Move every SERVER Markdown->ProseMirror path off the editor-ext markdown layer
(`markdownToHtml`, a second marked-based parser) onto the canonical
`@docmost/prosemirror-markdown` package, and add a foreign-markdown normalizer at
the import boundary.
Code:
- `ImportService.processMarkdown` (single `.md` upload) now parses
`markdownToProseMirror(normalizeForeignMarkdown(md))` directly — no HTML hop.
- `PageService.parseProsemirrorContent` markdown case (page create/update with
`format: 'markdown'`) same.
- `FileImportTaskService` (zip import) parses markdown with the package, then
serializes to HTML (`jsonToHtml`) so the SHARED HTML attachment / internal-link
pipeline (processAttachments + formatImportHtml + processHTML) keeps handling
`.md` and `.html` imports uniformly. The markdown PARSE — the drift source — no
longer goes through editor-ext; the PM->HTML->PM hop that follows is lossless
plumbing for attachment resolution, not a second parse.
- `canonicalizeFootnotes` stays as an idempotent #228 safety net for the HTML
path (a no-op on the already-canonical markdown output).
Normalizer (`integrations/import/utils/foreign-markdown.ts`): a TEXT pre-pass,
NOT a parser fork. The strict canonical parser does not accept GFM `[^id]`
reference footnotes (and would misread `[^id]: def` as a CommonMark link-ref
definition, silently corrupting the ref into a bogus link), so the normalizer
rewrites reference footnotes into canonical inline `^[def]` before parsing.
Callout surfaces (`:::type` and `> [!type]`) are intentionally NOT touched — the
canonical parser already accepts BOTH natively, so normalizing them would be
redundant and risk degrading its nesting/code-fence-aware handling.
Fixtures-first: foreign-markdown.spec pins the normalizer and the end-to-end
acceptance (no literal `[^id]`/`:::` leaks; re-export is canonical). The two
footnote-canonicalize specs are updated to the canonical output — the parser
assigns fresh `fn-*` ids, so they now assert by definition BODY order (still
reference-ordered, deduped, orphan-free).
FINAL CHECK: `grep -rn "htmlToMarkdown\|markdownToHtml" apps/server/src` (non
-test) is now empty — both editor-ext markdown-layer functions are gone from the
server.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Move every SERVER ProseMirror->Markdown path off the editor-ext markdown layer
(`htmlToMarkdown`, a second turndown-based converter) onto the canonical
`@docmost/prosemirror-markdown` package.
- `ExportService.exportPage` (page/space markdown export) and
`collaboration.util.jsonToMarkdown` (used by page.controller's markdown
responses and the AI public-share chat tool) now serialize DIRECTLY from
ProseMirror JSON via `convertProseMirrorToMarkdown` — no HTML intermediate, no
`<colgroup>` scrub (the converter emits GFM tables directly).
This is the SAME serializer the git-sync vault writer feeds, so an exported page
BODY is byte-identical to its vault representation: no more export-md vs vault-md
drift. The HTML export path is unchanged (still `jsonToHtml`).
Emitted markdown moves to the canonical forms: callouts `> [!type]` (not
`:::type`), inline footnotes `^[…]` (not `[^id]`), lossless images
` <!--img {…}-->` (editor-ext dropped width/height/align).
Fixtures-first: export-markdown.spec asserts those canonical forms and the
export==vault-by-construction equality (both call the package converter). The
one deliberate export/vault delta — export prepends the page title as an H1
while the vault carries it in frontmatter — is pinned by a test.
Test infra: declare the `@docmost/prosemirror-markdown` workspace dep; teach
jest to load its ESM build (babel-jest) and stub `@tiptap/react` (server code
imports editor-ext, whose node views reference React renderers only used in a
live browser editor — never on the server).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- F1: added in-app execute tests for the two wirings that ACTUALLY changed in the
migration (the contract-parity test only checks advertised schema keys, not
execute bodies): movePage forwards the newly-added optional `position` to
client.movePage (and passes undefined position + null parent when omitted); the
table trio (insert/delete/updateCell) forwards the unified `table` param
positionally. A field destructured under the wrong name would have silently
passed undefined to the client (execute is any-cast, tsc won't catch it).
- F2: rewrote the three migrated descriptions that hardcoded snake_case sibling
tool names (which the in-app camelCase layer exposes under different ids,
violating the registry's own transport-neutral-prose convention) into neutral
prose: getPage "use get_page_json" -> "use the lossless page-JSON read tool";
updatePageJson "get_page_json -> ... -> update_page_json" -> "read the page-JSON
view -> modify -> write it back", "prefer rename_page" -> "prefer the rename-page
tool"; exportPageMarkdown "import_page_markdown round-trip" -> "page-Markdown
import round-trip" (the last was a direct regress — the in-app base said the
camelCase importPageMarkdown). (stashPage's pre-existing get_page_json mention is
out of scope, per the reviewer.)
Gate: mcp build 0; ai-chat-tools.service + tool-tiers (catalog-partition) pass,
incl. the 5 new execute-wiring tests.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- F1: added '/icons/' to STATIC_PATH_PREFIXES — public/icons/ is copied verbatim
to client/dist (a sibling of the already-included brand/vad/locales), and
index.html references /icons/favicon-*.png on every page load, so those requests
were getting their own route labels instead of collapsing to `static`.
- F2: corrected the comment — only /assets/ is content-hashed (unbounded per
deploy); /vad//brand//locales//icons/ have stable names (repetitive, not
unbounded). Either way none belong in the API-route histogram.
- F3: the negative test now exercises the trailing-slash boundary (the actual
anti-false-collapse guard): '/assets' (no slash), '/assetsx/foo.js',
'/iconset/x.png' must NOT collapse to `static` — cases that a buggy
includes()/slashless-prefix impl would wrongly collapse. Plus '/icons/*' added
to the positive it.each.
Gate: server tsc 0; metrics.spec passes.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Migrates share_page / sharePage into the transport-agnostic spec registry
(schema + description declared once; each transport keeps only its execute/auth):
- sharePage (deferred) -> SHARED_TOOL_SPECS; index.ts uses registerShared(),
ai-chat uses sharedTool(); removed from INLINE_TOOL_TIERS.
Drift reconciled (documented inline): both inline copies already carried the
"only share when the user explicitly asked" security framing, so the old
"per-transport divergence" note in BOTH layers was STALE — there was no real
behavioral divergence, only wording drift. The canonical description merges the
MCP copy's URL-format + idempotency detail with the in-app copy's reversibility
note and keeps the shared security framing. pageId keeps the MCP copy's stricter
.min(1). The MCP execute keeps its own `searchIndexing ?? true` default
(per-layer, not part of the shared schema).
Intentionally NOT migrated (kept inline — genuinely divergent, as their existing
notes state):
- search / searchPages: the in-app tool is a semantic+keyword hybrid (RRF) with
in-process access control and a tuned schema (limit 1-20); the MCP `search` is
a plain REST full-text search (limit up to 100). Different behavior AND schema.
- docmost_transform / transformPage: the in-app tool deliberately omits the
`deleteComments` schema field (a comment-deletion guardrail) and carries a
shorter description. Different schema.
