() as any,
+    );
+    await Promise.resolve();
+    await Promise.resolve();
+    // Batch is a single insert call carrying only the applied event.
+    expect(inserted).toHaveLength(1);
+    expect(inserted[0]).toHaveLength(1);
+    expect(inserted[0][0].event).toBe(AuditEvent.COMMENT_SUGGESTION_APPLIED);
+  });
+
+  it('setActorId / setActorType mutate the ambient CLS context', () => {
+    const { service, store } = makeService(ctx({ actorId: null }));
+    service.setActorId('u-9');
+    service.setActorType('api_key');
+    expect(store[AUDIT_CONTEXT_KEY].actorId).toBe('u-9');
+    expect(store[AUDIT_CONTEXT_KEY].actorType).toBe('api_key');
+  });
+
+  it('updateRetention() writes the workspace retention window', async () => {
+    const { service, updated } = makeService(ctx());
+    await service.updateRetention('ws-1', 30);
+    expect(updated).toEqual([{ auditRetentionDays: 30 }]);
+  });
+});

apps/server/src/integrations/audit/database-audit.service.ts

@@ -0,0 +1,160 @@
+import { Injectable, Logger } from '@nestjs/common';
+import { InjectKysely } from 'nestjs-kysely';
+import { ClsService } from 'nestjs-cls';
+import { KyselyDB } from '@docmost/db/types/kysely.types';
+import {
+  AuditContext,
+  AUDIT_CONTEXT_KEY,
+} from '../../common/middlewares/audit-context.middleware';
+import {
+  AuditLogPayload,
+  ActorType,
+  EXCLUDED_AUDIT_EVENTS,
+} from '../../common/events/audit-events';
+import { AuditLogContext, IAuditService } from './audit.service';
+
+/**
+ * Minimal DB-backed audit trail (#496). Replaces NoopAuditService so that
+ * decision-bearing events — notably comment.suggestion_applied /
+ * comment.suggestion_dismissed, whose subject comment is HARD-DELETED on the
+ * childless path — leave a durable record of who decided what. Without this the
+ * events were emitted (comment.service / *.controller) but swallowed, so an
+ * applied/dismissed suggestion was unrecoverable once the row was gone.
+ *
+ * Rows land in the pre-existing `audit` table (migration 20260228T223532). The
+ * per-request actor/workspace/ip come from the CLS AuditContext populated by
+ * AuditContextMiddleware + AuditActorInterceptor; callers that run OUTSIDE a
+ * request (queue workers, imports) pass an explicit context via
+ * logWithContext / logBatchWithContext.
+ *
+ * Audit is a side-record: a write failure MUST NOT break the originating
+ * request, so every persistence path swallows its error with a warn. Events in
+ * EXCLUDED_AUDIT_EVENTS (high-volume/low-signal) are dropped.
+ */
+@Injectable()
+export class DatabaseAuditService implements IAuditService {
+  private readonly logger = new Logger(DatabaseAuditService.name);
+
+  constructor(
+    @InjectKysely() private readonly db: KyselyDB,
+    private readonly cls: ClsService,
+  ) {}
+
+  /**
+   * Persist a single event using the ambient request-scoped AuditContext. A
+   * no-op when there is no workspace in scope (the table's workspace_id is NOT
+   * NULL) or the event is excluded. Fire-and-forget: the returned promise is not
+   * awaited by hot callers, and its rejection is swallowed here.
+   */
+  log(payload: AuditLogPayload): void {
+    const context = this.cls?.get<AuditContext>(AUDIT_CONTEXT_KEY);
+    if (!context?.workspaceId) {
+      // No workspace in scope — nothing we can attribute the row to. This is
+      // expected for events emitted outside an HTTP request; those callers must
+      // use logWithContext instead.
+      return;
+    }
+    void this.persist(payload, {
+      workspaceId: context.workspaceId,
+      actorId: context.actorId ?? undefined,
+      actorType: context.actorType,
+      ipAddress: context.ipAddress ?? undefined,
+      userAgent: context.userAgent ?? undefined,
+    });
+  }
+
+  /** Persist a single event with an explicit (non-request) context. */
+  logWithContext(payload: AuditLogPayload, context: AuditLogContext): void {
+    if (!context?.workspaceId) return;
+    void this.persist(payload, context);
+  }
+
+  /** Persist a batch of events sharing one explicit context (imports). */
+  logBatchWithContext(
+    payloads: AuditLogPayload[],
+    context: AuditLogContext,
+  ): void {
+    if (!context?.workspaceId || payloads.length === 0) return;
+    const rows = payloads
+      .filter((p) => !EXCLUDED_AUDIT_EVENTS.has(p.event))
+      .map((p) => this.toRow(p, context));
+    if (rows.length === 0) return;
+    this.db
+      .insertInto('audit')
+      .values(rows)
+      .execute()
+      .catch((err: any) =>
+        this.logger.warn(`Failed to persist ${rows.length} audit events: ${err?.message}`),
+      );
+  }
+
+  /** Update the ambient request actor (e.g. after login resolves the user). */
+  setActorId(actorId: string): void {
+    const context = this.cls?.get<AuditContext>(AUDIT_CONTEXT_KEY);
+    if (context) {
+      context.actorId = actorId;
+      this.cls.set(AUDIT_CONTEXT_KEY, context);
+    }
+  }
+
+  /** Update the ambient request actor type (user | system | api_key). */
+  setActorType(actorType: ActorType): void {
+    const context = this.cls?.get<AuditContext>(AUDIT_CONTEXT_KEY);
+    if (context) {
+      context.actorType = actorType;
+      this.cls.set(AUDIT_CONTEXT_KEY, context);
+    }
+  }
+
+  /** Persist a workspace's audit-log retention window (days). */
+  async updateRetention(
+    workspaceId: string,
+    retentionDays: number,
+  ): Promise<void> {
+    try {
+      await this.db
+        .updateTable('workspaces')
+        .set({ auditRetentionDays: retentionDays })
+        .where('id', '=', workspaceId)
+        .execute();
+    } catch (err: any) {
+      this.logger.warn(
+        `Failed to update audit retention for workspace ${workspaceId}: ${err?.message}`,
+      );
+    }
+  }
+
+  private async persist(
+    payload: AuditLogPayload,
+    context: AuditLogContext,
+  ): Promise<void> {
+    if (EXCLUDED_AUDIT_EVENTS.has(payload.event)) return;
+    try {
+      await this.db
+        .insertInto('audit')
+        .values(this.toRow(payload, context))
+        .execute();
+    } catch (err: any) {
+      // Audit is a side-record; never let a failed write surface to the caller.
+      this.logger.warn(
+        `Failed to persist audit event ${payload.event}: ${err?.message}`,
+      );
+    }
+  }
+
+  private toRow(payload: AuditLogPayload, context: AuditLogContext) {
+    return {
+      workspaceId: context.workspaceId,
+      actorId: context.actorId ?? null,
+      actorType: context.actorType ?? 'user',
+      event: payload.event,
+      resourceType: payload.resourceType,
+      resourceId: payload.resourceId ?? null,
+      spaceId: payload.spaceId ?? null,
+      // jsonb columns: node-postgres serializes plain objects to JSON.
+      changes: payload.changes ? (payload.changes as any) : null,
+      metadata: payload.metadata ? (payload.metadata as any) : null,
+      ipAddress: context.ipAddress ?? null,
+    };
+  }
+}

