fix(#346 review F1-F4): no 206-compress + Vary + precompress VAD + cache test

- F1 [HIGH — data corruption]: @fastify/compress was compressing 206/Range
  attachment responses while Content-Range still described the RAW offsets, so a
  resuming client (curl -C -, download managers) appended encoded bytes as raw →
  corrupted file. sendFileResponse now sets the request header `x-no-compression`
  (the documented @fastify/compress opt-out — its onSend skips when the request
  carries it; the reviewer's `Content-Encoding: identity` does NOT work because
  compress explicitly excludes `identity` and overwrites it). This opts the whole
  download route (both 200 full-file and 206 range) out of on-the-fly compression
  — correct, since attachment bytes are final and mostly binary.
- F2: static responses now emit `Vary: Accept-Encoding` (the preCompressed
  content-negotiated /assets/* were `immutable` without Vary → shared-cache could
  serve a brotli variant to an identity/gzip-only client).
- F3: vite compression `include` extended to .wasm/.onnx so the VAD binaries
  (~26MB .wasm, ~2.3MB .onnx under public/vad) are precompressed at build (.br
  emitted) instead of runtime-brotli'd on every request. (include REPLACES the
  plugin default, so the default js/css/json/html set is re-listed.)
- F4: extracted the cache classification into a pure `resolveStaticAssetHeaders`
  + static.module.spec.ts (3 tests: /assets/* immutable+Vary, index.html
  no-store, non-hashed not-immutable).

Gate: server tsc 0 (deps present), static.module.spec 3/3, client build emits
.wasm.br/.onnx.br, frozen install 0.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

.github/workflows/develop.yml

@@ -151,6 +151,12 @@ jobs:
       - name: Build editor-ext
         run: pnpm --filter @docmost/editor-ext build
 
+      # @docmost/prosemirror-markdown is an ESM workspace package the server
+      # imports at runtime; its build/ is gitignored and test:e2e has no pretest
+      # hook, so build it before the e2e run (mirrors the test.yml job).
+      - name: Build prosemirror-markdown
+        run: pnpm --filter @docmost/prosemirror-markdown build
+
       - name: Run migrations
         run: pnpm --filter ./apps/server migration:latest
 

.github/workflows/test.yml

@@ -13,6 +13,49 @@ permissions:
   contents: read
 
 jobs:
+  # Guard against a long-lived branch adding a migration whose timestamped
+  # filename sorts BEFORE migrations already applied on the target branch (and
+  # thus in prod). The Kysely startup migrator rejects that as "corrupted
+  # migrations" and crash-loops the app on boot (incident #361). This gate fails
+  # the PR so the migration is renamed to a current timestamp before merge. Only
+  # runs for pull_request events (needs a base branch to diff against).
+  migration-order:
+    if: github.event_name == 'pull_request'
+    runs-on: ubuntu-latest
+    timeout-minutes: 5
+    steps:
+      - name: Checkout (full history for the base-branch diff)
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+      - name: Added migrations must sort after the newest on the base branch
+        env:
+          TARGET_BRANCH: ${{ github.base_ref }}
+        run: |
+          set -euo pipefail
+          MIG_DIR="apps/server/src/database/migrations"
+          # checkout above already did fetch-depth:0 (full history). Fetch the base
+          # WITHOUT --depth (a shallow graft would truncate the base history and
+          # break the merge-base when the base has moved ahead of the PR merge —
+          # exactly the long-branch-vs-moving-base case this gate guards, #361).
+          git fetch --no-tags origin "$TARGET_BRANCH"
+          newest_on_target=$(git ls-tree -r --name-only "origin/${TARGET_BRANCH}" "$MIG_DIR" | sort | tail -1)
+          # NO `|| true`: a diff failure (e.g. an unresolved merge-base) must fail
+          # the job CLOSED — a gate whose job is to BLOCK must never pass on error.
+          # `set -e` above already aborts on a non-zero diff exit.
+          added=$(git diff --diff-filter=A --name-only "origin/${TARGET_BRANCH}...HEAD" -- "$MIG_DIR")
+          bad=0
+          for f in $added; do
+            if [[ "$f" < "$newest_on_target" || "$f" == "$newest_on_target" ]]; then
+              echo "::error::Migration $f sorts at or before the newest on ${TARGET_BRANCH} ($newest_on_target) — rename it with a CURRENT timestamp before merge (do not change its contents). See incident #361."
+              bad=1
+            fi
+          done
+          if [ "$bad" -eq 0 ]; then
+            echo "Migration order OK (added migrations all sort after $newest_on_target)."
+          fi
+          exit $bad
+
   test:
     runs-on: ubuntu-latest
     timeout-minutes: 20

AGENTS.md

@@ -201,7 +201,7 @@ pnpm workspace (`pnpm@10.4.0`) orchestrated by **Nx**. Four workspace packages:
 | `apps/client` | `client` | React 18 + Vite + Mantine 8 + TanStack Query + Jotai | SPA frontend |
 | `packages/editor-ext` | `@docmost/editor-ext` | Tiptap/ProseMirror | Shared Tiptap node/mark extensions, imported by both the client and the server |
 | `packages/mcp` | `@docmost/mcp` | MCP SDK, Tiptap, Yjs | Standalone MCP server, also bundled into the server at `/mcp`. Consumes the shared converter/schema from `@docmost/prosemirror-markdown` (#293) — it no longer carries its own vendored converter/schema copy |
-| `packages/prosemirror-markdown` | `@docmost/prosemirror-markdown` | Tiptap, marked, jsdom | The single, canonical ProseMirror↔Markdown converter + Docmost schema mirror (#293). Consumed by `mcp` and `git-sync`; there is exactly ONE copy of the converter now |
+| `packages/prosemirror-markdown` | `@docmost/prosemirror-markdown` | Tiptap, marked, jsdom | The single, canonical ProseMirror↔Markdown converter + Docmost schema mirror (#293). Consumed by `mcp`, `git-sync`, AND `apps/server` (server-side markdown import/export, #345); there is exactly ONE copy of the converter now |
 
 `build` targets are Nx-cached and dependency-ordered (`dependsOn: ["^build"]`), so `editor-ext` builds before the apps. `nx.json` sets `affected.defaultBase: main`.
 
@@ -214,6 +214,12 @@ Run from the repo root unless noted. The dev workflow needs **Postgres (with the
 > server, `APP_SECRET` mismatch between processes, a stale `editor-ext` white-
 > screening the client, LAN exposure. See **[docs/dev-stand.md](docs/dev-stand.md)**
 > for the step-by-step and the traps.
+>
+> **Testing the app against a stand** (browser E2E + out-of-band verification) has
+> its own non-obvious traps — the page has two ProseMirror editors (only the body is
+> collab-bound), a ~10s store debounce, and API-seeding the thing under test is a
+> silent no-test. See **[docs/how-to-test.md](docs/how-to-test.md)** before writing
+> UI tests.
 
 ```bash
 pnpm install                 # install all workspaces (uses pnpm patches; see package.json `pnpm.patchedDependencies`)
@@ -250,7 +256,10 @@ pnpm --filter server migration:codegen                   # regenerate src/databa
 ```
 Migration files live in `apps/server/src/database/migrations/` and are named `YYYYMMDDThhmmss-description.ts`. Fork-specific migrations only **add** tables (`page_embeddings`, `ai_chats`, `ai_chat_messages`, `ai_provider_credentials`, `ai_mcp_servers`, `page_template_references`) and columns (e.g. `pages.is_template`, a `NOT NULL DEFAULT false` boolean) — never drop/rewrite Docmost data.
 
-**Migration ordering — always check when merging branches/features.** Kysely runs migrations in **alphabetical (= timestamp) order** and refuses to start if a *new* migration sorts **before** one already applied to the DB (`corrupted migrations: ... must always have a name that comes alphabetically after the last executed migration`). When you merge a branch or land a feature, verify your migration's timestamp still sorts **after every migration that may already be applied on the target** (`/bin/ls -1 apps/server/src/database/migrations | sort | tail`). Branches developed in parallel routinely break this: a feature branch adds `…T130000-…`, `main` meanwhile ships and deploys `…T150000-…`, and after the merge the older-timestamped file is rejected at boot. **Fix = rename your migration to a timestamp after the latest one already in the target** (content unchanged — the filename is the ordering key), then rebuild so the compiled `dist/database/migrations/` picks up the new name.
+**Migration ordering — always check when merging branches/features.** Kysely runs migrations in **alphabetical (= timestamp) order**. A *new* migration that sorts **before** one already applied to the DB is a "back-dated" migration, which branches developed in parallel routinely produce: a feature branch adds `…T130000-…`, `develop` meanwhile ships and deploys `…T150000-…`, and after the merge the older-timestamped file has been skipped. Two layers guard this (both added for incident #361, where a back-dated migration crash-looped prod for ~11 min):
+
+- **CI gate (primary):** the `migration-order` job in `.github/workflows/test.yml` fails a PR whose added migration sorts at/before the newest on the base branch. **So the fix is to rename your migration to a timestamp after the latest one already in the target** (`/bin/ls -1 apps/server/src/database/migrations | sort | tail`; content unchanged — the filename is the ordering key), then rebuild so the compiled `dist/database/migrations/` picks up the new name.
+- **Runtime safety net:** both Migrators (`migration.service.ts` startup auto-migrate + `migrate.ts` CLI) set `allowUnorderedMigrations: true`, so the app does **not** refuse to start on an out-of-order migration — it applies the skipped older one instead of crash-looping. Kysely's `#ensureNoMissingMigrations` guard is still on (a *removed* applied migration is still an error). Because apply order can then differ from lexicographic across instances, migrations must stay **independent** (each creates its own objects) — the CI gate remains the primary line; this net only covers a gate bypass (manual push / hotfix branch).
 
 ## Architecture — the big picture
 
@@ -284,7 +293,7 @@ The API server is a Fastify app with a global `/api` prefix (`main.ts` excludes
 ### Client structure
 Vite SPA. Code is organized by feature under `apps/client/src/features/*` (mirrors the server domains: `page`, `space`, `comment`, `ai-chat`, `editor`, …). Conventions:
 - **TanStack Query** for server state (one `queries/` file per feature), **Jotai** atoms for local/shared UI state, **Mantine 8** + CSS modules (`*.module.css`) + `postcss-preset-mantine` for UI.
-- The editor is Tiptap; shared node/mark extensions live in `packages/editor-ext` and are imported by **both the client and the server** (collaboration, import/export) — editor schema changes often need to be made in `editor-ext`, not just the client. The ProseMirror↔Markdown converter and its Docmost schema mirror now live in a SINGLE package, `@docmost/prosemirror-markdown` (#293), consumed by both `mcp` and `git-sync` — do NOT reintroduce a per-package copy. `editor-ext` is the upstream source of the Tiptap schema; the package's `docmost-schema.ts` mirrors it and a serializer-contract test (`packages/prosemirror-markdown/test/serializer-contract.test.ts`) guards the boundary (every schema node must have a converter case), so a drift surfaces as a failing test rather than silent divergence.
+- The editor is Tiptap; shared node/mark extensions live in `packages/editor-ext` and are imported by **both the client and the server** (collaboration, schema, `canonicalizeFootnotes`) — editor schema changes often need to be made in `editor-ext`, not just the client. Server-side markdown import/export no longer lives in `editor-ext`: it goes through the canonical converter (#345, see below). The ProseMirror↔Markdown converter and its Docmost schema mirror now live in a SINGLE package, `@docmost/prosemirror-markdown` (#293), consumed by `mcp`, `git-sync`, and `apps/server` (#345) — do NOT reintroduce a per-package copy. `editor-ext` is the upstream source of the Tiptap schema; the package's `docmost-schema.ts` mirrors it and a serializer-contract test (`packages/prosemirror-markdown/test/serializer-contract.test.ts`) guards the boundary (every schema node must have a converter case), so a drift surfaces as a failing test rather than silent divergence.
 - API access goes through `apps/client/src/lib/api-client.ts` (axios). The `@` alias maps to `apps/client/src`.
 - Runtime config is injected at build time by `vite.config.ts` via `define` (`APP_URL`, `COLLAB_URL`, `APP_VERSION`, …) — these come from the root `.env`, not from `import.meta.env`.
 