Gate: mcp build 0 + node --test 458/458 (page-search excluded — hangs only under
the local re2->RegExp type-shim, its source untouched), server jest 775 incl.
tool-tiers catalog-partition + shared-spec contract parity, server tsc 0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Migrates the three-layer page tools into the transport-agnostic spec registry
(schema + description declared once; each transport keeps only its execute/auth):
- getPage, listPages (core), createPage, movePage, renamePage, deletePage,
updatePageJson, exportPageMarkdown (deferred) -> SHARED_TOOL_SPECS; index.ts
uses registerShared(), ai-chat uses sharedTool(); removed from
INLINE_TOOL_TIERS. Tiers preserved from CORE_TOOL_KEYS (getPage/listPages =
core, the rest deferred).
delete_page is genuinely three-layer (in-app deletePage exists), so it IS
migrated — not MCP-only. Its H4 guardrail is preserved: the shared schema
exposes ONLY pageId, so no permanentlyDelete/forceDelete flag can reach the
client (still asserted by ai-chat-tools.service.spec.ts).
Descriptions merged (documented inline): each canonical text takes the MCP
copy's richer structural notes plus the in-app copy's reversibility framing.
Schema DRIFT reconciled (documented inline):
- createPage.content: MCP pinned .min(1) but the in-app copy left it unbounded
and DOCUMENTS an empty body as valid ("may be empty" — creating an empty page
to fill later is a real use). Kept the looser no-min form: create_page now also
accepts an empty body (harmless) and no previously-valid in-app input is
rejected. title/spaceId keep the MCP .min(1) (empty is never valid).
- movePage: MCP exposed an optional `position` (fractional-index) field the
in-app copy lacked. Unified by KEEPING position — the in-app client already
accepts an optional position arg, so the in-app execute now forwards it;
optional, so no previously-valid call is rejected. `parentPageId` is nullable
on both (real JSON null -> root); the MCP execute keeps its 'null'/'' string
coercion as a per-layer robustness fallback.
- getPage/renamePage/updatePageJson/exportPageMarkdown/listPages: kept the MCP
copy's stricter .min(1) on ids where the in-app copy was unbounded.
Per-transport execute logic preserved: getPage's {title,markdown} projection,
updatePageJson's JSON-string normalization, list_pages' default limit/tree, and
move_page's cycle guard + positive-confirmation check all stay in their execute
bodies.
Intentionally NOT touched: updatePageContent (Markdown-based body update; no MCP
equivalent) and getTable (name-convention divergence, see tables family) stay
inline.
Gate: mcp build 0 + node --test 458/458 (page-search excluded — hangs only under
the local re2->RegExp type-shim, its source untouched), server jest 770 incl.
tool-tiers catalog-partition + shared-spec contract parity + deletePage H4
guardrail, server tsc 0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Production OOM'd (JS heap 1.85 GB / 2 GB limit) during a ~20-step,
~28k-chunk autonomous agent turn. Heap snapshot analysis (memlab) showed a
single DefaultStreamTextResult retaining ~1.7 GB via the never-consumed
leftover tee() branch of its internal baseStream.
Root cause in ai@6.0.134: streamText substitutes the default text() output
strategy even when the caller passes NO `output` option. Its
createOutputTransformStream then accumulates the ENTIRE turn text and, on
EVERY text-delta, enqueues `{ part, partialOutput }` where partialOutput is
a flat snapshot of all text so far (JSON.stringify flattens the
cons-string) — O(n²) memory across the turn. Every consumer accessor tees
baseStream and keeps the second branch as the new baseStream; the final
leftover branch is never read, so its controller queue holds every chunk
(28,225 x ~164 KB in the OOM'd run) for the life of the turn.
Fix (pnpm patch on both dist/index.js and dist/index.mjs):
- pass the raw, possibly-undefined `output` option into
createOutputTransformStream instead of defaulting to text()
- when output == null, publish each text-delta immediately without
accumulating turn text or producing partialOutput snapshots; streaming
granularity is unchanged, and callers that DO request an output strategy
keep the original behavior
Our server never uses partialOutputStream / experimental_output / the
output option, so no behavior changes for us beyond memory.
Regression spec ai-sdk-partial-output.patch.spec.ts drives the real
patched SDK with MockLanguageModelV3: asserts per-delta textStream
granularity, an EMPTY experimental_partialOutputStream (tripwire — yields
one cumulative partial per delta when unpatched), and the PATCH(docmost
marker in both installed dist bundles. Also documents the patch in
AGENTS.md (must be re-created when bumping `ai`) and CHANGELOG.md.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Migrates the three-layer table WRITE tools into the transport-agnostic spec
registry (schema + description declared once; each transport keeps only its
execute/auth):
- tableInsertRow, tableDeleteRow, tableUpdateCell -> SHARED_TOOL_SPECS;
index.ts uses registerShared(), ai-chat uses sharedTool(); removed from
INLINE_TOOL_TIERS (all three are deferred; not in CORE_TOOL_KEYS).
Drift reconciled (documented inline): the four table tools previously carried a
"NOT shared" note in both layers over a single parameter-NAME drift — the MCP
layer named the table reference `table`, the in-app layer `tableRef`. Unified on
the MCP name `table` (renaming the public MCP parameter would break external MCP
clients; the in-app parameter is model-facing/prompt-only and safe to rename).
The in-app execute bodies now destructure `table`. Descriptions took the MCP
copy's richer wording (documents `#<index>`, padding, header-row behavior) plus
the in-app copy's "Reversible via page history" note; both fields keep the MCP
copy's stricter .min(1) (in-app left them unbounded); sibling tool references
phrased transport-neutrally.
Intentionally NOT migrated (kept inline): table_get / getTable. Its MCP tool
name is noun-first (`table_get`) while the in-app key is verb-first (`getTable`),
which breaks the snake_case(inAppKey) naming convention the registry enforces
(shared-tool-specs.contract.spec.ts). Renaming the public MCP tool would break
external clients, so it stays per-transport — but its in-app reference param was
still aligned to `table` (was `tableRef`) for consistency with the migrated trio.
Gate: mcp tsc 0 + node --test 458/458 (page-search excluded — hangs only under
the local re2->RegExp type-shim, its source is untouched), server jest 730 incl.
tool-tiers catalog-partition + shared-spec contract parity, server tsc 0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Migrates the three-layer comment tools into the single transport-agnostic spec
registry (schema + model-facing description declared once; each transport keeps
only its execute/auth):
- createComment, listComments, resolveComment, checkNewComments — moved to
SHARED_TOOL_SPECS; index.ts uses registerShared(), ai-chat uses sharedTool();
removed from INLINE_TOOL_TIERS (tier/catalogLine now on the spec). Tiers
preserved from CORE_TOOL_KEYS (create/list/resolve = core, check = deferred).
Intentionally NOT migrated (kept MCP-inline): update_comment / delete_comment —
they are MCP-only by design; the in-app AI-chat layer deliberately has no
updateComment/deleteComment (comment edits are irreversible / not
version-tracked), asserted by ai-chat-tools.service.spec.ts. A registry spec's
tier/catalogLine are in-app metadata and the catalog-partition test forbids a
deferred spec without a live in-app tool, so these stay per-transport.
Drift reconciled (documented inline): createComment/listComments/checkNewComments
took the more-maintained/superset description + stricter .min(1) guards.
resolveComment: `resolved` drifted (MCP optional+default(true) vs in-app
required) — kept the MCP superset, so in-app resolveComment now accepts an
omitted `resolved` (defaults to resolve) — a deliberate, backward-compatible
unification (never rejects a previously-valid input).