apps/server/src/integrations/import/services/file-import-task.service.ts

@@ -22,10 +22,12 @@ import { v7 } from 'uuid';
 import { generateJitteredKeyBetween } from 'fractional-indexing-jittered';
 import { FileTask, InsertablePage } from '@docmost/db/types/entity.types';
 import { canonicalizeFootnotes } from '@docmost/editor-ext';
-import { markdownToProseMirror } from '@docmost/prosemirror-markdown';
+import {
+  markdownToProseMirror,
+  normalizeForeignMarkdown,
+} from '@docmost/prosemirror-markdown';
 import { getProsemirrorContent } from '../../../common/helpers/prosemirror/utils';
 import { formatImportHtml } from '../utils/import-formatter';
-import { normalizeForeignMarkdown } from '../utils/foreign-markdown';
 import {
   buildAttachmentCandidates,
   collectMarkdownAndHtmlFiles,

apps/server/src/integrations/import/services/import.service.ts

@@ -18,8 +18,10 @@ import { generateJitteredKeyBetween } from 'fractional-indexing-jittered';
 import { TiptapTransformer } from '@hocuspocus/transformer';
 import * as Y from 'yjs';
 import { canonicalizeFootnotes } from '@docmost/editor-ext';
-import { markdownToProseMirror } from '@docmost/prosemirror-markdown';
-import { normalizeForeignMarkdown } from '../utils/foreign-markdown';
+import {
+  markdownToProseMirror,
+  normalizeForeignMarkdown,
+} from '@docmost/prosemirror-markdown';
 import {
   FileTaskStatus,
   FileTaskType,