@@ -294,7 +303,7 @@ Vite SPA. Code is organized by feature under `apps/client/src/features/*` (mirro
 - **Errors must never be swallowed or shown as generic messages.** Every caught error MUST (1) be logged in full to the console/logger — error name, message, stack, `cause`, and (for HTTP/provider failures) the status code and response body — and (2) be surfaced to the user with a *specific, human-readable explanation of what actually went wrong*, never a bare generic string like "Something went wrong" / "Could not start recording" / "Transcription failed". Include the real reason (the underlying error/provider message) in the user-facing text. On the server, wrap third-party/provider failures with `describeProviderError` (or equivalent) and rethrow as a meaningful HTTP status + message — never let them collapse into an opaque 500. On the client, `console.error(<context>, err)` the raw error AND show the extracted reason (e.g. `err.response?.data?.message`, or the error `name: message`) in the notification.
 - The version string shown in the UI comes from `APP_VERSION` (CI/Docker) or `git describe --tags --always` (local), resolved in `vite.config.ts` — not from `package.json`.
 - Server TS config is permissive (`noImplicitAny: false`, `strictNullChecks: false`, `no-explicit-any` lint disabled). Follow the existing relaxed style rather than tightening types broadly.
-- Dependency versions are heavily pinned via `pnpm.overrides` and `pnpm.patchedDependencies` (`scimmy`, `yjs`) in the root `package.json`. Don't bump pinned/patched deps casually; the patches and overrides exist for compatibility/security reasons.
+- Dependency versions are heavily pinned via `pnpm.overrides` and `pnpm.patchedDependencies` (`scimmy`, `yjs`, `ai`) in the root `package.json`. Don't bump pinned/patched deps casually; the patches and overrides exist for compatibility/security reasons. The `ai@6.0.134` patch disables the SDK's O(n²) cumulative `partialOutput` accumulation when no output strategy is requested (server heap OOM on long agent runs, #184; tripwire test: `apps/server/src/integrations/ai/ai-sdk-partial-output.patch.spec.ts`) — it MUST be re-created via `pnpm patch` when bumping `ai`.
 - **Adding/renaming/removing an MCP tool requires updating `SERVER_INSTRUCTIONS`** in `packages/mcp/src/index.ts` — the intent-routing guide MCP clients receive on initialize. This applies both to inline `server.registerTool(...)` calls in `index.ts` and to specs in `packages/mcp/src/tool-specs.ts`. Enforced by `packages/mcp/test/unit/server-instructions.test.mjs`, which fails when a registered tool is not mentioned in the guide (deliberate opt-outs go into its `EXCEPTIONS` list). `packages/mcp/build/` is gitignored and rebuilt in CI/Docker via `pnpm build` (same convention as `git-sync`/`prosemirror-markdown`) — never commit it; rebuild locally after editing to run the tests.
 
 ## CI / release

CHANGELOG.md

@@ -169,6 +169,14 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Fixed
 
+- **The server no longer runs out of heap during long autonomous agent runs.** A
+  new pnpm patch on `ai@6.0.134` stops the SDK from building a cumulative
+  snapshot of the ENTIRE turn text on every streamed text-delta when no output
+  strategy was requested (our server never requests one). Unpatched, those
+  O(n²) `partialOutput` snapshots piled up in a never-consumed internal
+  `tee()` branch of the stream result — a ~20-step, ~28k-chunk agent run
+  retained ~1.7 GB and OOM'd the 2 GB JS heap. Streaming granularity is
+  unchanged; the patch must be re-created if `ai` is ever bumped. (#184)
 - **Internal links in exported Markdown no longer lose their visible text.** A
   link whose target page name had no file extension (e.g. a bare title) was
   collapsed to empty text during export, producing an unclickable, label-less

apps/client/package.json

@@ -98,6 +98,7 @@
     "typescript": "5.9.3",
     "typescript-eslint": "8.57.1",
     "vite": "8.0.5",
+    "vite-plugin-compression2": "2.5.3",
     "vitest": "4.1.6"
   }
 }

apps/client/public/locales/en-US/translation.json

@@ -1373,6 +1373,39 @@
   "The role catalog is unavailable": "The role catalog is unavailable",
   "Please try again later.": "Please try again later.",
   "No bundles available": "No bundles available",
+  "Content": "Content",
+  "Content language of the roles": "Content language of the roles",
+  "{{count}} updates available in {{bundles}} bundles": "{{count}} updates available in {{bundles}} bundles",
+  "Update all ({{count}})": "Update all ({{count}})",
+  "Updating {{current}}/{{total}}…": "Updating {{current}}/{{total}}…",
+  "{{count}} roles are installed in another language. A different language installs separately and appears as new.": "{{count}} roles are installed in another language. A different language installs separately and appears as new.",
+  "{{count}} roles": "{{count}} roles",
+  "{{count}} new — none installed": "{{count}} new — none installed",
+  "All installed · up to date": "All installed · up to date",
+  "{{count}} updates · {{installed}} up to date": "{{count}} updates · {{installed}} up to date",
+  "{{count}} new": "{{count}} new",
+  "{{count}} installed": "{{count}} installed",
+  "{{count}} updates": "{{count}} updates",
+  "Install bundle": "Install bundle",
+  "Install {{count}} selected": "Install {{count}} selected",
+  "Install bundle ({{count}})": "Install bundle ({{count}})",
+  "{{selected}} of {{total}} selected": "{{selected}} of {{total}} selected",
+  "Select all": "Select all",
+  "Deselect all": "Deselect all",
+  "Skipped": "Skipped",
+  "v{{version}}": "v{{version}}",
+  "{{count}} roles installed": "{{count}} roles installed",
+  "{{count}} roles installed · {{renamed}} renamed": "{{count}} roles installed · {{renamed}} renamed",
+  "{{count}} roles updated": "{{count}} roles updated",
+  "Installed {{installed}} · {{skipped}} skipped": "Installed {{installed}} · {{skipped}} skipped",
+  "A role named \"{{name}}\" already exists in this workspace.": "A role named \"{{name}}\" already exists in this workspace.",
+  "\"{{name}}\" is already installed.": "\"{{name}}\" is already installed.",
+  "Rename & install": "Rename & install",
+  "Couldn’t load the catalog": "Couldn’t load the catalog",
+  "Check your connection and try again. Installed roles are not affected.": "Check your connection and try again. Installed roles are not affected.",
+  "Retry": "Retry",
+  "The catalog is empty": "The catalog is empty",
+  "No role bundles are published for this language yet. Try switching the content language.": "No role bundles are published for this language yet. Try switching the content language.",
   "Already up to date": "Already up to date",
   "Updated to the latest version": "Updated to the latest version",
   "This role is no longer in the catalog": "This role is no longer in the catalog",

apps/client/public/locales/ru-RU/translation.json

@@ -1235,6 +1235,39 @@
   "The role catalog is unavailable": "Каталог ролей недоступен",
   "Please try again later.": "Попробуйте позже.",
   "No bundles available": "Наборы недоступны",
+  "Content": "Язык контента",
+  "Content language of the roles": "Язык контента ролей",
+  "{{count}} updates available in {{bundles}} bundles": "Доступно обновлений: {{count}} в наборах: {{bundles}}",
+  "Update all ({{count}})": "Обновить все ({{count}})",
+  "Updating {{current}}/{{total}}…": "Обновление {{current}}/{{total}}…",
+  "{{count}} roles are installed in another language. A different language installs separately and appears as new.": "Ролей установлено на другом языке: {{count}}. Другой язык устанавливается отдельно и отображается как новый.",
+  "{{count}} roles": "ролей: {{count}}",
+  "{{count}} new — none installed": "новых: {{count}} — ничего не установлено",
+  "All installed · up to date": "Все установлены · актуальны",
+  "{{count}} updates · {{installed}} up to date": "обновлений: {{count}} · актуальны: {{installed}}",
+  "{{count}} new": "новых: {{count}}",
+  "{{count}} installed": "установлено: {{count}}",
+  "{{count}} updates": "обновлений: {{count}}",
+  "Install bundle": "Установить набор",
+  "Install {{count}} selected": "Установить выбранные ({{count}})",
+  "Install bundle ({{count}})": "Установить набор ({{count}})",
+  "{{selected}} of {{total}} selected": "выбрано {{selected}} из {{total}}",
+  "Select all": "Выбрать все",
+  "Deselect all": "Снять выбор",
+  "Skipped": "Пропущено",
+  "v{{version}}": "v{{version}}",
+  "{{count}} roles installed": "Установлено ролей: {{count}}",
+  "{{count}} roles installed · {{renamed}} renamed": "Установлено ролей: {{count}} · переименовано: {{renamed}}",
+  "{{count}} roles updated": "Обновлено ролей: {{count}}",
+  "Installed {{installed}} · {{skipped}} skipped": "Установлено: {{installed}} · пропущено: {{skipped}}",
+  "A role named \"{{name}}\" already exists in this workspace.": "Роль с именем «{{name}}» уже существует в этом рабочем пространстве.",
+  "\"{{name}}\" is already installed.": "«{{name}}» уже установлена.",
+  "Rename & install": "Переименовать и установить",
+  "Couldn’t load the catalog": "Не удалось загрузить каталог",
+  "Check your connection and try again. Installed roles are not affected.": "Проверьте подключение и попробуйте снова. Установленные роли не затронуты.",
+  "Retry": "Повторить",
+  "The catalog is empty": "Каталог пуст",
+  "No role bundles are published for this language yet. Try switching the content language.": "Для этого языка ещё не опубликовано ни одного набора ролей. Попробуйте сменить язык контента.",
   "No roles configured": "Роли не настроены",
   "Already up to date": "Уже актуальна",
   "Updated to the latest version": "Обновлено до последней версии",

apps/client/src/features/ai-chat/queries/ai-chat-query.ts

@@ -1,6 +1,7 @@
 import {
   useInfiniteQuery,
   useMutation,
+  useQueries,
   useQuery,
   useQueryClient,
 } from "@tanstack/react-query";
@@ -307,6 +308,29 @@ export function useAiRoleCatalogBundleQuery(
   });
 }
 