Gate: mcp build 0 + node --test 480/480, ai-chat 654, tool-tiers (incl. F3
catalog-partition) 16/16, server tsc 0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Follow-up to #355: http_request_duration_seconds's `route` label captured raw
content-hashed asset filenames (route="/assets/index-CAbxDtto.js",
"/assets/chunk-*.js"). @fastify/static serves each file through a route whose
matched routeOptions.url IS the raw hashed path, so the label was unbounded — a
new set of names every deploy, growing the series forever (the exact cardinality
leak the API routes were protected against).
resolveRouteLabel now detects a static request by its path prefix (/assets/,
/vad/, /brand/, /locales/) FIRST and collapses it to a single `static` label
(query string stripped before the check); API routes still use the template and
404s still collapse to `unknown`. Static edge latency is already measured by
Traefik's traefik_router_request_duration_*.
Gate: server tsc 0; metrics.spec passes (added static-collapse + query-strip +
"real API route mentioning assets is NOT collapsed" cases).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A long-lived branch can add a migration whose timestamped filename sorts BEFORE
migrations already applied in prod (#234's 20260627T130000-ai-chat-runs merged
after 20260704T120000-client-metrics was live). Kysely's migrator with the
default ordered setting then rejects the applied set as "corrupted migrations"
(no longer a prefix of the sorted list), throws, and the app crash-loops on boot
— exactly incident #361 (502s for ~11 min after a develop deploy). #119 and #120
(June branches) are the next such threats.
Two levels, both:
1. CI migration-order gate (a new `migration-order` job in test.yml, PR-only):
fails the PR when an added migration sorts at/before the newest migration on
the base branch, with an actionable message to rename it to a current
timestamp before merge. This is the primary defense — makes back-dating
impossible to merge accidentally.
2. `allowUnorderedMigrations: true` on BOTH Migrators (migration.service.ts
startup auto-migrate + migrate.ts CLI): the runtime safety net — Kysely applies
a not-yet-applied older migration instead of bricking startup, so a back-dated
migration that bypasses the gate (manual push / hotfix branch) still boots.
Trade-off documented inline: apply order across instances may differ from
lexicographic, so migrations must stay independent (ours each create their own
objects); the CI gate remains the primary line.
Verified: allowUnorderedMigrations is a valid Kysely 0.28.17 Migrator option;
server tsc clean; the gate script rejects a back-dated filename and passes a
current one. No new deps, no migration, no runtime behavior change beyond the
migrator resilience.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
20260627T130000-ai-chat-runs sorted before the already-executed
20260702T120000-ai-chat-page-snapshot, so Kysely's strict ordering
check ("corrupted migrations") crash-looped the server on startup.
- rename 20260627T130000-ai-chat-runs.ts -> 20260704T130000-ai-chat-runs.ts
- update the mirror comment in database/types/db.d.ts
- F10 [stability]: closeMetricsServer() now calls server.closeIdleConnections()
+ server.unref() after server.close(). server.close()'s callback doesn't fire
until keep-alive sockets drain, and the scraper (VictoriaMetrics/vmagent) holds
an idle keep-alive socket — so onModuleDestroy's awaited close would hang until
the scraper disconnects or the orchestrator SIGKILLs on the kill-grace window.
closeIdleConnections() drops idle keep-alive sockets so shutdown completes
immediately (Node 22, per the Dockerfile base).
- F9 [test]: client-telemetry.module.spec.ts pins the E1=B register() gate — the
core of the "public endpoint OFF by default" decision: flag unset / any non-
"true" value ("false"/""/"0"/…) → empty controllers+providers (route absent);
"true"/"TRUE" → registers VitalsController + VitalsService. A flag-inversion or
truthiness regression that reopened the anonymous disk-fill surface now fails.
- F11 [regression/perf]: the db_query_duration_seconds token work (firstSqlToken
regex + Set lookup) is now gated on isMetricsEnabled() in database.module.ts, so
a non-metrics deployment pays NOTHING per query (previously observeDbQuery
no-op'd but the token was still computed on every query). Also hoisted the
13-element known-token Set to a module const (KNOWN_SQL_TOKENS) so it's built
once, not per query.
Gate: server tsc 0; metrics + vitals + client-telemetry suites pass (incl. the
new register-gate test).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Maintainer resolved E1 as variant B: the public vitals sink + client collection
must be OFF by default (else client_metrics grows unbounded on a self-host deploy
with no external pruner, via an unauthenticated public endpoint).
- F1: new operator flag CLIENT_TELEMETRY_ENABLED (default OFF), SEPARATE from
METRICS_PORT (Grafana reads the table directly, independent of the scrape port).
ClientTelemetryModule.register() provides VitalsController ONLY when the flag is
true (route absent otherwise); the flag reaches the client via window.CONFIG
(config.ts isClientTelemetryEnabled), and initVitals() early-returns when off.
- F2/F3 [throttler]: this repo's ThrottlerGuard applies EVERY named throttler to
every guarded route unless skipped. The new VITALS bucket therefore (a) newly
bound collab-token → 429 behind shared/NAT IPs, and (b) the vitals route didn't
skip the stricter public-share-ai (5/min) bucket → effective 5/min not 120.
Fix (additive, global config unchanged): vitals.controller @SkipThrottle the
other buckets + @Throttle VITALS 120/min; collab-token adds VITALS_THROTTLER to
its existing @SkipThrottle (restoring its prior effectively-unthrottled state).
- F4: metrics node:http server is closed on shutdown (MetricsServerLifecycle
OnModuleDestroy → closeMetricsServer(), fired by enableShutdownHooks).
- F5: docSize outside [0, int4-max] drops to null (keeping the event) instead of
overflowing int4 and failing the WHOLE batch insert (+ 2 tests).
- F6: .env.example documents METRICS_PORT (no default — unset = subsystem OFF) +
CLIENT_TELEMETRY_ENABLED; fixed the inaccurate "default 9464" wording.
- F7: disabled/non-sampled sessions install ZERO observers — isVitalsActive()
(enabled && sampled) gates reportClientMetric AND the page-editor
measurePageOpen + dispatchTransaction wrapping.
- F8: kept db.d.ts hand-added (wontfix) — this repo HAND-CURATES db.d.ts (verified
across recent fork migrations a32fba63/8c5b57eb/fdeede00); codegen would be the
deviation. The ClientMetrics interface maps the migration 1:1.
Gate: server tsc 0, client tsc 0, server metrics/vitals/telemetry/throttle 21
tests, client route-template 5. No new deps.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Squashed for a clean rebase onto develop (was 19 commits; the reviewer approved
the net diff at fb246080). Detaches an agent run from the HTTP request/browser
window: a run is a first-class lifecycle object (ai_chat_runs), a browser
disconnect no longer kills it, a concurrent-run insert-gate prevents double runs,
and a reopened chat live-follows a still-running run via a polled observer merge.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The metrics INFRA is already deployed (VictoriaMetrics scraping docmost:9464,
Grafana dashboards, alerts) with a target `gitmost-app` that is red because the
app half didn't exist. This is that half. The contract (metric names, port,
table, endpoint) is FIXED by the deployed infra and matched exactly.
Server (prom-client):
- A bare node:http `/metrics` server on METRICS_PORT (default 9464), SEPARATE
from the Fastify :3000 listener so /metrics never exists publicly; the whole
subsystem is OFF when METRICS_PORT is unset.
- collectDefaultMetrics() + http_request_duration_seconds{method,route,status}
via a Fastify onResponse hook using the ROUTE TEMPLATE (req.routeOptions.url,
never the raw URL — bounded cardinality; 404 -> "unknown"), EXCLUDING SSE/
streaming responses (would record the connection lifetime and poison p95).
- db_query_duration_seconds (Kysely log callback, labelled by the leading SQL
token), bullmq_queue_depth{queue} (getJobCounts every 15s) +
bullmq_job_duration_seconds{queue} (worker completed/failed),
collab_store_duration_seconds (around onStoreDocument).