apps/server/src/integrations/import/utils/foreign-markdown.spec.ts

@@ -1,218 +0,0 @@
-import {
-  convertProseMirrorToMarkdown,
-  markdownToProseMirror,
-} from '@docmost/prosemirror-markdown';
-import { normalizeForeignMarkdown } from './foreign-markdown';
-
-/**
- * STEP 2 goldens for issue #345: the foreign-markdown normalizer that runs at the
- * import boundary BEFORE the strict canonical parser (`markdownToProseMirror`).
- *
- * Two layers:
- *  1. PURE string→string cases pinning the normalizer's own behavior (GFM
- *     reference footnotes → inline `^[…]`).
- *  2. END-TO-END acceptance: for a foreign corpus, `normalizeForeignMarkdown`
- *     then `markdownToProseMirror` then `convertProseMirrorToMarkdown` must leave
- *     NO literal `[^id]` / `:::` garbage in the document and must re-export in the
- *     canonical forms.
- */
-
-describe('normalizeForeignMarkdown — GFM reference footnotes', () => {
-  it('inlines a single-line reference footnote and drops its definition', () => {
-    const out = normalizeForeignMarkdown(
-      'A note[^1] here.\n\n[^1]: The definition.',
-    );
-    expect(out).toBe('A note^[The definition.] here.\n');
-  });
-
-  it('inlines every reference to a reused id (downstream dedups)', () => {
-    const out = normalizeForeignMarkdown(
-      'X[^a] and Y[^a].\n\n[^a]: shared.',
-    );
-    expect(out).toBe('X^[shared.] and Y^[shared.].\n');
-  });
-
-  it('joins indented continuation lines of a definition with a space', () => {
-    const out = normalizeForeignMarkdown(
-      'See[^n].\n\n[^n]: line one\n    line two',
-    );
-    expect(out).toBe('See^[line one line two].\n');
-  });
-
-  it('never rewrites a reference inside a fenced code block', () => {
-    const out = normalizeForeignMarkdown(
-      '```\ncode[^1] here\n```\n\n[^1]: def.',
-    );
-    expect(out).toContain('code[^1] here');
-    // The (now orphaned) definition line is still removed.
-    expect(out).not.toContain('[^1]: def.');
-  });
-
-  it('never rewrites a reference inside an INLINE-code span (backticks)', () => {
-    // The `[^1]` inside backticks is literal code and must survive verbatim;
-    // the one outside is rewritten. (Bug #1: only fenced blocks were protected.)
-    const out = normalizeForeignMarkdown(
-      'Use `arr[^1]` in code but note[^1] in prose.\n\n[^1]: def.',
-    );
-    expect(out).toBe('Use `arr[^1]` in code but note^[def.] in prose.\n');
-  });
-
-  it('escapes brackets in a body so an unbalanced ] cannot truncate the footnote', () => {
-    // A foreign definition body with a stray `]` would, unescaped, close the
-    // canonical `^[...]` early and leak the tail as text (bug #2). The body's
-    // brackets are backslash-escaped so the footnote stays whole.
-    const out = normalizeForeignMarkdown(
-      'Ref[^1] here.\n\n[^1]: see item ] and [more] later',
-    );
-    expect(out).toBe('Ref^[see item \\] and \\[more\\] later] here.\n');
-    // The tokenizer must see exactly one unescaped closing bracket (our own).
-    expect(out.match(/(?<!\\)\]/g)).toHaveLength(1);
-  });
-
-  it('leaves a reference with no matching definition literal (no body to inline)', () => {
-    const out = normalizeForeignMarkdown('Dangling[^x] ref.');
-    expect(out).toBe('Dangling[^x] ref.');
-  });
-
-  it('returns the input unchanged when there are no reference footnotes', () => {
-    const md = '# Title\n\nJust text with `inline code` and a [link](/x).';
-    expect(normalizeForeignMarkdown(md)).toBe(md);
-  });
-
-  it('does NOT touch callout surfaces — the canonical parser handles them', () => {
-    const callouts = ':::info\nHi\n:::\n\n> [!warning]\n> Careful';
-    expect(normalizeForeignMarkdown(callouts)).toBe(callouts);
-  });
-
-  it('strips a leading YAML front-matter block (Obsidian/Hugo/git-sync files)', () => {
-    const out = normalizeForeignMarkdown(
-      '---\ntitle: My Page\ntags: [a, b]\n---\n\n# Heading\n\nBody.',
-    );
-    expect(out).toBe('# Heading\n\nBody.');
-    // The front-matter must not leak into the body as a setext heading.
-    expect(out).not.toContain('title: My Page');
-    expect(out).not.toContain('---');
-  });
-
-  it('does not strip a horizontal rule that is not leading front-matter', () => {
-    const md = 'Intro paragraph.\n\n---\n\nAfter the rule.';
-    expect(normalizeForeignMarkdown(md)).toBe(md);
-  });
-
-  it('is linear on a document with thousands of definitions (no quadratic blowup)', () => {
-    // F2(a): the pass-2 rewrite must be O(text), not O(text × defs). Build a
-    // pathological doc (many defs + many plain text lines) and assert it
-    // completes well under a second — a quadratic implementation took ~14s.
-    const N = 4000;
-    const refs = Array.from({ length: N }, (_, i) => `line ${i} plain text`).join('\n');