+/**
+ * Eagerly open EVERY listed bundle's content in parallel for one language. The
+ * redesigned catalog shows each bundle's status summary in its COLLAPSED header,
+ * which needs every role's install state up front — so contents can no longer be
+ * lazy-loaded on expand. The catalog is small, so a fan-out of `useQueries` (one
+ * cached read per bundle, sharing the same cache keys as
+ * `useAiRoleCatalogBundleQuery`) is cheap. Gated by `enabled` (modal open + a
+ * resolved language) so nothing fetches while the modal is closed.
+ */
+export function useAiRoleCatalogBundlesQueries(
+  bundleIds: string[],
+  language: string,
+  enabled: boolean,
+) {
+  return useQueries({
+    queries: bundleIds.map((bundleId) => ({
+      queryKey: AI_ROLE_CATALOG_BUNDLE_RQ_KEY(bundleId, language),
+      queryFn: () => getAiRoleCatalogBundle(bundleId, language),
+      enabled: enabled && !!language,
+    })),
+  });
+}
+
 export function useImportAiRolesFromCatalogMutation() {
   const queryClient = useQueryClient();
   const { t } = useTranslation();

apps/client/src/features/ai-chat/queries/import-from-catalog-message.test.tsx

@@ -77,7 +77,14 @@ describe("useImportAiRolesFromCatalogMutation — success notifications", () =>
   });
 
   it("errors:[] -> only the summary notification (counts interpolated)", async () => {
-    await runMutation({ created: 3, renamed: 1, skipped: 2, errors: [] });
+    await runMutation({
+      created: 3,
+      renamed: 1,
+      skipped: 2,
+      errors: [],
+      createdRoles: [],
+      skippedRoles: [],
+    });
     expect(notificationsShowMock).toHaveBeenCalledTimes(1);
     expect(notificationsShowMock).toHaveBeenCalledWith({
       message: "Imported 3, renamed 1, skipped 2",
@@ -93,6 +100,8 @@ describe("useImportAiRolesFromCatalogMutation — success notifications", () =>
         { slug: "a", message: "name taken" },
         { slug: "b", message: "name taken" },
       ],
+      createdRoles: [{ slug: "ok", name: "Ok" }],
+      skippedRoles: [],
     });
     expect(notificationsShowMock).toHaveBeenCalledTimes(2);
     expect(notificationsShowMock).toHaveBeenNthCalledWith(1, {

apps/client/src/features/ai-chat/types/ai-chat.types.ts

@@ -108,12 +108,25 @@ export interface IAiRoleImportPayload {
   conflict: "skip" | "rename";
 }
 
-/** Import result counts (mirrors `importFromCatalog()`). */
+/**
+ * Import result (mirrors `importFromCatalog()`). The counters (`created`,
+ * `skipped`, `renamed`) drive the summary notification; the per-role lists
+ * (`createdRoles`, `skippedRoles`) drive the redesigned catalog modal's inline
+ * result plaque — which roles were installed (and any rename) and which were
+ * skipped and why (so the plaque can name the conflicting role and offer
+ * "Rename & install").
+ */
 export interface IAiRoleImportResult {
   created: number;
   skipped: number;
   renamed: number;
   errors: { slug: string; message: string }[];
+  createdRoles: { slug: string; name: string; renamedTo?: string }[];
+  skippedRoles: {
+    slug: string;
+    name: string;
+    reason: "name-conflict" | "already-installed";
+  }[];
 }
 
 /**

apps/client/src/features/ai-chat/utils/catalog-bundle-model.test.ts

@@ -0,0 +1,234 @@
+import { describe, it, expect } from "vitest";
+import {
+  bundleCounts,
+  bundlePhase,
+  installedLangForRole,
+  mapBundleRolesToView,
+  mapCatalogRoleToView,
+  nameConflictSlugs,
+  partialOffersRename,
+  type CatalogViewRole,
+} from "./catalog-bundle-model.ts";
+import type {
+  IAiRole,
+  IAiRoleCatalogRole,
+} from "@/features/ai-chat/types/ai-chat.types.ts";
+
+function installedRole(
+  source: { slug: string; language: string; version: number },
+  overrides: Partial<IAiRole> = {},
+): IAiRole {
+  return {
+    id: `role-${source.slug}-${source.language}`,
+    name: source.slug,
+    emoji: null,
+    description: null,
+    enabled: true,
+    autoStart: true,
+    launchMessage: null,
+    source,
+    ...overrides,
+  };
+}
+
+function catalogRole(
+  overrides: Partial<IAiRoleCatalogRole> = {},
+): IAiRoleCatalogRole {
+  return {
+    slug: "writer",
+    emoji: "✍️",
+    name: "Writer",
+    description: "Drafts copy.",
+    instructions: "be a writer",
+    autoStart: true,
+    launchMessage: null,
+    version: 3,
+    ...overrides,
+  };
+}
+
+// Build a minimal view role for bundlePhase tests.
+function viewRole(status: CatalogViewRole["status"]): CatalogViewRole {
+  return { slug: `s-${status}`, name: status, description: "", version: 1, status };
+}
+
+describe("bundlePhase", () => {
+  it("empty bundle -> empty", () => {
+    expect(bundlePhase([])).toBe("empty");
+  });
+
+  it("all importable, none installed -> allNew", () => {
+    expect(bundlePhase([viewRole("import"), viewRole("import")])).toBe(
+      "allNew",
+    );
+  });
+
+  it("nothing to import or update -> allInstalled", () => {
+    expect(bundlePhase([viewRole("installed"), viewRole("installed")])).toBe(
+      "allInstalled",
+    );
+  });
+
+  it("updates present, nothing to import -> updates", () => {
+    expect(bundlePhase([viewRole("update"), viewRole("installed")])).toBe(
+      "updates",
+    );
+  });
+
+  it("import + installed (no updates) -> mixed", () => {
+    expect(bundlePhase([viewRole("import"), viewRole("installed")])).toBe(
+      "mixed",
+    );
+  });
+
+  it("import + update -> mixed", () => {
+    expect(bundlePhase([viewRole("import"), viewRole("update")])).toBe("mixed");
+  });
+
+  it("a skipped role with nothing installed -> mixed (NOT allInstalled)", () => {
+    // F1: a bundle whose only non-installed role was skipped has 0 installed for
+    // it, so the collapsed 'All installed · up to date' header would contradict
+    // the open 'Installed 0 · 1 skipped' plaque. It must be mixed until resolved.
+    expect(bundlePhase([viewRole("skipped")])).toBe("mixed");
+  });
+
+  it("installed + a skipped role -> mixed (partial success is not allInstalled)", () => {
+    expect(bundlePhase([viewRole("installed"), viewRole("skipped")])).toBe(
+      "mixed",
+    );
+  });
+});
+
+describe("bundleCounts", () => {
+  it("tallies each status once", () => {
+    expect(
+      bundleCounts([
+        viewRole("import"),
+        viewRole("import"),
+        viewRole("installed"),
+        viewRole("update"),
+        viewRole("skipped"),
+      ]),
+    ).toEqual({ importable: 2, installed: 1, update: 1, skipped: 1 });
+  });
+});
+
+describe("nameConflictSlugs / partialOffersRename (reason -> action)", () => {
+  it("only name-conflict skips become the transient overlay / offer rename", () => {
+    const skipped = [
+      { slug: "writer", name: "Writer", reason: "name-conflict" as const },
+      { slug: "editor", name: "Editor", reason: "already-installed" as const },
+    ];
+    expect(nameConflictSlugs(skipped)).toEqual(["writer"]);
+    expect(partialOffersRename(skipped)).toBe(true);
+  });
+
+  it("an already-installed-only skip is informational: no overlay, no rename", () => {
+    const skipped = [
+      { slug: "editor", name: "Editor", reason: "already-installed" as const },
+    ];
+    expect(nameConflictSlugs(skipped)).toEqual([]);
+    expect(partialOffersRename(skipped)).toBe(false);
+  });
+});
+
+describe("installedLangForRole", () => {
+  it("returns the other language when the same slug is installed elsewhere", () => {
+    const roles = [installedRole({ slug: "writer", language: "ru", version: 2 })];
+    expect(installedLangForRole("writer", roles, "en")).toBe("ru");
+  });
+
+  it("returns undefined when the same slug is installed in the SAME language", () => {
+    const roles = [installedRole({ slug: "writer", language: "en", version: 2 })];
+    expect(installedLangForRole("writer", roles, "en")).toBeUndefined();
+  });
+
+  it("returns undefined when no install of the slug exists", () => {
+    expect(installedLangForRole("writer", [], "en")).toBeUndefined();
+  });
+
+  it("ignores manually-created roles (no source)", () => {
+    const roles = [
+      installedRole({ slug: "writer", language: "ru", version: 2 }, {
+        source: null,
+      }),
+    ];
+    expect(installedLangForRole("writer", roles, "en")).toBeUndefined();
+  });
+});
+
+describe("mapCatalogRoleToView", () => {
+  it("no install -> import status, catalog version, emoji preserved", () => {
+    const view = mapCatalogRoleToView(catalogRole(), [], "en");
+    expect(view).toMatchObject({
+      slug: "writer",
+      emoji: "✍️",
+      name: "Writer",
+      description: "Drafts copy.",
+      status: "import",
+      version: 3,
+    });
+    expect(view.installedRoleId).toBeUndefined();
+    expect(view.installedLang).toBeUndefined();
+  });
+
+  it("import with the slug installed in another language -> installedLang set", () => {
+    const roles = [installedRole({ slug: "writer", language: "ru", version: 9 })];
+    const view = mapCatalogRoleToView(catalogRole(), roles, "en");
+    expect(view.status).toBe("import");
+    expect(view.installedLang).toBe("ru");
+  });
+
+  it("installed (up to date) -> installed status, catalog version, installedRoleId", () => {
+    const installed = installedRole({
+      slug: "writer",
+      language: "en",
+      version: 3,
+    });
+    const view = mapCatalogRoleToView(catalogRole(), [installed], "en");
+    expect(view).toMatchObject({
+      status: "installed",
+      version: 3,
+      installedRoleId: installed.id,
+    });
+  });
+
+  it("update -> version=from, newVersion=to, installedRoleId", () => {
+    const installed = installedRole({
+      slug: "writer",
+      language: "en",
+      version: 1,
+    });
+    const view = mapCatalogRoleToView(catalogRole(), [installed], "en");
+    expect(view).toMatchObject({
+      status: "update",
+      version: 1,
+      newVersion: 3,
+      installedRoleId: installed.id,
+    });
+  });
+
+  it("missing emoji -> emoji undefined; null description -> empty string", () => {
+    const view = mapCatalogRoleToView(
+      catalogRole({ emoji: null, description: null }),
+      [],
+      "en",
+    );
+    expect(view.emoji).toBeUndefined();
+    expect(view.description).toBe("");
+  });
+});
+
+describe("mapBundleRolesToView", () => {
+  it("maps a bundle's roles preserving order", () => {
+    const roles = [
+      catalogRole({ slug: "a", name: "A", version: 1 }),
+      catalogRole({ slug: "b", name: "B", version: 1 }),
+    ];
+    const installed = [installedRole({ slug: "a", language: "en", version: 1 })];
+    const view = mapBundleRolesToView(roles, installed, "en");
+    expect(view.map((r) => r.slug)).toEqual(["a", "b"]);
+    expect(view[0].status).toBe("installed");
+    expect(view[1].status).toBe("import");
+  });
+});