- POST /api/telemetry/vitals — PUBLIC (sendBeacon) but IP-throttled; ~16KB body
cap, <=50 events/batch, metric-name + rating whitelist, attr truncated to 120
chars, batch insert; malformed/foreign/oversized silently dropped and 200'd (no
browser retry). New migration `client_metrics` (schema byte-identical to the
contract, both indexes, conditional grafana_ro GRANT; no app-side retention —
the maintenance container prunes >90d).
Client (web-vitals):
- initVitals() decides sampling ONCE per session (25%, sessionStorage) BEFORE
subscribing; onINP/onLCP/onCLS/onTTFB (attribution) buffered + flushed via
navigator.sendBeacon on visibilitychange:hidden and a timer (not fetch-per-
metric). Custom: editor_tx_ms (dispatchTransaction sync-part timer, >8ms, with
doc_size), page_open_ms, longtask_ms. Route labels are templates only; no
titles/slugs/text.
Gate: server + client tsc 0, frozen install 0 (added prom-client + web-vitals +
regenerated the lock), server metrics/vitals tests 11, client route-template 5,
and the migration verified valid against real Postgres.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- F1: document AI_CHAT_DEFERRED_TOOLS in .env.example (AI_* section) — default
ON = deferred loading (compact catalog + loadTools), =false restores the old
"all tools always active" behavior.
- F2: integration test of the ON path in ai-chat-stream.int-spec.ts — a deferred
tool activated via loadTools is active on the SAME turn's next step but a fresh
turn starts cold (CORE + loadTools only), proving the per-turn activatedTools
Set does not leak across turns/chats. Drives the real streamText loop with a
MockLanguageModelV3 and inspects recorded per-step activeTools-filtered tools.
- F3: replace the magic toHaveLength(28) in tool-tiers.spec.ts with a two-way
partition against the LIVE in-app toolset (AiChatToolsService.forUser keys):
every non-core tool must appear in buildInAppDeferredCatalog and every catalog
entry must map to a real non-core tool — so a future tool forgotten in
INLINE_TOOL_TIERS fails the suite instead of silently vanishing from the agent.
No production logic change (mechanism was already reviewed correct).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The in-app AI agent shipped all ~41 tool schemas on every model step. This
adds a two-tier catalog: core tools (frequent or one-line) stay always-active;
the rest are advertised as a compact catalog and their full schema is fetched
on demand via the loadTools meta-tool, wired through ai@6 prepareStep's
per-step activeTools.
- tools/tool-tiers.ts: CORE_TOOL_KEYS, INLINE_TOOL_TIERS, applyLoadTools,
catalog builders (+ tool-tiers.spec.ts, 13 cases).
- ai-chat.service.ts prepareAgentStep: returns activeTools =
[...CORE_TOOL_KEYS, loadTools, ...activatedTools]; per-turn activated Set.
- ai-chat.prompt.ts: buildToolCatalogBlock renders the deferred catalog.
- mcp/tool-specs.ts: tier + catalogLine metadata (external snake_case /mcp
transport unchanged).
- EnvironmentService.isAiChatDeferredToolsEnabled(): AI_CHAT_DEFERRED_TOOLS,
default ON per issue intent (kill-switch =false restores old behavior).
Gate: server ai-chat 631/631, tool-tiers 13/13, mcp 472/472, tsc clean.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
F4 [critical] — the anti-join `DELETE … WHERE NOT EXISTS(child)` was still racy
under Postgres READ COMMITTED: a reply INSERT holds FOR KEY SHARE on the parent;
the DELETE's start snapshot doesn't see the uncommitted child (NOT EXISTS true),
blocks on the reply's lock, and when the reply commits the parent was only LOCKED
(not modified) so EvalPlanQual does NOT re-check → the DELETE proceeds and CASCADE
destroys the just-committed reply. Replaced with a transaction: SELECT the parent
FOR UPDATE (conflicts with the reply's FOR KEY SHARE → serializes the concurrent
reply), re-check for a child with a FRESH statement in the same tx (a new RC
snapshot sees a just-committed reply), delete only if still childless (return 1)
else return 0 (caller resolves). The FOR UPDATE lock is held to end-of-tx so no
reply can insert between the re-check and the delete. Signature unchanged, so the
service + its mocked unit tests are untouched; docstrings updated.
F5 [warning] — the client Dismiss button was gated only on canComment, but the
server now gates dismiss on owner-or-space-admin, so a non-owner non-admin saw a
button the server 403s. `canShowDismiss` now also requires
`isOwnerOrAdmin = currentUser?.user?.id === comment.creatorId || userSpaceRole ===
"admin"` (the same gate the comment delete-menu already uses); threaded into both
call sites.
F6 [warning] — added a REAL-DB int-spec
(apps/server/test/integration/comment-delete-if-childless.int-spec.ts, + a
createComment seeder): (a) childless → returns 1, row gone; (b) committed reply →
returns 0, parent+reply survive; (c) CONCURRENCY — a second connection inserts a
reply (FOR KEY SHARE) and commits mid-operation while deleteCommentIfChildless
blocks on FOR UPDATE → asserts it returns 0 and both rows survive (a blind
anti-join would lose the reply here). Ran against live Postgres — 3/3 pass.
server tsc clean; comment jest 53 + int-spec 3 (live Postgres) pass. client tsc
clean; comment vitest 56 pass.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Maintainer escalation decision (B) + reviewer findings on the ephemeral-
suggestion PR.
Authz (decision B): POST /comments/dismiss-suggestion now gates the destructive
branch on owner-OR-space-admin, mirroring POST /comments/delete exactly (same
SpaceCaslAction.Manage / SpaceCaslSubject.Settings, same owner short-circuit,
same ForbiddenException). A non-owner non-admin who tries to dismiss another's
childless suggestion gets Forbidden before the service runs. Apply stays on
canEdit (accepting an edit is the editor's semantics), unchanged.
F1 [blocking] — atomic conditional delete closes the hasChildren→delete race.
New repo `deleteCommentIfChildless(id)` runs a single
`DELETE FROM comments WHERE id=:id AND NOT EXISTS (SELECT 1 FROM comments child
WHERE child.parent_comment_id = comments.id)` (verified by compiling the Kysely
expression to SQL — the correlated subquery references the OUTER comments.id).
deleteEphemeralSuggestion strips the mark first, then the conditional delete: if
it removed the row → commentDeleted + outcome 'deleted'; if a reply raced in
(0 rows) → fall back to resolveComment (outcome 'resolved') so the discussion and
the new reply survive. No reply can be cascade-deleted anymore.
F2 [warning] — the apply/dismiss onError success-noop is narrowed from 404||400
to 404 ONLY. A 400 means the comment is ALIVE (apply's 400 = the thread was
resolved-not-applied), so it now shows a real error (surfacing the server
message) and KEEPS the comment in cache instead of a false "applied" + dropping a
live thread.
F3 [suggestion] — the 404-race client tests assert the success toast fired.
Tests: server — dismiss authz (owner ok / non-owner-non-admin Forbidden /
space-admin ok), the delete→resolve race (hasChildren=false but conditional
delete returns 0 → resolve, no commentDeleted), delete-path asserts switched to
deleteCommentIfChildless; client — apply-400 and dismiss-400 (kept in cache, red,
not success) + the toast assertions.
server tsc clean, comment+collaboration jest green; client tsc clean, comment
vitest 54 passed.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Agent suggestion-edits (comments with suggestedText, #315) piled up: Apply
auto-resolved the thread, cluttering the resolved tab, and the anchors stayed in
the document. Make them ephemeral: resolving (Apply OR the new Dismiss) makes the
comment DISAPPEAR — hard-delete + remove the Yjs `comment` mark — UNLESS the
thread has replies, in which case resolve it (preserve the discussion). Manual
Resolve is unchanged. Scope: only comments with `suggestedText`.