-    const defs = Array.from({ length: N }, (_, i) => `[^n${i}]: def ${i}`).join('\n');
-    const doc = `start[^n0] and[^n${N - 1}] end\n\n${refs}\n\n${defs}`;
-    const t0 = Date.now();
-    const out = normalizeForeignMarkdown(doc);
-    const elapsed = Date.now() - t0;
-    expect(elapsed).toBeLessThan(2000);
-    // Sanity: the two real references were still inlined.
-    expect(out).toContain('^[def 0]');
-    expect(out).toContain(`^[def ${N - 1}]`);
-  });
-
-  it('is bounded on a long unclosed backtick run (no inline-split ReDoS)', () => {
-    // F2(b): a huge unterminated backtick run must not cause quadratic
-    // backtracking in the inline-code split. Oversized lines skip the split
-    // entirely (left untouched), so this returns promptly.
-    const line = 'x' + '`'.repeat(200000);
-    const doc = `${line}\n\n[^1]: def`;
-    const t0 = Date.now();
-    normalizeForeignMarkdown(doc);
-    expect(Date.now() - t0).toBeLessThan(2000);
-  });
-
-  it('does not crash or slow down on thousands of prefix-chain definition ids', () => {
-    // F7: the rewrite must use a FIXED generic scanner, not an alternation built
-    // from the ids. A `(a|aa|aaa|…)` alternation over prefix-chain ids blows the
-    // V8 regex compiler (FATAL RegExpCompiler Allocation failed — uncatchable,
-    // kills the process). A fixed scanner has no id-dependent compilation cost.
-    const N = 4000;
-    const ids = Array.from({ length: N }, (_, i) => 'a'.repeat(i + 1));
-    const defs = ids.map((id) => `[^${id}]: body ${id.length}`).join('\n');
-    const doc = `ref[^${ids[0]}] and[^${ids[N - 1]}] end\n\n${defs}`;
-    const t0 = Date.now();
-    const out = normalizeForeignMarkdown(doc);
-    expect(Date.now() - t0).toBeLessThan(2000);
-    // Prefix disambiguation is correct: [^a] and [^aaaa...] inline their OWN body.
-    expect(out).toContain('^[body 1]');
-    expect(out).toContain(`^[body ${N}]`);
-  });
-
-  it('strips a CRLF (Windows) front-matter block, not just LF', () => {
-    // F9: the line-anchored regex needs LF after the opening `---`, so a Windows
-    // file (`---\r\n…`) would slip past the strip and leak the front-matter into
-    // the body. normalizeForeignMarkdown normalizes CRLF -> LF first.
-    const out = normalizeForeignMarkdown(
-      '---\r\ntitle: Foo\r\ntags: [a]\r\n---\r\n\r\n# Heading\r\n\r\nBody.',
-    );
-    expect(out).toBe('# Heading\n\nBody.');
-    expect(out).not.toContain('title: Foo');
-    expect(out).not.toContain('---');
-  });
-
-  it('strips front-matter whose value contains a triple-dash (line-anchored)', () => {
-    // F8: the block must close only on a `\n---` LINE, not the first inline
-    // `---`. A value like `title: Q1 --- Q2` must not truncate the front-matter
-    // and leak the rest (author/closing ---) into the body.
-    const out = normalizeForeignMarkdown(
-      '---\ntitle: Q1 --- Q2 results\nauthor: bob\n---\n\nReal body.',
-    );
-    expect(out).toBe('Real body.');
-    expect(out).not.toContain('author: bob');
-    expect(out).not.toContain('Q2 results');
-  });
-});
-
-describe('foreign markdown import acceptance (normalizer + canonical parser)', () => {
-  const FOREIGN = [
-    '# Doc',
-    '',
-    'Body refs [^c] and [^a] and [^b] and again [^a].',
-    '',
-    ':::info',
-    'A legacy callout.',
-    ':::',
-    '',
-    '| h1 | h2 |',
-    '| --- | --- |',
-    '| 1 | 2 |',
-    '',
-    '[^a]: note A',
-    '[^b]: note B',
-    '[^c]: note C',
-    '[^z]: orphan note',
-  ].join('\n');
-
-  it('leaves no literal [^id] or ::: in the imported doc and re-exports canonically', async () => {
-    const normalized = normalizeForeignMarkdown(FOREIGN);
-    const doc = await markdownToProseMirror(normalized);
-    const reexport = convertProseMirrorToMarkdown(doc);
-
-    // No foreign garbage leaks into the document.
-    expect(reexport).not.toMatch(/\[\^/); // no reference footnote refs/defs
-    expect(reexport).not.toContain(':::'); // no legacy callout fences
-
-    // Canonical forms are present.
-    expect(reexport).toContain('^[note C]');
-    expect(reexport).toContain('> [!info]');
-    expect(reexport).toContain('| h1 | h2 |');
-
-    // Footnotes: ordered by first reference (C, A, B), reused [^a] deduped to one,
-    // orphan [^z] dropped (it had no reference after normalization).
-    const list = doc.content.find((n: any) => n.type === 'footnotesList');
-    const bodies = list.content.map(
-      (d: any) => d.content[0].content[0].text,
-    );
-    expect(bodies).toEqual(['note C', 'note A', 'note B']);
-    expect(bodies).not.toContain('orphan note');
-    expect(
-      doc.content.filter((n: any) => n.type === 'footnotesList'),
-    ).toHaveLength(1);
-  });
-});