apps/client/src/features/ai-chat/utils/catalog-bundle-model.ts

@@ -0,0 +1,206 @@
+import type {
+  IAiRole,
+  IAiRoleCatalogRole,
+} from "@/features/ai-chat/types/ai-chat.types.ts";
+import { catalogRoleInstallState } from "@/features/ai-chat/utils/catalog-role-install-state.ts";
+
+/**
+ * The redesigned catalog modal renders bundles as cards with a summary status
+ * (readable without expanding) and a single primary action. The per-role and
+ * per-bundle view model that drives that UI is derived here as PURE functions so
+ * the mapping, the "installed in another language" hint, and the bundle-phase
+ * computation are unit-testable without mounting the component (mirrors the
+ * `catalogRoleInstallState` precedent).
+ */
+
+/**
+ * A role's status in the catalog view model.
+ *  - `import`    — not installed in the current content language.
+ *  - `installed` — installed and up to date.
+ *  - `update`    — installed, but the catalog ships a newer version.
+ *  - `skipped`   — TRANSIENT client-only status set after a conflicted import
+ *                  (a name collision under `conflict:'skip'`); never from the
+ *                  backend.
+ */
+export type RoleStatus = "import" | "installed" | "update" | "skipped";
+
+/** A catalog role mapped into the modal's view model. */
+export interface CatalogViewRole {
+  // Slug is the stable identity within a bundle; used as the row key and as the
+  // `slugs[]` payload for import.
+  slug: string;
+  // Optional in the catalog — the row reserves space and renders nothing when
+  // absent.
+  emoji?: string;
+  name: string;
+  description: string;
+  // For `installed`/`import`: the catalog version. For `update`: the installed
+  // (from) version, with `newVersion` holding the catalog (to) version.
+  version: number;
+  newVersion?: number;
+  status: RoleStatus;
+  // The language a same-slug role is installed under, when it differs from the
+  // current content language (drives the Р5 hint). Only set for `import` roles.
+  installedLang?: string;
+  // The workspace role id, present for `installed`/`update` — needed to call the
+  // update-from-catalog mutation.
+  installedRoleId?: string;
+}
+
+/**
+ * The summary phase of a bundle, derived from its roles' statuses. Determines
+ * the collapsed-header summary and the bundle's single primary action.
+ *  - `empty`        — the bundle has no roles.
+ *  - `allNew`       — everything is importable, nothing installed.
+ *  - `allInstalled` — everything installed & up to date; nothing else pending.
+ *  - `updates`      — updates available and nothing left to import.
+ *  - `mixed`        — any other combination.
+ */
+export type BundlePhase =
+  | "empty"
+  | "allNew"
+  | "allInstalled"
+  | "updates"
+  | "mixed";
+
+/** Per-status tallies for a bundle's roles (the single source of truth). */
+export interface BundleCounts {
+  importable: number;
+  installed: number;
+  update: number;
+  skipped: number;
+}
+
+/**
+ * Count a bundle's roles by status ONCE. Both `bundlePhase` and the panel derive
+ * from this, so the tally logic lives in exactly one place (no rescans / drift).
+ */
+export function bundleCounts(roles: CatalogViewRole[]): BundleCounts {
+  const counts: BundleCounts = {
+    importable: 0,
+    installed: 0,
+    update: 0,
+    skipped: 0,
+  };
+  for (const r of roles) {
+    if (r.status === "import") counts.importable += 1;
+    else if (r.status === "installed") counts.installed += 1;
+    else if (r.status === "update") counts.update += 1;
+    else if (r.status === "skipped") counts.skipped += 1;
+  }
+  return counts;
+}
+
+export function bundlePhase(roles: CatalogViewRole[]): BundlePhase {
+  if (roles.length === 0) return "empty";
+  const { importable, installed, update, skipped } = bundleCounts(roles);
+  // A `skipped` role is a pending post-import conflict (0 installed for it), so a
+  // bundle that has ANY skipped role is NOT "all installed & up to date" — that
+  // would make the collapsed green "up to date" header contradict the open
+  // panel's "Installed 0 · 1 skipped" plaque. It is `mixed` until resolved.
+  if (importable === 0 && update === 0 && skipped === 0) return "allInstalled";
+  if (update > 0 && importable === 0 && skipped === 0) return "updates";
+  if (importable > 0 && installed === 0 && update === 0 && skipped === 0)
+    return "allNew";
+  return "mixed";
+}
+
+/**
+ * The subset of a skip result that should be shown as a TRANSIENT `skipped`
+ * overlay in the bundle (so the row offers a re-import path). Only NAME-CONFLICT
+ * skips qualify: an `already-installed` skip (a concurrent-import race) has
+ * nothing to act on — re-importing the same slug would just skip again — so it
+ * must NOT be overlaid (else the row shows a misleading "Rename & install" that
+ * self-heals into a false "installed"). Pure so both reason branches are tested.
+ */
+export function nameConflictSlugs(
+  skipped: { slug: string; reason: "name-conflict" | "already-installed" }[],
+): string[] {
+  return skipped
+    .filter((s) => s.reason === "name-conflict")
+    .map((s) => s.slug);
+}
+
+/**
+ * Whether a partial-import result should offer the "Rename & install" action:
+ * only when at least one skip is a name conflict (renameable). An
+ * `already-installed`-only partial is informational.
+ */
+export function partialOffersRename(
+  skipped: { reason: "name-conflict" | "already-installed" }[],
+): boolean {
+  return skipped.some((s) => s.reason === "name-conflict");
+}
+
+/**
+ * For a role NOT installed in the current `language`, find a workspace role with
+ * the same catalog `slug` installed under a DIFFERENT language, and return that
+ * language. Drives the "installed in another language" hint (Р5): a different
+ * language of the same slug is a separate install and appears as `import`.
+ */
+export function installedLangForRole(
+  slug: string,
+  workspaceRoles: IAiRole[],
+  language: string,
+): string | undefined {
+  const other = workspaceRoles.find(
+    (r) =>
+      r.source?.slug === slug &&
+      !!r.source?.language &&
+      r.source.language !== language,
+  );
+  return other?.source?.language;
+}
+
+/**
+ * Map one catalog role to the view model, computing its install status against
+ * the workspace roles (via `catalogRoleInstallState`) and, for importable roles,
+ * the other-language hint.
+ */
+export function mapCatalogRoleToView(
+  role: IAiRoleCatalogRole,
+  workspaceRoles: IAiRole[],
+  language: string,
+): CatalogViewRole {
+  const state = catalogRoleInstallState(role, workspaceRoles, language);
+  const base = {
+    slug: role.slug,
+    emoji: role.emoji ?? undefined,
+    name: role.name,
+    description: role.description ?? "",
+  };
+  if (state.state === "update") {
+    return {
+      ...base,
+      status: "update",
+      version: state.fromVersion,
+      newVersion: state.toVersion,
+      installedRoleId: state.installed.id,
+    };
+  }
+  if (state.state === "installed") {
+    return {
+      ...base,
+      status: "installed",
+      version: role.version,
+      installedRoleId: state.installed.id,
+    };
+  }
+  return {
+    ...base,
+    status: "import",
+    version: role.version,
+    installedLang: installedLangForRole(role.slug, workspaceRoles, language),
+  };
+}
+
+/**
+ * Map a whole bundle's catalog roles to the view model, preserving order.
+ */
+export function mapBundleRolesToView(
+  roles: IAiRoleCatalogRole[],
+  workspaceRoles: IAiRole[],
+  language: string,
+): CatalogViewRole[] {
+  return roles.map((r) => mapCatalogRoleToView(r, workspaceRoles, language));
+}

apps/client/src/features/editor/gitmost/gitmost-global-bridge.tsx

@@ -24,6 +24,7 @@ import {
   GitmostListPagesResult,
   GitmostListSpacesResult,
   gitmostDecodePayloadToFile,
+  gitmostInsertTranscriptIntoEditor,
   gitmostUploadFileToEditor,
 } from "@/features/editor/gitmost/gitmost-recording.ts";
 
@@ -281,6 +282,18 @@ export default function GitmostGlobalBridge() {
             pageId: page.id,
           };
         }
+
+        // Best-effort: append the transcript (heading + one paragraph per line)
+        // below the just-inserted audio node. The audio insert already
+        // succeeded, so a transcript failure must NOT turn this into an error —
+        // wrap it and, on any throw, log and still return ok. A missing/empty/
+        // non-string transcript is a no-op inside the helper (audio only).
+        try {
+          gitmostInsertTranscriptIntoEditor(editor, payload?.transcript);
+        } catch (err) {
+          console.error("[gitmost] transcript insert failed", err);
+        }
+
         return { ok: true, pageId: page.id };
       } catch (err: any) {
         console.error("[gitmost] createPageWithRecording failed", err);