Server:
- New collab event `deleteCommentMark` (collaboration.handler) mirroring
resolveCommentMark, wiring the existing removeYjsMarkByAttribute to strip the
anchor from the doc.
- `finalizeAppliedSuggestion` forks on `hasChildren`: replies → apply + resolve
(outcome 'resolved'); none → apply + hard-delete + mark removal (outcome
'deleted').
- New `dismissSuggestion` (validates top-level + suggestedText + not applied/not
resolved) with the same fork; permission `canComment` (NOT canEdit — dismiss
doesn't change page text); audit COMMENT_SUGGESTION_DISMISSED. New
POST /comments/dismiss-suggestion; apply stays canEdit.
- Both return `{ outcome: 'deleted' | 'resolved' }` so the client picks the
optimistic action.
Data-integrity (review F1): the shared `deleteEphemeralSuggestion` removes the
anchor mark FIRST and FATALLY, then deletes the DB row only on success. The row
delete is irreversible, so a mark-removal failure — including the
COLLAB_DISABLE_REDIS "no live instance" hard-error — must abort the whole
operation (→ 5xx, repeatable) rather than swallow the error and leave a permanent
orphan anchor pointing at a deleted comment. `deleteCommentMark` is no longer
best-effort (unlike resolve, where the row is kept and a failed mark is
recoverable).
Client:
- `canShowDismiss` (canComment) alongside `canShowApply` (canEdit); a "Dismiss"
button next to Apply in the suggestion block.
- `useApplySuggestionMutation`/`useDismissSuggestionMutation` reconcile the cache
on `outcome` ('deleted' → remove; 'resolved' → relocate to the resolved tab).
- Idempotent races (review F2): BOTH apply and dismiss onError reduce 404/400 to
success (comment already gone/resolved), dropping it from the cache instead of
a red error — restores the #315 apply idempotency the ephemeral delete would
otherwise break.
- i18n Dismiss / "Не применять" (ru/en).
Not done (flagged): deleteCommentMark on the normal /comments/delete path — left
out (would change every non-suggestion delete + needs gateway injection; the
interactive client already strips the mark via unsetComment). Out of scope per
the issue.
Tests: server — apply/dismiss delete-vs-resolve fork, all four dismiss state
guards, the deleteCommentMark handler, controller authz (dismiss=canComment,
apply=canEdit), AND a mark-removal-failure test proving the row is NOT deleted +
the error propagates (F1). client — Dismiss show-conditions, outcome cache
reconciliation, and 404 idempotent race for BOTH dismiss and apply (F2).
Verified: server tsc clean; comment+collaboration jest 144 passed. client tsc
clean; vitest 905 passed | 1 expected-fail.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Editorial roles (Corrector/Factchecker) brute-forced `get_node` block-by-block to
find occurrences (unquoted «ё», straight quotes, «т.е.»), burning tokens. New
`search_in_page(pageId, query, {regex?, caseSensitive?, limit?})` reads the page's
ProseMirror JSON via the existing getPageRaw and searches it IN MEMORY — no server
endpoint, no DB/schema change, no touch to the packages/mcp/src/lib schema mirror.
New pure `searchInDoc(doc, query, opts)` (packages/mcp/src/lib/page-search.ts):
recursive descent to each TEXT CONTAINER (paragraph/heading/table-cell paragraph),
glues its inline text via `blockPlainText` (a match survives inline-mark
boundaries — e.g. «т.е.» split across bold/italic), searches literal (indexOf) or
regex, and returns `{ total, truncated, matches:[{ nodeId, blockIndex, type,
before, match, after }] }`. `nodeId` is the container's attrs.id or the
`#<topLevelIndex>` of the enclosing top-level block — the SAME ref format
get_node/patch_node/comment-anchoring accept (verified identical to getNodeByRef),
so the agent goes straight from a hit to a targeted comment; `before`/`after` are
~40-char windows for a unique selection. `total`/`truncated` always reported (never
silent truncation). Lives in the SHARED_TOOL_SPECS registry → exposed in BOTH
transports (external /mcp + in-app AI-chat), with a SERVER_INSTRUCTIONS line and a
DocmostClientLike signature + contract-test entry. Corrector/Factchecker prompts
get a one-line "use search_in_page first" hint (versions bumped, catalog hash lock
refreshed).
Guards: empty/whitespace query → clear error; invalid regex → clear error (not a
generic 500); zero-length regex matches (`\b`, `a*`) skipped with lastIndex
advanced (no loop/flood); MAX_PATTERN_LENGTH=1000, MAX_CONTAINER_TEXT=100k bound
each exec; limit clamped [1,200] (default 50).
Tests: new page-search.test.mjs (17) — literal+regex, case-sensitivity,
mark-boundary glue, nodeId for paragraph/heading (attrs.id) and table-cell
(#<index> fallback), context bounds, limit/total/truncated + clamp, invalid
regex/empty/over-long errors, zero-length skip, empty-doc null-safety.
mcp: tsc clean; node --test 467 passed (+17). apps/server: tsc --noEmit clean
(DocmostClientLike + wiring). catalog check.mjs OK.
Known limitations (from internal review, non-blocking):
- Residual ReDoS: a crafted catastrophic-backtracking pattern (e.g. `(a+)+$`)
against a large single container can hang the event loop — JS regex is not
interruptible, so the length caps bound the base but not the backtracking.
Realistic exposure is low (containers are small; the pattern is supplied by the
authenticated model). Candidate for a follow-up hardening (safe-regex validation
or a worker+timeout) if it matters.
- Case-insensitive LITERAL search folds via toLowerCase; a char whose lowercase
differs in length (e.g. Turkish İ) BEFORE a match could shift the context
window — negligible for the RU/EN editorial scenario.
- On a `#<index>` table-cell fallback, `type` is the inline container ("paragraph")
while nodeId addresses the top-level block — addressing is correct; the field is
documented as the container's type.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The AI agent (MCP + in-app chat) saw ALL comments incl. resolved via two
channels, cluttering its context and breaking fragment search. Default now:
the agent sees only ACTIVE discussions; resolved is opt-in. Active anchors and
threads are always kept.
Channel 1 — resolved comment anchors on agent reads (converter option):
`convertProseMirrorToMarkdown(content, options?)` gains
`options.dropResolvedCommentAnchors` (default false — zero change for every
existing caller incl. git-sync). Both `case "comment"` emitters (top-level and
the raw-HTML inlineToHtml path) emit BARE text (no `<span data-comment-id>`) when
`resolved && the flag`; active anchors keep their wrapper. mcp `getPage` passes
the flag; `export_page_markdown` does NOT (lossless export must preserve resolved
anchors — that is why it is an opt-in option, not unconditional); `get_page_json`
is untouched (lossless PM JSON). Built on the #293 package converter.
Channel 2 — `list_comments` default active-only: `listComments(pageId,
includeResolved=false)` now returns `{ items, resolvedThreadsHidden }` (was a
bare array). By default a RESOLVED top-level thread is hidden wholesale — the
root AND every reply anchored to it (a thread is gated only by its root's
resolvedAt; a resolved reply under an ACTIVE root stays). `resolvedThreadsHidden`
counts hidden threads so the agent knows to re-query. `includeResolved:true`
returns everything. The `includeResolved` param is added to both tool
registrations (MCP index.ts + in-app ai-chat-tools.service.ts); `DocmostClientLike`
signature updated. Server `findPageComments` is NOT touched — the web UI's tabs
depend on the full feed; filtering is only at the mcp-client level. All internal
call sites (export_page_markdown / checkNewComments / transformPage) updated to
`.items` with `includeResolved:true` to keep their full-feed behavior.