apps/server/src/integrations/import/utils/foreign-markdown.ts

@@ -1,266 +0,0 @@
-/**
- * Foreign-markdown normalizer — an input-liberal / output-canonical adapter that
- * runs at the IMPORT boundary, BEFORE the canonical parser
- * (`markdownToProseMirror` from `@docmost/prosemirror-markdown`).
- *
- * The canonical parser is deliberately STRICT: it only understands Docmost's
- * canonical markdown surface (Obsidian-style `> [!type]` callouts, Pandoc/Obsidian
- * inline footnotes `^[body]`, lossless `![alt](src) <!--img {...}-->` images, …).
- * Import, however, ingests FOREIGN files (GitHub/GFM, Notion, old Docmost
- * exports). Those use surfaces the canonical parser does not accept, most notably
- * GitHub-flavoured *reference* footnotes:
- *
- *     Text with a note[^1] and another[^long].
- *
- *     [^1]: The first definition.
- *     [^long]: A second one.
- *
- * Left untouched, the parser does NOT recognise `[^id]` (it only parses `^[body]`),
- * so the reference leaks as literal text — and worse, the trailing `[^id]: def`
- * line is a valid CommonMark *link-reference definition*, so `[^id]` is silently
- * rendered as a bogus link. This normalizer rewrites reference footnotes into the
- * canonical inline form so the parser materialises real footnote nodes.
- *
- * This is a TEXT pre-pass, NOT a second parser fork: it does not re-implement any
- * converter logic. Callout surfaces (`:::type` and `> [!type]`) are intentionally
- * NOT touched here — the canonical parser already accepts BOTH natively (its
- * `preprocessCallouts` pass), so normalizing them would be redundant and would
- * only risk degrading the parser's nesting/code-fence-aware handling.
- */
-
-/** Matches a fenced code block delimiter (``` or ~~~), capturing the marker run. */
-const CODE_FENCE_RE = /^(\s*)(`{3,}|~{3,})/;
-
-/**
- * Matches a GFM footnote DEFINITION line: `[^id]: body`. The id is any run of
- * non-`]` characters; the body is the remainder of the line (possibly empty).
- */
-const FOOTNOTE_DEF_RE = /^\[\^([^\]]+)\]:[ \t]?(.*)$/;
-
-/** True when a line is a code-fence delimiter that toggles fenced-code state. */
-function fenceMarker(line: string): string | null {
-  const m = line.match(CODE_FENCE_RE);
-  return m ? m[2] : null;
-}
-
-/** True when a line is indented (leading space/tab) and not blank — a continuation. */
-function isIndentedContinuation(line: string): boolean {
-  return /^[ \t]+\S/.test(line);
-}
-
-function escapeRegExp(value: string): string {
-  return value.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
-}
-
-/**
- * Backslash-escape any square bracket in a footnote body before it is wrapped in
- * `^[...]`. The canonical inline-footnote tokenizer scans the body with bracket
- * balancing and closes on the first UNMATCHED `]`, so an unbalanced bracket in a
- * foreign definition (e.g. `[^1]: see item ] later`) would otherwise truncate the
- * footnote and leak the tail as literal text. Escaping every `[`/`]` makes the
- * body an inert run of characters — the tokenizer then closes only on our own
- * closing `]`. (A balanced `[link](url)` inside a body still round-trips because
- * the escaped form renders the literal brackets, which is the safe reading for a
- * footnote body; the alternative — brittle balance tracking — risks worse.)
- */
-function escapeFootnoteBody(body: string): string {
-  return body.replace(/[[\]]/g, '\\$&');
-}
-
-/**
- * Rewrite every `[^id]` reference on a line to its `^[body]` form, but ONLY in the
- * text OUTSIDE inline-code spans. A `[^id]` inside backticks is literal code
- * content and must be preserved verbatim (a footnote ref never lives inside code).
- * We split the line on inline-code spans (paired backtick runs) and rewrite only
- * the non-code segments.
- */
-// Above this length a single line is not split into inline-code spans (see
-// below). A genuine markdown line carrying a footnote reference is never tens of
-// KB; the cap only bypasses the inline-code protection for pathological lines.