apps/client/src/features/editor/gitmost/gitmost-recording.test.ts

@@ -0,0 +1,150 @@
+import { describe, it, expect } from "vitest";
+import { Editor } from "@tiptap/core";
+import { Document } from "@tiptap/extension-document";
+import { Paragraph } from "@tiptap/extension-paragraph";
+import { Text } from "@tiptap/extension-text";
+import { Heading } from "@tiptap/extension-heading";
+import { Bold } from "@tiptap/extension-bold";
+import { Italic } from "@tiptap/extension-italic";
+import { Link } from "@tiptap/extension-link";
+import { gitmostInsertTranscriptIntoEditor } from "./gitmost-recording.ts";
+
+const ZWSP = "​"; // U+200B, the helper's block-trigger neutralizer
+
+/**
+ * #377 — the web-side bridge must append the native host's transcript below the
+ * recording. These exercise the pure insert helper through a REAL Tiptap editor
+ * (Document/Paragraph/Text/Heading + Bold/Italic/Link marks so an HTML-parsing
+ * regression would be caught), asserting the resulting document rather than
+ * mocking the editor: transcript present -> "Transcript" heading + one paragraph
+ * per non-empty line; content is inserted as LITERAL TEXT (no HTML/markdown
+ * parsing); col-0 markdown block triggers are neutralized so git-sync keeps them
+ * paragraphs; absent/empty/non-string -> no-op.
+ */
+describe("gitmostInsertTranscriptIntoEditor", () => {
+  const makeEditor = () =>
+    new Editor({
+      // Bold/Italic/Link are registered specifically so that IF the helper ever
+      // regressed to inserting an HTML/markdown string (instead of a text node),
+      // TipTap would parse `<b>`/`*..*`/`[..](..)` into marks and the literal-
+      // text assertions below would fail.
+      extensions: [Document, Paragraph, Text, Heading, Bold, Italic, Link],
+      // Start from a single empty paragraph (a fresh page's baseline). The
+      // helper appends at the end of the doc, i.e. below existing content.
+      content: { type: "doc", content: [{ type: "paragraph" }] },
+    });
+
+  it("inserts a Transcript heading + one paragraph per non-empty line, verbatim", () => {
+    const editor = makeEditor();
+
+    const inserted = gitmostInsertTranscriptIntoEditor(
+      editor,
+      "You: hello there\nSpeaker 1: hi\n\nYou: bye",
+    );
+
+    expect(inserted).toBe(true);
+
+    const nodes = (editor.getJSON().content ?? []) as any[];
+    // A level-2 "Transcript" heading is present.
+    const heading = nodes.find((n) => n.type === "heading");
+    expect(heading?.attrs?.level).toBe(2);
+    expect(heading?.content?.[0]?.text).toBe("Transcript");
+
+    // Every non-empty transcript line becomes a paragraph, in order, verbatim;
+    // the blank line between them is dropped.
+    const texts = nodes
+      .filter((n) => n.type === "paragraph")
+      .map((n) => n.content?.[0]?.text)
+      .filter((t) => typeof t === "string");
+    expect(texts).toEqual(["You: hello there", "Speaker 1: hi", "You: bye"]);
+
+    editor.destroy();
+  });
+
+  it("inserts HTML + markdown metacharacters as LITERAL text (no injection / no mark parsing)", () => {
+    const editor = makeEditor();
+    const line =
+      "You: <b>bold</b> <script>alert(1)</script> and *stars* and [link](x)";
+
+    const inserted = gitmostInsertTranscriptIntoEditor(editor, line);
+    expect(inserted).toBe(true);
+
+    const paras = (editor.getJSON().content ?? []).filter(
+      (n: any) => n.type === "paragraph",
+    ) as any[];
+    // The transcript line is exactly ONE paragraph holding a SINGLE text node
+    // whose text is the verbatim string — not split into bold/link/other nodes,
+    // not carrying any marks, not raw HTML. This FAILS if the helper switched to
+    // insertContent(htmlString): TipTap would then parse <b>/[link](x)/*stars*.
+    const content = paras[paras.length - 1].content;
+    expect(content).toHaveLength(1);
+    expect(content[0].type).toBe("text");
+    expect(content[0].marks ?? []).toEqual([]);
+    expect(content[0].text).toBe(line);
+    // And no bold/italic/link mark exists anywhere in the document.
+    const html = editor.getHTML();
+    expect(html).not.toMatch(/<(strong|b|em|i|a)\b/);
+    // The angle brackets survived as escaped entities (literal text), not a live
+    // <script>/<b> element.
+    expect(html).not.toMatch(/<script/i);
+
+    editor.destroy();
+  });
+
+  it("neutralizes col-0 markdown block triggers with a leading ZWSP (git-sync safety)", () => {
+    const editor = makeEditor();
+    // Trigger lines (some with a leaked indent) + a normal prefixed line.
+    const inserted = gitmostInsertTranscriptIntoEditor(
+      editor,
+      [
+        "- dash",
+        "  > quote", // leading indent must be trimmed then neutralized
+        "# hash",
+        "1. one",
+        "> [!info] note",
+        "```js",
+        "---", // solid thematic break -> horizontalRule (text-losing) if unneutralized
+        "***",
+        "___",
+        "You: normal line",
+      ].join("\n"),
+    );
+    expect(inserted).toBe(true);
+
+    const texts = (editor.getJSON().content ?? [])
+      .filter((n: any) => n.type === "paragraph")
+      .map((n: any) => n.content?.[0]?.text)
+      .filter((t: any) => typeof t === "string") as string[];
+
+    // Every block-trigger line is prefixed with the invisible ZWSP (indent
+    // trimmed first); the normal `You:` line is left byte-exact.
+    expect(texts).toEqual([
+      ZWSP + "- dash",
+      ZWSP + "> quote",
+      ZWSP + "# hash",
+      ZWSP + "1. one",
+      ZWSP + "> [!info] note",
+      ZWSP + "```js",
+      ZWSP + "---",
+      ZWSP + "***",
+      ZWSP + "___",
+      "You: normal line",
+    ]);
+
+    editor.destroy();
+  });
+
+  it("is a no-op for undefined / empty / whitespace-only / non-string transcripts", () => {
+    for (const value of [undefined, "", "   \n  \n", 42, {}, null]) {
+      const editor = makeEditor();
+      const before = JSON.stringify(editor.getJSON());
+
+      const inserted = gitmostInsertTranscriptIntoEditor(editor, value as any);
+
+      expect(inserted).toBe(false);
+      // Document is untouched (audio-only behavior preserved).
+      expect(JSON.stringify(editor.getJSON())).toBe(before);
+      editor.destroy();
+    }
+  });
+});

apps/client/src/features/editor/gitmost/gitmost-recording.ts

@@ -65,6 +65,11 @@ export interface GitmostCreatePagePayload {
   base64: string;
   filename: string;
   mimeType: string;
+  // Optional transcript for the recording: plain text, `\n`-separated, each
+  // line already formatted as `You: ...` / `Speaker N: ...` by the native host
+  // (ready to insert, no parsing needed). Omitted (no speech / no models) ->
+  // audio only.
+  transcript?: string;
 }
 
 export interface GitmostCreatePageResult {
@@ -235,6 +240,83 @@ export async function gitmostUploadFileToEditor(
   }
 }
 
+// Zero-width space (U+200B). Prepended to a transcript line that begins with a
+// markdown BLOCK trigger: it is invisible in the rendered doc but shifts the
+// trigger off column 0, so the git-sync doc->markdown->doc round-trip keeps the
+// line a plain paragraph (see GITMOST_MD_BLOCK_TRIGGER_RE).
+const GITMOST_ZWSP = "​";
+
+// A markdown BLOCK-level construct that, sitting at column 0 of a paragraph
+// line, the git-sync markdown serializer (packages/prosemirror-markdown
+// markdown-converter.ts, `case "paragraph"`) would re-parse into a NON-paragraph
+// block on the doc->markdown->doc cycle. That serializer emits paragraph text
+// verbatim with NO block-escape (the pre-existing root cause), so a leading
+// `#`/`-`/`*`/`+`/`>`, an ordered-list `N.`/`N)`, a code fence ```/~~~, a table
+// `|`, or a `> [!info]` callout opener would silently become a heading / list /
+// quote / code block / table / callout. The final alternative matches a WHOLE-
+// LINE thematic break — solid `---`/`***`/`___` or spaced `- - -`/`_ _ _` (3+ of
+// the same `-`/`*`/`_`) — which round-trips into a `horizontalRule`; because
+// that node carries NO text, an un-neutralized separator line would LOSE its
+// text entirely (worse than the list/quote case). This matches a TRIMMED line's
+// start; the transcript's own `You:` / `Speaker N:` prefix begins with a letter
+// and never matches, so prefixed lines are left byte-exact.
+const GITMOST_MD_BLOCK_TRIGGER_RE =
+  /^(?:#{1,6}(?:\s|$)|[-*+](?:\s|$)|>|\d+[.)](?:\s|$)|```|~~~|\||([-*_])(?:\s*\1){2,}\s*$)/;
+
+// Append a transcript block BELOW the recording's audio node in a live editor:
+// a "Transcript" heading followed by one paragraph per non-empty transcript
+// line. The transcript is plain text, `\n`-separated, each line already
+// formatted as `You: ...` / `Speaker N: ...` by the native host — line text is
+// inserted as a TEXT node (never HTML/markdown), so there is no injection or
+// mark-parsing surface. Each kept line is trimmed (drops an indent that would
+// both leak into the display and, at col 0, form a markdown block trigger) and,
+// if it still begins with a col-0 markdown block trigger, gets an invisible
+// zero-width space prepended so the git-sync round-trip cannot turn it into a
+// list/quote/heading/callout/code/table (defensive boundary against the
+// serializer's missing block-escape). This is best-effort and meant to run
+// AFTER the audio has already been inserted; the caller must guard against a
+// throw so a transcript failure never fails the (already successful) recording.
+// Returns true when a block was inserted, false when there was nothing to
+// insert (transcript undefined/empty/not-a-string). A non-string value is a
+// no-op, not an error.
+export function gitmostInsertTranscriptIntoEditor(
+  editor: Editor,
+  transcript: unknown,
+): boolean {
+  if (typeof transcript !== "string") return false;
+  const lines = transcript
+    .split("\n")
+    // Trim each line and drop blank (whitespace-only) ones.
+    .map((line) => line.trim())
+    .filter((line) => line.length > 0)
+    // Neutralize a col-0 markdown block trigger with an invisible ZWSP so the
+    // git-sync round-trip keeps the line a paragraph. Host lines (`You:` /
+    // `Speaker N:`) never match and stay byte-exact.
+    .map((line) =>
+      GITMOST_MD_BLOCK_TRIGGER_RE.test(line) ? GITMOST_ZWSP + line : line,
+    );
+  if (lines.length === 0) return false;
+
+  const content = [
+    {
+      type: "heading",
+      attrs: { level: 2 },
+      content: [{ type: "text", text: "Transcript" }],
+    },
+    ...lines.map((line) => ({
+      type: "paragraph",
+      content: [{ type: "text", text: line }],
+    })),
+  ];
+
+  // Append at the end of the document. On a freshly-created recording page the
+  // audio node is the last block, so the end position places the transcript
+  // directly below it.
+  const endPos = editor.state.doc.content.size;
+  editor.chain().focus().insertContentAt(endPos, content).run();
+  return true;
+}
+
 // Full insert path used by the open-page bridge (insertRecording): guard the
 // editor, validate/decode the payload, then upload. Never throws — resolves to
 // a result code.

apps/client/vite.config.ts

@@ -1,5 +1,6 @@
 import { defineConfig, loadEnv } from "vite";
 import react from "@vitejs/plugin-react";
+import { compression } from "vite-plugin-compression2";
 import * as path from "path";
 import { execSync } from "node:child_process";
 
@@ -53,7 +54,25 @@ export default defineConfig(({ mode }) => {
       },
       APP_VERSION: JSON.stringify(resolveAppVersion(envPath)),
     },
-    plugins: [react()],
+    plugins: [
+      react(),
+      // Emit .br and .gz next to every built asset so the server can serve the
+      // precompressed copy (see @fastify/static preCompressed in static.module.ts).
+      compression({
+        algorithms: ["brotliCompress", "gzip"],
+        // vite-plugin-compression2's default `include` only covers text-ish
+        // bundle output (js/mjs/json/css/html/svg/…). Extend it with the large
+        // VAD binaries copied from public/vad (.wasm ~26MB, .onnx ~2.3MB) so
+        // they are brotli/gzip'd once at build time and served via
+        // @fastify/static preCompressed — otherwise @fastify/compress would
+        // re-brotli them on EVERY request. The default types are repeated here
+        // because setting `include` replaces (does not extend) the default.
+        include: /\.(html|xml|css|json|js|mjs|svg|yaml|yml|toml|wasm|onnx)$/,
+        // index.html is rewritten at server boot (window.CONFIG injection); a
+        // precompressed copy would go stale — NEVER precompress it.
+        exclude: [/index\.html$/],
+      }),
+    ],
     build: {
       rolldownOptions: {
         output: {