The comment model is assumed FLAT (a reply's parentCommentId points at the
thread root) — documented in the filter; a future reply-of-reply model would
need a root-walk there.
Tests: resolved-comment-anchors.test.ts (6 — anchor dropped with flag / kept
without, for BOTH emitters; active always kept); list-comments-resolved.test.mjs
(4 — resolved thread+reply hidden + counter; includeResolved:true returns all;
an ACTIVE thread with a RESOLVED reply is NOT hidden).
package vitest: 664 passed; tsc clean. mcp: node --test 458 passed; tsc clean.
apps/server + git-sync: tsc clean (converter option default-off).
NOTE: based on feat/293-B (#293/#326 STEP 5) — the converter lives in the
package; this PR is stacked on #333 and its base retargets to develop once #333
merges. mcp/build is gitignored (not committed).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Audit of all 41 tool descriptions against the actual implementation found
factually wrong or misleading texts:
- list_comments claimed '(paginated)' — it takes only pageId and returns ALL
comments in one call (internal pagination); now also states that RESOLVED
threads are included and how to filter them. In-app twin synced.
- search claimed the limit default is 'applied by the client' — the client
deliberately omits it so the SERVER applies its default.
- create_page's '(automatically moves it to the correct hierarchy)' said
nothing useful — now documents parentPageId nesting semantics; move_page
drops the stale 'essential for organizing pages created via create_page'.
- share_page now warns the page becomes accessible to ANYONE with the URL.
- get_page (both transports) now explains inline <span data-comment-id> tags
are comment anchors (incl. resolved) — markup, not page text.
- patch_node/delete_node/insert_node pointed only at the expensive page-JSON
view for block ids — now route through the cheap page outline first.
- docmost_transform marks 'Примечания переводчика' as the DEFAULT
notesHeading, overridable for non-Russian pages.
Checks: @docmost/mcp tests 450/450 (incl. the server-instructions guard);
server ai-chat-tools spec 20/20; mcp build/ artifacts rebuilt.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- editorial roles (ru/en): proofreader and line editor attach suggestedText
replacements to targeted fixes; fact-checker ALWAYS attaches the ready
correction for [Incorrect] verdicts; structural editor and narrator get a
light-touch rule for in-place rewordings; role versions bumped and the
content-hash lock refreshed
- MCP SERVER_INSTRUCTIONS: route 'propose a concrete text fix for one-click
human approval' to create_comment with suggestedText (unique-selection
reminder); build/ artifacts rebuilt
- AI-chat SAFETY_FRAMEWORK: mention the comment-suggestion capability so the
default assistant offers ready fixes instead of only describing changes
Checks: catalog check.mjs OK; @docmost/mcp tests 448/448; server
ai-chat.prompt spec 28/28.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
F1 [blocking]: a suggestion whose anchor matched via normalization could never
be applied (spurious 409). The comment mark lands on the doc's ACTUAL text
(Docmost auto-converts to typographic quotes/dashes/nbsp), but the stored
selection — used as expectedText at apply — was the raw ASCII agent input
(+substring(0,250)). So replaceYjsMarkedText's strict joined!==expectedText
always failed and threw "text changed" though nobody edited. Fix: new pure
getAnchoredText(doc, selection) reconstructs the exact raw doc substring the mark
covers (slicing identical to spliceCommentMark); on the suggestion path
client.createComment stores THAT as selection, so expectedText equals the marked
text and apply returns applied:true. Live anchoring still uses the raw agent
selection (normalization still finds the anchor). Truncation raised 250->2000
(+ DTO @MaxLength(2000)) so the anchored substring is never cut below the mark
span. Ordinary comments unchanged. AI-chat shares client.createComment, so
covered. Regression tests: getAnchoredText raw-vs-ASCII; create payload selection
is the typographic substring; apply with typographic expectedText -> applied.
F2 [blocking]: added comment.controller.spec.ts pinning that validateCanEdit runs
before applySuggestion (Forbidden -> applySuggestion never called; happy path ->
called; missing comment -> 404 without authorizing).
MCP 448 pass; server comment+yjs 54 pass. MCP build/ rebuilt.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Agents can attach a suggested replacement when creating an inline comment, via
both the MCP create_comment tool and the AI-chat createComment tool.
Because applying a suggestion edits the EXACT anchored text, an ambiguous anchor
would let Apply corrupt the wrong occurrence. So when suggestedText is set the
selection must occur EXACTLY ONCE:
- new countAnchorMatches(doc, selection) counts occurrences across all blocks
(same normalization/traversal as canAnchorInDoc), counting occurrences (2 in
one block => 2) — stricter than block-count, never under-counting distinct
occurrences (false-unique is the dangerous direction).
- client.createComment gains suggestedText: a pre-check (getPageJson +
countAnchorMatches: 0 => not-found, >=2 => ambiguity error) before create, and
an AUTHORITATIVE live check inside the anchoring mutation that recomputes on the
live doc and, if != 1, aborts and rolls back the just-created comment (reusing
the existing safeDeleteComment "anchor not found" path). Ordinary comments keep
first-occurrence behavior unchanged.
- suggestedText is rejected on a reply or without selection in all three layers
(MCP handler, MCP client, AI-chat tool), mirroring the server DTO/service.
- filterComment surfaces suggestedText/suggestionAppliedAt/suggestionAppliedById.
- DocmostClientLike.createComment signature updated. MCP build/ rebuilt.
Tests: countAnchorMatches (0/1/N, within/across/nested block, span nodes,
quote normalization); createComment (ambiguous refused pre-create, reply and
no-selection rejected, unique succeeds and forwards suggestedText, filterComment
surfaces it); ai-chat schema accepts suggestedText. MCP 443 pass; ai-chat 601 pass.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Server side of agent comment suggestions.
- CreateCommentDto gains optional suggestedText (<=2000). CommentService.create
accepts it ONLY for a top-level inline comment with a non-empty selection,
requires it be non-empty and differ from selection (else BadRequest), and
stores it.
- POST /comments/apply-suggestion (ApplySuggestionDto { commentId }): authorizes
with validateCanEdit (applying edits page text) BEFORE any structural check or
mutation, then CommentService.applySuggestion:
- runs the phase-3 collab event applyCommentSuggestion on `page.<pageId>` to
atomically check-and-replace the marked text, returning { applied, currentText };
- applied → stamp suggestion_applied_at/by, auto-resolve the thread, ws
commentUpdated, audit COMMENT_SUGGESTION_APPLIED;
- already-applied (DB) → idempotent success (no re-apply), self-healing the
resolve if it was missed — satisfies the issue's double-click / two-user
race requirement;
- collab verdict applied:false && currentText===suggestedText → idempotent
success (crash between doc mutation and DB write);
- text changed → 409 ConflictException carrying currentText;
- gateway undefined/throw → hard error, never a silent success.
- audit-events: COMMENT_SUGGESTION_APPLIED.
Tests: create validation (reply/no-selection/equal-to-selection rejected;
valid stored) + applySuggestion verdict branches incl. both idempotent paths.
jest src/core/comment: 33 passed.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
New custom collab event applyCommentSuggestion runs replaceYjsMarkedText inside
the document's Yjs transaction on the owning instance and returns the
{ applied, currentText } verdict to the API-server caller (cross-process via the
Redis bridge, whose customEventComplete/replyId already carries handler return
values).