-const INLINE_SPLIT_MAX_LINE = 8192;
-
-function rewriteRefsOutsideInlineCode(
-  line: string,
-  replace: (text: string) => string,
-): string {
-  // The inline-code split alternation `(`+)(?:(?!\1)[\s\S])*\1` backtracks
-  // quadratically on a long UNCLOSED backtick run (its middle can consume the
-  // rest of the line, then fail to find a closing run and retry from each
-  // position). On an untrusted import this is a request-thread ReDoS. A real
-  // footnote line is short, so for an oversized line we skip the inline-code
-  // protection entirely and leave the line UNTOUCHED (rewriting it wholesale
-  // could corrupt a `[^id]` that legitimately lives inside inline code). This is
-  // a conservative bypass: an over-8KB line simply does not get its reference
-  // footnotes inlined — acceptable for a pathological input.
-  if (line.length > INLINE_SPLIT_MAX_LINE) return line;
-
-  // Alternation: an inline-code span (one or more backticks, then anything up to
-  // the SAME run of backticks) OR a run of non-backtick text. Unterminated
-  // backticks fall through as ordinary text (matched by the second branch on the
-  // leftover), so a stray backtick never swallows the rest of the line.
-  const parts = line.match(/(`+)(?:(?!\1)[\s\S])*\1|[^`]+|`+/g);
-  if (!parts) return line;
-  return parts
-    .map((seg) => (seg.startsWith('`') ? seg : replace(seg)))
-    .join('');
-}
-
-/**
- * Convert GFM reference footnotes (`[^id]` + `[^id]: def`) into canonical inline
- * footnotes (`^[def]`).
- *
- * - Definitions are collected first (a leading `[^id]: text` line plus any
- *   immediately-following indented continuation lines, joined with a space) and
- *   removed from the output.
- * - Each in-text reference `[^id]` for which a definition was found is replaced by
- *   `^[def]`. References with no matching definition are left literal (there is no
- *   body to inline; the parser fails them open the same way).
- * - Code is respected on both passes: `[^id]` inside a fenced ``` / ~~~ block is
- *   never rewritten and a `[^id]:` line inside a fence is never a definition; and
- *   on the rewrite pass a `[^id]` inside an INLINE-code span (backticks) is left
- *   literal too.
- * - The inlined body is bracket-escaped so an unbalanced `[`/`]` in a foreign
- *   definition cannot truncate the resulting `^[...]` footnote.
- *
- * Deduplication / reference-ordering / orphan-dropping of the resulting footnotes
- * is handled downstream by the canonical parser (`assembleFootnotes`); this pass
- * only changes the surface syntax.
- */
-function convertReferenceFootnotes(markdown: string): string {
-  const lines = markdown.split('\n');
-
-  // Pass 1: collect definitions and mark their lines for removal.
-  const defs = new Map<string, string>();
-  const dropped = new Array<boolean>(lines.length).fill(false);
-  let inFence = false;
-  let fence = '';
-
-  for (let i = 0; i < lines.length; i++) {
-    const line = lines[i];
-    const marker = fenceMarker(line);
-    if (inFence) {
-      if (marker && marker[0] === fence[0] && marker.length >= fence.length) {
-        inFence = false;
-        fence = '';
-      }
-      continue;
-    }
-    if (marker) {
-      inFence = true;
-      fence = marker;
-      continue;
-    }
-
-    const def = line.match(FOOTNOTE_DEF_RE);
-    if (!def) continue;
-
-    const id = def[1];
-    const body: string[] = [def[2].trim()];
-    dropped[i] = true;
-
-    // Consume immediately-following indented continuation lines (GFM lazy
-    // continuation is not supported by design — keep it simple and predictable).
-    let j = i + 1;
-    while (j < lines.length && isIndentedContinuation(lines[j])) {
-      body.push(lines[j].trim());
-      dropped[j] = true;
-      j++;
-    }
-    i = j - 1;
-
-    // Last definition wins for a duplicated id (matches CommonMark link-ref
-    // semantics closely enough for a foreign-input adapter).
-    defs.set(id, body.filter((s) => s.length > 0).join(' '));
-  }
-
-  if (defs.size === 0) {