apps/server/package.json

@@ -23,7 +23,7 @@
     "migration:reset": "tsx src/database/migrate.ts down-to NO_MIGRATIONS",
     "migration:codegen": "kysely-codegen --dialect=postgres --camel-case --env-file=../../.env --out-file=./src/database/types/db.d.ts",
     "lint": "eslint \"{src,apps,libs,test}/**/*.ts\" --fix",
-    "pretest": "pnpm --filter @docmost/editor-ext build",
+    "pretest": "pnpm --filter @docmost/editor-ext build && pnpm --filter @docmost/prosemirror-markdown build",
     "test": "jest",
     "test:int": "jest --config test/jest-integration.json",
     "test:watch": "jest --watch",
@@ -43,6 +43,8 @@
     "@clickhouse/client": "^1.18.2",
     "@docmost/mcp": "workspace:*",
     "@docmost/pdf-inspector": "1.9.6",
+    "@docmost/prosemirror-markdown": "workspace:*",
+    "@fastify/compress": "^9.0.0",
     "@fastify/cookie": "^11.0.2",
     "@fastify/multipart": "^10.0.0",
     "@fastify/static": "^9.1.3",
@@ -175,7 +177,7 @@
       "/node_modules/"
     ],
     "transform": {
-      "happy-dom.+\\.js$": [
+      "(happy-dom.+|prosemirror-markdown/build/.+)\\.js$": [
         "babel-jest",
         {
           "presets": [
@@ -193,7 +195,7 @@
       "^.+\\.(t|j)sx?$": "ts-jest"
     },
     "transformIgnorePatterns": [
-      "/node_modules/(?!(\\.pnpm/)?(nanoid|uuid|image-dimensions|marked|happy-dom|lib0)(@|/))"
+      "/node_modules/(?!(\\.pnpm/)?(nanoid|uuid|image-dimensions|marked|happy-dom|lib0|@docmost/prosemirror-markdown)(@|/))"
     ],
     "collectCoverageFrom": [
       "**/*.(t|j)s"
@@ -204,7 +206,8 @@
       "^@docmost/db/(.*)$": "<rootDir>/database/$1",
       "^@docmost/transactional/(.*)$": "<rootDir>/integrations/transactional/$1",
       "^@docmost/ee/(.*)$": "<rootDir>/ee/$1",
-      "^src/(.*)$": "<rootDir>/$1"
+      "^src/(.*)$": "<rootDir>/$1",
+      "^@tiptap/react$": "<rootDir>/../test/stubs/tiptap-react.js"
     }
   }
 }

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

@@ -43,7 +43,6 @@ import {
   Column,
   Status,
   addUniqueIdsToDoc,
-  htmlToMarkdown,
   TransclusionSource,
   TransclusionReference,
   FootnoteReference,
@@ -51,6 +50,7 @@ import {
   FootnoteDefinition,
   PageEmbed,
 } from '@docmost/editor-ext';
+import { convertProseMirrorToMarkdown } from '@docmost/prosemirror-markdown';
 import { generateText, getSchema, JSONContent } from '@tiptap/core';
 import { generateHTML, generateJSON } from '../common/helpers/prosemirror/html';
 // @tiptap/html library works best for generating prosemirror json state but not HTML
@@ -239,6 +239,10 @@ export function prosemirrorNodeToYElement(node: any): Y.XmlElement | Y.XmlText {
 }
 
 export function jsonToMarkdown(tiptapJson: any): string {
-  const html = jsonToHtml(tiptapJson);
+  // Direct ProseMirror JSON -> Markdown via the canonical converter
-  return htmlToMarkdown(html);
+  // (`@docmost/prosemirror-markdown`) — no HTML intermediate, no second
+  // editor-ext markdown layer. Same serializer as the page/space export and the
+  // git-sync vault writer, so every server PM->MD path emits identical canonical
+  // markdown (issue #345).
+  return convertProseMirrorToMarkdown(tiptapJson);
 }

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

@@ -610,6 +610,63 @@ describe('AiAgentRolesService guards', () => {
       expect(repo.insert.mock.calls[0][0].name).toBe('Researcher (2)');
     });
 
+    it('createdRoles lists the installed role (no renamedTo when not renamed)', async () => {
+      const { service } = makeImportService({});
+      const res = await service.importFromCatalog('ws-1', 'u1', dto());
+      expect(res.createdRoles).toEqual([
+        { slug: 'researcher', name: 'Researcher' },
+      ]);
+      expect(res.skippedRoles).toEqual([]);
+    });
+
+    it('createdRoles carries renamedTo on a rename', async () => {
+      const existing = [makeRow({ id: 'r-x', name: 'Researcher' })];
+      const { service } = makeImportService({ existing });
+      const res = await service.importFromCatalog(
+        'ws-1',
+        'u1',
+        dto({ conflict: 'rename' }),
+      );
+      expect(res.createdRoles).toEqual([
+        { slug: 'researcher', name: 'Researcher', renamedTo: 'Researcher (2)' },
+      ]);
+      expect(res.skippedRoles).toEqual([]);
+    });
+
+    it('skippedRoles: already-installed slug carries reason "already-installed"', async () => {
+      const existing = [
+        makeRow({
+          id: 'r-existing',
+          name: 'Old researcher',
+          source: { slug: 'researcher', language: 'en', version: 1 } as never,
+        }),
+      ];
+      const { service } = makeImportService({ existing });
+      const res = await service.importFromCatalog('ws-1', 'u1', dto());
+      expect(res.skippedRoles).toEqual([
+        {
+          slug: 'researcher',
+          name: 'Researcher',
+          reason: 'already-installed',
+        },
+      ]);
+      expect(res.createdRoles).toEqual([]);
+    });
+
+    it('skippedRoles: a name collision under conflict:skip carries reason "name-conflict"', async () => {
+      const existing = [makeRow({ id: 'r-x', name: 'Researcher' })];
+      const { service } = makeImportService({ existing });
+      const res = await service.importFromCatalog(
+        'ws-1',
+        'u1',
+        dto({ conflict: 'skip' }),
+      );
+      expect(res.skippedRoles).toEqual([
+        { slug: 'researcher', name: 'Researcher', reason: 'name-conflict' },
+      ]);
+      expect(res.createdRoles).toEqual([]);
+    });
+
     it('dto.slugs filters; an unknown slug becomes an error entry', async () => {
       const { service, repo } = makeImportService({
         bundleRoles: [catalogRole()],
@@ -677,6 +734,15 @@ describe('AiAgentRolesService guards', () => {
       // 'a' converged on the concurrent install (skip); 'b' imported; no errors.
       expect(res).toMatchObject({ created: 1, skipped: 1, renamed: 0 });
       expect(res.errors).toEqual([]);
+      // The per-role list records 'a' as an already-installed skip (the UI reads
+      // skippedRoles, not the counter, to render its plaque — assert the array,
+      // not just the count).
+      expect(res.skippedRoles).toContainEqual({
+        slug: 'a',
+        name: 'A',
+        reason: 'already-installed',
+      });
+      expect(res.createdRoles.map((r) => r.slug)).toEqual(['b']);
       // Both inserts were attempted (the batch did not abort on the 23505).
       expect(repo.insert).toHaveBeenCalledTimes(2);
     });

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

@@ -305,6 +305,16 @@ export class AiAgentRolesService {
     skipped: number;
     renamed: number;
     errors: { slug: string; message: string }[];
+    // Per-role lists alongside the counters (kept for back-compat). The redesigned
+    // catalog UI needs the actual roles — which were created (and any rename) and
+    // which were skipped and why — to render an inline result plaque with the
+    // conflicting role's name and a "Rename & install" affordance.
+    createdRoles: { slug: string; name: string; renamedTo?: string }[];
+    skippedRoles: {
+      slug: string;
+      name: string;
+      reason: 'name-conflict' | 'already-installed';
+    }[];
   }> {
     const { file, versions } = await this.loadBundleById(
       dto.bundleId,
@@ -312,6 +322,13 @@ export class AiAgentRolesService {
     );
 
     const errors: { slug: string; message: string }[] = [];
+    const createdRoles: { slug: string; name: string; renamedTo?: string }[] =
+      [];
+    const skippedRoles: {
+      slug: string;
+      name: string;
+      reason: 'name-conflict' | 'already-installed';
+    }[] = [];
 
     // Resolve the selected catalog roles (honor dto.slugs; flag unknown ones).
     let selected = file.roles;
@@ -351,16 +368,27 @@ export class AiAgentRolesService {
       // Already installed from the catalog in THIS language => skip (use
       // update-from-catalog). A different language of the same slug still imports.
       const installKey = `${role.slug}:${dto.language}`;
+      const originalName = role.name.trim();
       if (installedKeys.has(installKey)) {
         skipped++;
+        skippedRoles.push({
+          slug: role.slug,
+          name: originalName,
+          reason: 'already-installed',
+        });
         continue;
       }
 
-      let name = role.name.trim();
+      let name = originalName;
       let didRename = false;
       if (takenNames.has(name.toLowerCase())) {
         if (dto.conflict === 'skip') {
           skipped++;
+          skippedRoles.push({
+            slug: role.slug,
+            name: originalName,
+            reason: 'name-conflict',
+          });
           continue;
         }
         // conflict === 'rename': find a free " (N)" suffix.
@@ -380,6 +408,11 @@ export class AiAgentRolesService {
         });
         created++;
         if (didRename) renamed++;
+        createdRoles.push({
+          slug: role.slug,
+          name: originalName,
+          ...(didRename ? { renamedTo: name } : {}),
+        });
         takenNames.add(name.toLowerCase());
         installedKeys.add(installKey);
       } catch (err) {
@@ -391,6 +424,11 @@ export class AiAgentRolesService {
         // skipped (already installed) and continue; do NOT abort or error.
         if (isSourceUniqueViolation(err)) {
           skipped++;
+          skippedRoles.push({
+            slug: role.slug,
+            name: originalName,
+            reason: 'already-installed',
+          });
           installedKeys.add(installKey);
           continue;
         }
@@ -407,7 +445,7 @@ export class AiAgentRolesService {
       }
     }
 
-    return { created, skipped, renamed, errors };
+    return { created, skipped, renamed, errors, createdRoles, skippedRoles };
   }
 