- withYdocConnection is now generic and returns the callback's result (captured
in a closure, since hocuspocus connection.transact does not forward it). The
callback is typed synchronous-only: transact runs fn synchronously without
awaiting, so an async fn would mutate outside the transaction and lose
atomicity.
- collaboration.gateway.handleYjsEvent: when Redis is disabled
(COLLAB_DISABLE_REDIS), dispatch the handler locally against the single
hocuspocus instance and return its verdict instead of silently returning
undefined (which would make apply a no-op). Also fixes the pre-existing silent
no-op of setCommentMark/resolveCommentMark without Redis.
Tests: handler spec (applied mutates doc + returns verdict; changed-text returns
{applied:false} without mutating; args forwarded; withYdocConnection returns the
value) and gateway spec (no-Redis path dispatches locally, returns the verdict,
not undefined).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The primitive behind "Apply comment suggestion": walk the XmlFragment, collect
the delta segments carrying the `comment` mark for a commentId, and replace them
with new text ONLY if the run is intact (single Y.XmlText, contiguous, and the
joined text still equals the expected anchor). Otherwise return a verdict
{ applied:false, currentText } — null when the anchor is gone, else the current
text — so the caller can report "someone changed it". On apply it deletes the
run and re-inserts the new text re-attaching the same comment mark (thread stays
anchored). Mutates in place for the caller's connection.transact(); opens no
transaction of its own.
Non-string inserts (embeds) advance the offset by their 1-unit index length so a
marked segment after an embed gets the right position and an embed inside a run
is correctly rejected as a changed anchor.
Tests (yjs.util.spec.ts): happy path (mark preserved, surrounding text and no
mark-bleed), resolved-mark match, changed text, deleted anchor, paragraph split,
interleaved unmarked text, and embed before/inside the run. 17 passed.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add suggested_text / suggestion_applied_at / suggestion_applied_by_id to the
comments table (migration) and mirror them in the hand-curated db.d.ts Comments
interface. suggested_text holds a proposed replacement for the comment's
anchored selection; the applied_* columns record who applied it and when.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
In prod the AI provider resets the connection pre-response (ECONNRESET); the
#175 pre-response retry recovers it, but 2 of the 3 allowed attempts were burned
in a single turn — no headroom, and one more reset would surface an error to the
user. This is tuning for resilience (not a diagnosis of who resets):
- Retry budget 2 → 4 (total 5 attempts), env-configurable via
AI_STREAM_PRE_RESPONSE_RETRIES (0 = no retry; empty/invalid → default 4).
- Backoff: linear 150*(attempt+1) → capped exponential + full jitter
(preResponseBackoffMs, a pure injectable helper): base 150ms, ×2 per attempt,
capped 2000ms, delay = random in [0, capped]. Avoids a synchronized retry
storm and spreads reconnects across the reset window.
- Keep-alive default 10_000 → 4_000 ms so undici recycles idle sockets before a
~5s upstream/middlebox idle cutoff can poison them (a common pre-response
reset cause). Still env-overridable via AI_STREAM_KEEPALIVE_MS.
- .env.example documents both knobs.
Timeout (900s), RETRYABLE_CONNECT_CODES, and the instrumentation are unchanged.
refs #310
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
POST /api/ai-chat/bound-chat 500'd with Postgres 22P02 because the client
sends a page slugId (10-char nanoid) in the request `pageId` field, which the
server passed straight into the UUID `page_id` column. The chat-to-document
binding silently broke (client fail-softs to a new chat) and every slug-URL
page open logged a 500.
Fix: resolve the incoming id to a real page UUID on the server. PageRepo.findById
already accepts both a uuid and a slugId (isValidUUID→slugId fallback), so
boundChat now resolves the page first, guards it against a foreign/unknown
workspace (returns {chatId:null} before any chat lookup — no cross-workspace
probe), and looks up the latest chat by the resolved page.id (real uuid).
Client: renamed the local pageId→slugId for clarity (the value is a slugId);
the wire body key stays `pageId` so the DTO is unchanged. DTO left @IsString()
(a @IsUUID() would only turn the 500 into a 400 and still break binding).
Test: bound-chat spec asserts a slugId resolves and findLatestByPage is called
with the real uuid; a foreign-workspace page → {chatId:null} without a chat
lookup (no leak); an unknown id → {chatId:null}, no throw.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
F1: remove an accidentally-committed self-referential symlink
packages/mcp/node_modules/node_modules -> an absolute build-machine path (leaked a dev
home path, a pnpm artifact useless in the repo), and add a targeted ignore so it can't
recommit.
F2: the commentUpdated broadcast re-emitted the caller's pre-loaded comment mutated in
place, so the {agent,launcher} stack survived only because the controller happened to
load it with includeCreator:true — the fragile coupling that let the stack vanish on
edit once already. update() now RE-FETCHES the enriched comment before broadcasting,
symmetric with create()/resolveComment() (the row is already persisted), so all three
broadcasts carry the stack regardless of any caller's pre-load. Adds a caller-contract
test asserting all three broadcasts emit agent/launcher for an agent comment and neither
for a non-agent one, spotlighting the update path (non-vacuous vs the old re-emit).
F3: add a direct test of the page-history attachPageHistoryAgent mapping (its distinct
lastUpdatedSource/lastUpdatedAiChatId/lastUpdatedBy column set): role / no-role / MCP /
non-agent, and that the internal agentRole join column is stripped.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The same tool metadata (zod schema + model-facing description) was hand-duplicated
between the standalone MCP server and the in-app AI-chat agent, so every tweak had to
land in two places and copies drifted (a materialized parity bug). The shared
transport-agnostic registry (packages/mcp/src/tool-specs.ts) already de-duplicates 14
tools; this migrates two more genuinely-identical ones — patch_node/patchNode and
insert_node/insertNode. The canonical description is a strict SUPERSET of both originals
(keeps MCP's "without resending the whole document" + table-structure/anchor guidance
AND the in-app "reversible via page history" / "exactly one of anchorNodeId or
anchorText" framing — no model-facing guidance dropped); the schema is identical (the
in-app side just gains MCP's .min(1) on ids, a safe tightening). Each transport keeps its
own execute/auth wrapper, and the in-app parseNodeArg node-arg normalization is unchanged.
The three table tools are intentionally NOT merged (a real param-name divergence:
table vs tableRef) — documented on both sides. Other per-transport divergences
(search/share/create_comment/transform/list_pages) are left separate with a short comment
explaining why (the issue asked to flag these as intentional). DocmostClientLike stays a
hand-mirror (the ESM/CJS boundary blocks a compile-time type import; a runtime drift-guard
already pins it). Also fixes a latent contract-spec bug: derive `required` from
`instanceof z.ZodOptional` (matches the emitted JSON schema) instead of `isOptional()`,
which wrongly reported z.any() fields as optional.
Partially addresses #294.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
For AI-agent-authored content (comments + page history), replace the text AI-AGENT
badge with an avatar stack: the agent in front, the human who launched it smaller and
behind. This fixes the inverted hierarchy (the action was the agent's; the human just
launched it). closes#300.
Backend: a single server-authoritative resolver resolveAgentProvenance normalizes to
{ agent, launcher } from server columns only (createdSource/lastUpdatedSource, aiChatId,
creator, chat role) — nothing from request input, so agent identity can't be spoofed.