-    return markdown;
-  }
-
-  // ONE fixed, generic scanner regex — NOT one built from the definition ids.
-  // It matches ANY `[^id]` shape, and the replacer decides per match via a map
-  // lookup whether that id is a real definition (replace) or not (leave as-is).
-  // This is genuinely O(total text) with no per-document regex compilation.
-  //
-  // Do NOT rebuild this as an alternation over `[...defs.keys()]`: a giant
-  // `(id1|id2|...)` alternation over thousands of ids can blow the V8 regex
-  // compiler's stack — a fatal, UNCATCHABLE "RegExpCompiler Allocation failed"
-  // on prefix-chain ids (`a`, `aa`, `aaa`, ...) that kills the whole process
-  // (worse than the earlier per-def thread-hang). A fixed scanner has no
-  // id-dependent compilation cost and cannot blow up.
-  const refRe = /\[\^([^\]]+)\]/g;
-  const rewriteSegment = (segment: string): string =>
-    segment.replace(refRe, (whole, id: string) => {
-      const body = defs.get(id);
-      // Only real definitions are inlined; an unknown id is left literal (same as
-      // the old per-def loop, which simply never matched it).
-      return body === undefined ? whole : `^[${escapeFootnoteBody(body)}]`;
-    });
-
-  // Pass 2: rewrite in-text references, skipping fenced code and dropped lines.
-  const out: string[] = [];
-  inFence = false;
-  fence = '';
-  for (let i = 0; i < lines.length; i++) {
-    if (dropped[i]) continue;
-    let line = lines[i];
-
-    const marker = fenceMarker(line);
-    if (inFence) {
-      out.push(line);
-      if (marker && marker[0] === fence[0] && marker.length >= fence.length) {
-        inFence = false;
-        fence = '';
-      }
-      continue;
-    }
-    if (marker) {
-      inFence = true;
-      fence = marker;
-      out.push(line);
-      continue;
-    }
-
-    line = rewriteRefsOutsideInlineCode(line, rewriteSegment);
-    out.push(line);
-  }
-
-  return out.join('\n');
-}
-
-/**
- * Strip a single leading YAML front-matter block (`---\n…\n---`). Foreign files
- * from Obsidian / Hugo / Jekyll / Notion — and Docmost's OWN git-sync page files
- * — open with front-matter that the canonical parser does not consume, so
- * without this it leaks into the body (and `title: Foo` above the closing `---`
- * renders as a setext `<h2>` that `extractTitleAndRemoveHeading` can hijack as
- * the page title). It is a no-op for front-matter-free input.
- *
- * LINE-ANCHORED (the same shape the canonical parser uses in
- * prosemirror-markdown/page-file.ts): the block opens only on `---\n` at the
- * very start and closes only on a `\n---` line. The retired editor-ext
- * `markdownToHtml` front-matter strip (removed in #347) closed on the FIRST
- * `---` ANYWHERE (an unanchored close), so a value
- * containing a triple-dash (e.g. `title: Q1 --- Q2`) truncated the front-matter
- * and leaked the rest into the body. An optional leading BOM is tolerated.
- */
-const YAML_FRONT_MATTER_RE = /^\uFEFF?---\n[\s\S]*?\n---\n?/;
-
-/**
- * Normalize a foreign markdown string into Docmost's canonical markdown surface
- * so the strict canonical parser accepts it losslessly: normalize line endings,
- * strip a leading YAML front-matter block, then rewrite GFM reference footnotes
- * into inline footnotes. Add further fixture-driven foreign-surface cases here as
- * they are found.
- */
-export function normalizeForeignMarkdown(markdown: string): string {
-  if (!markdown) return markdown;
-  // Normalize CRLF -> LF FIRST. The line-anchored front-matter regex requires a
-  // bare `\n` after the opening `---`, and convertReferenceFootnotes splits on
-  // `\n`; a Windows/CRLF foreign file (`---\r\n…`) would otherwise slip past the
-  // front-matter strip and leak into the body. The canonical parser
-  // (page-file.ts parsePageFile) normalizes the same way before its FRONTMATTER_RE.
-  const src = markdown.replace(/\r\n/g, '\n');
-  const withoutFrontMatter = src.replace(YAML_FRONT_MATTER_RE, '').trimStart();
-  return convertReferenceFootnotes(withoutFrontMatter);
-}