   /**

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

@@ -539,3 +539,115 @@ describe('AiChatToolsService model-friendly input validation (#190)', () => {
     expect(result.error?.message).toContain('parameter "pageId": missing (required)');
   });
 });
+
+/**
+ * #294 F1 — the contract-parity test introspects only the ADVERTISED schema keys
+ * (buildShape), not the execute bodies. Most execs are unchanged pass-throughs,
+ * but two wirings actually CHANGED in the migration and are otherwise untested:
+ *   - movePage now forwards the newly-added optional `position` field to the
+ *     client (client.movePage(pageId, parentPageId, position));
+ *   - the table trio unified its `tableRef` param to `table` and must forward it
+ *     positionally. A field destructured under the wrong name would silently pass
+ *     `undefined` to the client (execute is `any`-cast, so tsc won't catch it).
+ */
+describe('AiChatToolsService #294 changed execute wirings', () => {
+  const calls: Record<string, unknown[][]> = {
+    movePage: [],
+    tableInsertRow: [],
+    tableDeleteRow: [],
+    tableUpdateCell: [],
+  };
+  const fakeClient: Partial<DocmostClientLike> = {
+    movePage: (...args: unknown[]) => {
+      calls.movePage.push(args);
+      return Promise.resolve({ success: true });
+    },
+    tableInsertRow: (...args: unknown[]) => {
+      calls.tableInsertRow.push(args);
+      return Promise.resolve({ ok: true });
+    },
+    tableDeleteRow: (...args: unknown[]) => {
+      calls.tableDeleteRow.push(args);
+      return Promise.resolve({ ok: true });
+    },
+    tableUpdateCell: (...args: unknown[]) => {
+      calls.tableUpdateCell.push(args);
+      return Promise.resolve({ ok: true });
+    },
+  };
+  const tokenServiceStub = {
+    generateAccessToken: jest.fn().mockResolvedValue('access-token'),
+    generateCollabToken: jest.fn().mockResolvedValue('collab-token'),
+  };
+  let service: AiChatToolsService;
+
+  beforeEach(() => {
+    for (const k of Object.keys(calls)) calls[k].length = 0;
+    jest.spyOn(loader, 'loadDocmostMcp').mockResolvedValue(
+      mockLoaded(function () {
+        return fakeClient as DocmostClientLike;
+      } as unknown as loader.DocmostClientCtor),
+    );
+    service = new AiChatToolsService(
+      tokenServiceStub as never,
+      {} as never,
+      {} as never,
+      {} as never,
+      {} as never,
+      {
+        asSink: () => ({ put: jest.fn(), has: jest.fn(), evict: jest.fn() }),
+      } as never,
+    );
+  });
+  afterEach(() => jest.restoreAllMocks());
+
+  const buildTools = () =>
+    service.forUser(
+      { id: 'user-1', email: 'u@example.com', workspaceId: 'ws-1' } as never,
+      'session-1',
+      'ws-1',
+      'chat-1',
+    );
+
+  it('movePage forwards the optional position to the client', async () => {
+    const tools = await buildTools();
+    await tools.movePage.execute(
+      { pageId: 'p1', parentPageId: 'parent1', position: 'a5' } as never,
+      {} as never,
+    );
+    expect(calls.movePage).toEqual([['p1', 'parent1', 'a5']]);
+  });
+
+  it('movePage passes undefined position and null parent when omitted (unchanged behavior)', async () => {
+    const tools = await buildTools();
+    await tools.movePage.execute({ pageId: 'p2' } as never, {} as never);
+    expect(calls.movePage).toEqual([['p2', null, undefined]]);
+  });
+
+  it('tableInsertRow forwards the unified `table` param positionally', async () => {
+    const tools = await buildTools();
+    await tools.tableInsertRow.execute(
+      { pageId: 'p1', table: '#0', cells: ['a', 'b'], index: 2 } as never,
+      {} as never,
+    );
+    expect(calls.tableInsertRow).toEqual([['p1', '#0', ['a', 'b'], 2]]);
+  });
+
+  it('tableDeleteRow forwards `table` positionally', async () => {
+    const tools = await buildTools();
+    await tools.tableDeleteRow.execute(
+      { pageId: 'p1', table: '#0', index: 1 } as never,
+      {} as never,
+    );
+    expect(calls.tableDeleteRow).toEqual([['p1', '#0', 1]]);
+  });
+
+  it('tableUpdateCell forwards `table` positionally', async () => {
+    const tools = await buildTools();
+    await tools.tableUpdateCell.execute(
+      { pageId: 'p1', table: '#0', row: 1, col: 2, text: 'x' } as never,
+      {} as never,
+    );
+    expect(calls.tableUpdateCell).toEqual([['p1', '#0', 1, 2, 'x']]);
+  });
+});

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

@@ -316,50 +316,27 @@ export class AiChatToolsService {
         execute: async () => resolveCurrentPageResult(openedPage),
       }),
 
-      getPage: tool({
+      // Schema + description now live in @docmost/mcp's SHARED_TOOL_SPECS (#294).
-        description:
+      // The execute body keeps this layer's { title, markdown } projection.
-          'Fetch a single page as Markdown by its page id. Returns the page ' +
+      getPage: sharedTool(sharedToolSpecs.getPage, async ({ pageId }) => {
-          'title and its Markdown content. Inline <span data-comment-id> tags ' +
+        // getPage(pageId) -> { data: filterPage(page, markdown), success }.
-          'in the markdown are comment highlight anchors (also present for ' +
+        const result = await client.getPage(pageId);
-          'RESOLVED threads) — treat them as markup, not page text.',
+        const data = (result?.data ?? {}) as {
-        inputSchema: modelFriendlyInput({
+          title?: string;
-          pageId: z.string().describe('The id (or slugId) of the page.'),
+          content?: string;
-        }),
+        };
-        execute: async ({ pageId }) => {
+        return {
-          // getPage(pageId) -> { data: filterPage(page, markdown), success }.
+          title: data.title ?? '',
-          const result = await client.getPage(pageId);
+          markdown: typeof data.content === 'string' ? data.content : '',
-          const data = (result?.data ?? {}) as {
+        };
-            title?: string;
-            content?: string;
-          };
-          return {
-            title: data.title ?? '',
-            markdown: typeof data.content === 'string' ? data.content : '',
-          };
-        },
       }),
 
       // --- WRITE tools (all reversible — history/trash; §6.5 / D3) ---
 
-      createPage: tool({
+      // Schema + description now live in @docmost/mcp's SHARED_TOOL_SPECS (#294).
-        description:
+      createPage: sharedTool(
-          'Create a new page with a Markdown body in a space, optionally under ' +
+        sharedToolSpecs.createPage,
-          'a parent page. Returns the new page id and title. Reversible: a page ' +
+        async ({ title, content, spaceId, parentPageId }) => {
-          'can be moved to trash later.',
-        inputSchema: modelFriendlyInput({
-          title: z.string().describe('The title of the new page.'),
-          content: z
-            .string()
-            .describe('The page body as Markdown (may be empty).'),
-          spaceId: z
-            .string()
-            .describe('The id of the space to create the page in.'),
-          parentPageId: z
-            .string()
-            .optional()
-            .describe('Optional parent page id to nest the new page under.'),
-        }),
-        execute: async ({ title, content, spaceId, parentPageId }) => {
           // createPage(title, content, spaceId, parentPageId?) ->
           // { data: filterPage(page, markdown), success }.
           const result = await client.createPage(
@@ -375,7 +352,7 @@ export class AiChatToolsService {
           };
           return { id: data.id ?? data.slugId, title: data.title ?? title };
         },
-      }),
+      ),
 
       updatePageContent: tool({
         description:
@@ -399,115 +376,46 @@ export class AiChatToolsService {
         },
       }),
 
-      renamePage: tool({
+      // Schema + description now live in @docmost/mcp's SHARED_TOOL_SPECS (#294).
-        description:
+      renamePage: sharedTool(
-          "Rename a page (change its title only; the body is untouched). " +
+        sharedToolSpecs.renamePage,
-          'Reversible: rename back at any time.',
+        async ({ pageId, title }) => {
-        inputSchema: modelFriendlyInput({
-          pageId: z.string().describe('The id of the page to rename.'),
-          title: z.string().describe('The new title.'),
-        }),
-        execute: async ({ pageId, title }) => {
           // renamePage(pageId, title) -> { success, pageId, title }.
           await client.renamePage(pageId, title);
           return { pageId, title };
         },
-      }),
+      ),
 
-      movePage: tool({
+      // Schema + description now live in @docmost/mcp's SHARED_TOOL_SPECS (#294).
-        description:
+      // The shared schema adds the optional `position` field this layer lacked
-          'Move a page under a new parent page, or to the space root when no ' +
+      // before; the execute now forwards it (the client already accepted it).
-          'parent is given. Reversible: move it back at any time.',
+      movePage: sharedTool(
-        inputSchema: modelFriendlyInput({
+        sharedToolSpecs.movePage,
-          pageId: z.string().describe('The id of the page to move.'),
+        async ({ pageId, parentPageId, position }) => {
-          parentPageId: z
-            .string()
-            .nullable()
-            .optional()
-            .describe(
-              'Target parent page id. Null/omitted moves the page to the ' +
-                'space root.',
-            ),
-        }),
-        execute: async ({ pageId, parentPageId }) => {
           // movePage(pageId, parentPageId, position?) -> raw move response.
-          await client.movePage(pageId, parentPageId ?? null);
+          await client.movePage(pageId, parentPageId ?? null, position);
           return { pageId, parentPageId: parentPageId ?? null, moved: true };
         },
+      ),
+
+      // Schema + description now live in @docmost/mcp's SHARED_TOOL_SPECS (#294).
+      // GUARDRAIL (§14 H4) preserved: the shared schema exposes ONLY pageId, so
+      // permanentlyDelete/forceDelete are never part of the input and can never
+      // be forwarded — the agent physically cannot permanently delete a page.
+      deletePage: sharedTool(sharedToolSpecs.deletePage, async ({ pageId }) => {
+        // deletePage(pageId) hits POST /pages/delete with { pageId } only,
+        // which is the soft-delete (trash) path on the server.
+        await client.deletePage(pageId);
+        return { pageId, trashed: true };
       }),
 
-      deletePage: tool({
+      // Schema + description now live in @docmost/mcp's SHARED_TOOL_SPECS (#294).
-        description:
+      // This layer keeps only its own execute-side guards (require a selection
-          'Move a page to the trash (SOFT delete only — fully reversible; the ' +
+      // for a top-level comment; reject suggestedText on a reply / without a
-          'page can be restored from trash). This NEVER permanently deletes.',
+      // selection) — the schema+description are shared.
-        inputSchema: modelFriendlyInput({
+      createComment: sharedTool(
-          pageId: z.string().describe('The id of the page to move to trash.'),
+        sharedToolSpecs.createComment,
-        }),
+        async ({
-        // GUARDRAIL (§14 H4): the only field ever passed to the client is
-        // pageId. permanentlyDelete/forceDelete are not part of the schema and
-        // are never forwarded, so the agent physically cannot permanently
-        // delete a page through this tool.
-        execute: async ({ pageId }) => {
-          // deletePage(pageId) hits POST /pages/delete with { pageId } only,
-          // which is the soft-delete (trash) path on the server.
-          await client.deletePage(pageId);
-          return { pageId, trashed: true };
-        },
-      }),
-
-      // INTENTIONAL per-transport divergence (not shared): the description is
-      // tuned for the in-app agent (e.g. "retry with a corrected EXACT selection"
-      // and "Reversible via the comment UI"); the standalone MCP `create_comment`
-      // keeps its own wording. Kept per-layer.
-      createComment: tool({
-        description:
-          'Add an INLINE comment to a page, or reply to an existing top-level ' +
-          'comment (one level only — the backend rejects replies to replies). ' +
-          'The comment is anchored inline to the given exact `selection` text ' +
-          '(which gets highlighted); page-level comments are NOT supported. A ' +
-          "new top-level comment REQUIRES a `selection`. Replies inherit the " +
-          "parent's anchor and take no selection. If the call fails with a " +
-          '"selection not found" error, retry with a corrected EXACT selection ' +
-          'copied verbatim from a single paragraph/block. You may also attach a ' +
-          '`suggestedText` proposing a replacement for the `selection` (a human ' +
-          'applies it from the UI); when set, the `selection` must occur exactly ' +
-          'once in the page. Reversible via the comment UI.',
-        inputSchema: modelFriendlyInput({
-          pageId: z.string().describe('The id of the page to comment on.'),
-          content: z.string().describe('The comment body as Markdown.'),
-          selection: z
-            .string()
-            .min(1)
-            .max(250)
-            .optional()
-            .describe(
-              'EXACT contiguous text from a SINGLE paragraph/block to anchor ' +
-                '(highlight) the comment on (<=250 chars, avoid spanning across ' +
-                'formatting boundaries). Required for a new top-level comment; ' +
-                'omit only when replying via parentCommentId.',
-            ),
-          parentCommentId: z
-            .string()
-            .optional()
-            .describe(
-              'Optional id of a TOP-LEVEL comment to reply to (one level ' +
-                'of replies only).',
-            ),
-          suggestedText: z
-            .string()
-            .min(1)
-            .max(2000)
-            .optional()
-            .describe(
-              'Optional proposed replacement (PLAIN TEXT) for the `selection`, ' +
-                'applied by a human via the UI (never auto-applied). REQUIRES a ' +
-                '`selection`; NOT allowed on a reply. When set, the `selection` ' +
-                'must be UNIQUE in the page — expand it with surrounding context ' +
-                '(still <=250 chars) if it occurs more than once, or the call is ' +
-                'refused.',
-            ),
-        }),
-        execute: async ({
           pageId,
           content,
           selection,
@@ -548,26 +456,17 @@ export class AiChatToolsService {
           const data = (result?.data ?? {}) as { id?: string };
           return { commentId: data.id, pageId };
         },
-      }),
+      ),
 