Internal chat -> agent = chat role (name/emoji), launcher = human; external MCP
(aiChatId null) -> agent = the agent account, launcher = null; non-agent -> neither.
The role join (aiChatId -> ai_chats.role_id -> ai_agent_roles) deliberately does NOT
filter enabled/deleted_at, so a later-disabled role still labels historical content
(mirrors findById, not findLiveEnabled). Enrichment is applied on BOTH findPageComments
(list) AND findById (the create/resolve/update broadcast path), so the stack shows on
live comment events and doesn't vanish on resolve/edit.
Frontend: new AgentAvatarStack + AgentGlyph (avatarUrl -> role emoji on violet ->
IconSparkles on violet), integrated into comment-list-item and history-item where the
badge was; the deep-link-to-chat click moved onto the stack. ai-agent-badge removed.
Tests: AgentAvatarStack (role/no-role/MCP/click/non-clickable), the provenance resolver
+ recorder tests proving the role join never filters enabled/deleted, and findById
enrichment (guards the live-broadcast regression).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Page/space export (Markdown & HTML, both via jsonToHtml -> generateHTML) crashed with
"Export failed:undefined" on any page carrying a `comment` mark. Root cause:
comment.renderHTML returned a LIVE DOM node (document.createElement + a click listener)
whenever a global `document` existed — and the in-process MCP module injects a jsdom
global.window+global.document into the Node server, defeating the old
`typeof document === "undefined"` guard. The server export runs happy-dom's
DOMSerializer, which crashes appending the foreign jsdom node
(NodeUtility.isInclusiveAncestor -> "Cannot read properties of undefined (reading
'length')"). comment is the only extension returning a live node.
Fix: widen the guard with an isNodeRuntime check (process.versions.node) so on any Node
runtime renderHTML returns the plain, serializable spec array — even when MCP injected
jsdom globals. The browser branch (createElement + click -> ACTIVE_COMMENT_EVENT) is
untouched, so in-editor comment interactivity is preserved (Vite defines only
process.env as a member-expression substitution, no `process` object in the browser
bundle, so isNodeRuntime is false there). The mcp schema mirror already returns a spec
array and is not on the export path (tiptapExtensions imports Comment from
@docmost/editor-ext), so no mirror change is needed.
Also: export-modal now reads the real error text from the response Blob
(responseType:'blob' made err.response.data.message always undefined) so a failed export
shows the server's message instead of "undefined".
Adds a regression test that runs the real jsonToHtml on a comment-marked doc with
jsdom globals injected (reproduces the crash on the unpatched code, passes after).
closes#298
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
F1: pc.title (untrusted cross-user page title) was interpolated raw into the
markdown export heading. Reusing escapeAttr alone (the prompt sink's XML-attribute
sanitizer, strips < > ") is insufficient here because the sink is MARKDOWN: link
/image syntax survives, so a title like  or [phish](http://evil)
injects a remote image / clickable link into the downloaded .md disguised as a
trusted system annotation. Add markdownHeadingSafe() = escapeAttr() + backslash-
escape [ and ] (disables both [text](url) and ; a bare (url) is inert).
F2: cover the title branch — a title that collapses to empty via escapeAttr falls
to the bare heading (no ("")), and a link/image-injection title is neutralized
(non-vacuous vs the escapeAttr-only version).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The #274 page_changed marker lived only in the ephemeral system prompt, so the
diff the agent saw was invisible in the chat export/history, and the note was
too weak — the agent still overwrote the user's manual edits with a full-page
replace.
- Persist the diff the agent saw as metadata.pageChanged on the assistant row
(flushAssistant), threaded into all five flush call sites in stream(). Model
replay (rowToUiMessage/rowParts) reads only metadata.parts, so the sibling
never re-injects the note into the model context on later turns.
- Render the persisted diff as a labelled block (en/ru) before the message body
in the server-side Markdown export (chat-markdown.util.ts).
- Strengthen PAGE_CHANGED_NOTE: mandate a fresh getPage re-read and targeted
edits (editPageText/patchNode/insertNode/deleteNode) instead of a whole-page
replace, and never revert or overwrite the user's edits.
Tests: prompt, export and service specs updated; 114 pass, tsc clean.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
F1: escape the collaborative page title before interpolating into
<page_changed page="..."> (and the pre-existing openedPage attr) — strip
<>" and collapse whitespace, so a crafted title can't break out of the
attribute into the system prompt (cross-user injection).
F2: neutralize <page_changed>/</page_changed> occurrences inside the diff body
so a crafted line can't close the block early.
F3: remove the dead content_hash column (written every turn, never read) —
migration, repo, service hashing + crypto import, db.d.ts, spec asserts.
F4: test the best-effort catch branches (detectPageChange / snapshotOpenPage
swallow errors and don't break the turn).
F5: soften the overstated 'diff cannot smuggle instructions' comment to
defense-in-depth framing referencing the F1/F2 mitigations + safety sandwich.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The agent rebuilds context from DB each turn and didn't know the user manually
edited the open page since its last response, so it could overwrite those edits.
Add a per-turn ephemeral <page_changed> note in the system prompt (twin of
INTERRUPT_NOTE, self-clearing) carrying a unified Markdown diff of what changed
since the END of the agent's previous turn.
- New ai_chat_page_snapshots table (migration + hand-declared db.d.ts/entity
types) storing the page Markdown per (chat,page) at each turn's end.
- Pure computePageChange util (whitespace-normalized unified diff via the
existing jsdiff dep, 6KB cap + getPage hint).
- Turn start: if the open page's updatedAt moved past the snapshot, diff current
vs snapshot; non-empty -> PAGE_CHANGED_NOTE in the safety sandwich.
- Turn end: upsert the snapshot on EVERY terminal path (onFinish/onError/onAbort,
once) so the agent's own edits are excluded by construction even on aborted
turns.
All best-effort (never breaks/latency-regresses a turn); fast path when updatedAt
is unchanged. Server-only.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Real root cause of the silent MCP edit loss: the web editor always opens the
collaboration document by the page UUID (`page.${page.id}`), but the MCP
opened it by the agent-supplied id — usually a slugId — so `page.${pageId}`
became `page.<slugId>`. For one DB page that is TWO independent Yjs documents;
both persist to the same `pages` row (findById/updatePage resolve id or
slugId), so the human tab's debounced store overwrites the agent edit
(last-store-wins) — gone after reload, never shown live. The slugId doc also
made the server's transclusion sync + embedding reindex throw Postgres 22P02.
Fix:
- MCP (primary): resolvePageId(pageId) returns the canonical UUID — a UUID
short-circuits with no network call, a slugId resolves once via getPageRaw
and is cached both ways. Every collab-write path (mutatePageContent /
updatePageContentRealtime / replacePageContent and the mutate/replace/
unlocked seams) now opens by the resolved UUID, so the MCP and the editor
share ONE Yjs doc. replaceImage's whole-operation page lock also keys on the
UUID so it serializes against the other (now-UUID-keyed) writes.
- Server (defense + kills the 22P02 noise): onStoreDocument passes the resolved
page.id — not the raw doc-name id — to syncTransclusion, the embedding queue,
the mention-notification job, addContributors, and the in-tx history read.
Content store and the empty-guard are untouched.
Tests: a new MCP test stands up a real Hocuspocus server and asserts a slugId
input opens `page.<uuid>` (never `page.<slugId>`), with UUID short-circuit and
single-resolve caching; the server spec asserts the side-effects receive the
UUID for a `page.<slugId>` doc. closes#260
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
agent_coder
agent_vscode
vvzvlad
claude code agent 227
agent_coder