apps/server/src/integrations/queue/processors/general-queue.processor.comment-mark.spec.ts

@@ -139,13 +139,19 @@ describe('GeneralQueueProcessor — COMMENT_MARK_UPDATE (#399)', () => {
     );
   });
 
-  it('skips (no throw) when the comment row has vanished', async () => {
+  it('reconcile (#496): comment row vanished → strips the orphan anchor mark', async () => {
     const { proc, collaborationGateway, commentRepo } = makeProc();
     commentRepo.findById.mockResolvedValue(undefined);
 
     await expect(
       proc.process(job({ ...base, action: 'resolve', ts: 1000 })),
     ).resolves.toBeUndefined();
-    expect(collaborationGateway.handleYjsEvent).not.toHaveBeenCalled();
+    // A resolve/unresolve mark job whose comment row is gone leaves a silent
+    // orphan; the worker self-heals by stripping the anchor instead of returning.
+    expect(collaborationGateway.handleYjsEvent).toHaveBeenCalledWith(
+      'deleteCommentMark',
+      'page.page-1',
+      { commentId: 'c-1', user: { id: 'user-1' } },
+    );
   });
 });

apps/server/src/integrations/queue/processors/general-queue.processor.ts

@@ -95,7 +95,8 @@ export class GeneralQueueProcessor
    * #399: apply a comment's inline-mark mirror in the collab Y.Doc, off the HTTP
    * critical path. Runs the SAME gateway path the synchronous comment.service
    * code used (byte-identical mark op):
-   *   - resolve / unresolve → resolveCommentMark (flip the `resolved` attribute);
+   *   - resolve / unresolve → resolveCommentMark (flip the `resolved` attribute),
+   *     OR strip an orphan anchor when the comment row has vanished (#496);
    *   - delete → deleteCommentMark (strip the ephemeral-suggestion anchor #329).
    * The op is idempotent, so a BullMQ retry is safe. Throwing propagates to
    * WorkerHost → the job is retried and, on exhaustion, surfaces in failed-job
@@ -133,7 +134,18 @@ export class GeneralQueueProcessor
     // of this resolve), skip it rather than flip the mark to a stale state.
     const comment = await this.commentRepo.findById(commentId);
     if (!comment) {
-      // The comment vanished (e.g. hard-deleted) → nothing left to mirror.
+      // #496 reconcile: the comment row is GONE (e.g. an ephemeral apply/dismiss
+      // hard-deleted it while this resolve/unresolve mark job sat in the queue),
+      // but its inline anchor may still live in the doc — a silent orphan mark
+      // pointing at a comment that no longer exists. Self-heal by stripping it
+      // instead of just returning: this closes the divergence the fire-and-forget
+      // resolve/unresolve enqueue (comment.service resolveComment) could leave.
+      // Idempotent — deleteCommentMark on an already-absent mark is a no-op.
+      await this.getCollaborationGateway().handleYjsEvent(
+        'deleteCommentMark',
+        documentName,
+        { commentId, user },
+      );
       return;
     }
     const wantResolved = action === 'resolve';