-      resolveComment: tool({
+      // Schema + description now live in @docmost/mcp's SHARED_TOOL_SPECS (#294).
-        description:
+      resolveComment: sharedTool(
-          'Resolve or reopen a top-level comment thread (reversible — toggle ' +
+        sharedToolSpecs.resolveComment,
-          'the resolved flag). Only top-level comments can be resolved.',
+        async ({ commentId, resolved }) => {
-        inputSchema: modelFriendlyInput({
-          commentId: z
-            .string()
-            .describe('The id of the top-level comment to resolve/reopen.'),
-          resolved: z
-            .boolean()
-            .describe('true to resolve the thread, false to reopen it.'),
-        }),
-        execute: async ({ commentId, resolved }) => {
           // resolveComment(commentId, resolved) -> { success, commentId, resolved }.
           await client.resolveComment(commentId, resolved);
           return { commentId, resolved };
         },
-      }),
+      ),
 
       // --- READ tools (added) ---
 
@@ -585,33 +484,12 @@ export class AiChatToolsService {
       // hierarchy mode but is worded for the in-app agent; the standalone MCP
       // `list_pages` carries its own wording. Kept per-layer so each side tunes
       // its own guidance.
-      listPages: tool({
+      // Schema + description now live in @docmost/mcp's SHARED_TOOL_SPECS (#294).
-        description:
+      listPages: sharedTool(
-          'List the most recent pages, optionally scoped to a single space. ' +
+        sharedToolSpecs.listPages,
-          'Returns a bounded list (default 50, max 100). Pass tree:true (with ' +
+        async ({ spaceId, limit, tree }) =>
-          "spaceId) to instead get the space's full page hierarchy as a nested tree.",
-        inputSchema: modelFriendlyInput({
-          spaceId: z
-            .string()
-            .optional()
-            .describe('Optional space id to scope the listing to.'),
-          limit: z
-            .number()
-            .int()
-            .min(1)
-            .max(100)
-            .optional()
-            .describe('Maximum number of pages (1-100).'),
-          tree: z
-            .boolean()
-            .optional()
-            .describe(
-              'When true, return the full page hierarchy of the given space as a nested tree (children arrays) instead of the recent-pages flat list. Requires spaceId; ignores limit.',
-            ),
-        }),
-        execute: async ({ spaceId, limit, tree }) =>
           await client.listPages(spaceId, limit, tree),
-      }),
+      ),
 
       listSidebarPages: tool({
         description:
@@ -656,41 +534,34 @@ export class AiChatToolsService {
           }),
       ),
 
+      // NOT shared (kept inline): the MCP tool name `table_get` is noun-first
+      // while this key is `getTable` (verb-first), breaking the
+      // snake_case(inAppKey) convention the shared registry enforces. Its
+      // reference parameter is still named `table` (was `tableRef`) so it matches
+      // the migrated table row/cell tools below.
       getTable: tool({
         description:
           'Read a table as a matrix of cell texts (plus a parallel cellIds ' +
           'matrix so cells can be addressed for rich edits).',
         inputSchema: modelFriendlyInput({
           pageId: z.string().describe('The id of the page.'),
-          tableRef: z
+          table: z
             .string()
             .describe(
-              '"#<index>" from getOutline, or a block id of any node inside ' +
+              '"#<index>" from the page outline, or a block id of any node ' +
-                'the table.',
+                'inside the table.',
             ),
         }),
-        execute: async ({ pageId, tableRef }) =>
+        execute: async ({ pageId, table }) =>
-          await client.getTable(pageId, tableRef),
+          await client.getTable(pageId, table),
       }),
 
-      listComments: tool({
+      // Schema + description now live in @docmost/mcp's SHARED_TOOL_SPECS (#294).
-        description:
+      listComments: sharedTool(
-          'List comments on a page in one call. By DEFAULT only ACTIVE ' +
+        sharedToolSpecs.listComments,
-          'threads are returned; resolved threads (a resolved top-level ' +
+        async ({ pageId, includeResolved }) =>
-          'comment and all its replies) are hidden and their count reported ' +
-          'as `resolvedThreadsHidden` so you can re-query with ' +
-          '`includeResolved: true` to see everything. Returns ' +
-          '`{ items, resolvedThreadsHidden }`. Content is returned as Markdown.',
-        inputSchema: modelFriendlyInput({
-          pageId: z.string().describe('The id of the page.'),
-          includeResolved: z
-            .boolean()
-            .optional()
-            .describe('default only active threads; true — include resolved'),
-        }),
-        execute: async ({ pageId, includeResolved }) =>
           await client.listComments(pageId, includeResolved),
-      }),
+      ),
 
       getComment: tool({
         description: 'Fetch a single comment by id (content as Markdown).',
@@ -700,26 +571,12 @@ export class AiChatToolsService {
         execute: async ({ commentId }) => await client.getComment(commentId),
       }),
 
-      checkNewComments: tool({
+      // Schema + description now live in @docmost/mcp's SHARED_TOOL_SPECS (#294).
-        description:
+      checkNewComments: sharedTool(
-          'Find new comments across a space (optionally scoped to a subtree) ' +
+        sharedToolSpecs.checkNewComments,
-          'created after a given timestamp.',
+        async ({ spaceId, since, parentPageId }) =>
-        inputSchema: modelFriendlyInput({
-          spaceId: z.string().describe('The id of the space to scan.'),
-          since: z
-            .string()
-            .describe('An ISO-8601 timestamp; only comments created after it.'),
-          parentPageId: z
-            .string()
-            .optional()
-            .describe(
-              'Optional page id to scope the scan to that page and its ' +
-                'descendants.',
-            ),
-        }),
-        execute: async ({ spaceId, since, parentPageId }) =>
           await client.checkNewComments(spaceId, since, parentPageId),
-      }),
+      ),
 
       listShares: sharedTool(
         sharedToolSpecs.listShares,
@@ -749,19 +606,14 @@ export class AiChatToolsService {
           await client.diffPageVersions(pageId, from, to),
       ),
 
-      exportPageMarkdown: tool({
+      // Schema + description now live in @docmost/mcp's SHARED_TOOL_SPECS (#294).
-        description:
+      exportPageMarkdown: sharedTool(
-          'Export a page to a single self-contained Docmost-flavoured ' +
+        sharedToolSpecs.exportPageMarkdown,
-          'Markdown file (meta + body + comment threads). Lossless round-trip ' +
+        async ({ pageId }) => {
-          'with importPageMarkdown.',
-        inputSchema: modelFriendlyInput({
-          pageId: z.string().describe('The id of the page to export.'),
-        }),
-        execute: async ({ pageId }) => {
           const markdown = await client.exportPageMarkdown(pageId);
           return { markdown };
         },
-      }),
+      ),
 
       // --- WRITE tools (added; reversible via page history/trash) ---
 
@@ -811,28 +663,12 @@ export class AiChatToolsService {
         async ({ pageId, nodeId }) => await client.deleteNode(pageId, nodeId),
       ),
 
-      updatePageJson: tool({
+      // Schema + description now live in @docmost/mcp's SHARED_TOOL_SPECS (#294).
-        description:
+      // The execute body keeps this layer's content normalization (parity with
-          "Replace a page's body with a full ProseMirror document — a full " +
+      // the standalone MCP server, index.ts update_page_json).
-          'overwrite — and/or update its title. Minimal example content: ' +
+      updatePageJson: sharedTool(
-          '{"type":"doc","content":[{"type":"paragraph","content":' +
+        sharedToolSpecs.updatePageJson,
-          '[{"type":"text","text":"Hi"}]}]}. The content arg may be a JSON ' +
+        async ({ pageId, content, title }) => {
-          'object or a JSON string (both accepted). Omit content for a ' +
-          'title-only update. Reversible: the previous version is kept in page ' +
-          'history.',
-        inputSchema: modelFriendlyInput({
-          pageId: z.string().describe('The id of the page to update.'),
-          content: z
-            .any()
-            .optional()
-            .describe(
-              'Full ProseMirror doc {"type":"doc","content":[...]} (JSON ' +
-                'object or JSON string); omit for a title-only update.',
-            ),
-          title: z.string().optional().describe('Optional new title.'),
-        }),
-        execute: async ({ pageId, content, title }) => {
-          // Parity with the standalone MCP server (index.ts update_page_json):
           // undefined/null pass through as undefined (title-only / no-op); any
           // string is JSON.parsed (so an empty string "" throws, matching the
           // MCP server); an object is passed through unchanged.
@@ -845,66 +681,29 @@ export class AiChatToolsService {
           }
           return await client.updatePageJson(pageId, doc, title);
         },
-      }),
+      ),
 
-      // NOT in the shared registry: this layer names the table argument
+      // Schema + description now live in @docmost/mcp's SHARED_TOOL_SPECS (#294).
-      // `tableRef`, while the standalone MCP tool names it `table` (index.ts).
+      // The table reference parameter was unified to `table` (was `tableRef`).
-      // Sharing one buildShape would rename a model-facing parameter on one
+      tableInsertRow: sharedTool(
-      // transport, so the table row/cell tools stay per-layer by design.
+        sharedToolSpecs.tableInsertRow,
-      tableInsertRow: tool({
+        async ({ pageId, table, cells, index }) =>
-        description:
+          await client.tableInsertRow(pageId, table, cells, index),
-          'Insert a row of plain-text cells into a table. Reversible via ' +
+      ),
-          'page history.',
-        inputSchema: modelFriendlyInput({
-          pageId: z.string().describe('The id of the page.'),
-          tableRef: z
-            .string()
-            .describe('"#<index>" from getOutline, or a block id in the table.'),
-          cells: z.array(z.string()).describe('The cell texts for the row.'),
-          index: z
-            .number()
-            .int()
-            .optional()
-            .describe('0-based insert position (omit/out-of-range to append).'),
-        }),
-        execute: async ({ pageId, tableRef, cells, index }) =>
-          await client.tableInsertRow(pageId, tableRef, cells, index),
-      }),
 
-      // NOT shared — same `tableRef` (here) vs `table` (MCP) parameter-name
+      // Schema + description now live in @docmost/mcp's SHARED_TOOL_SPECS (#294).
-      // divergence as tableInsertRow.
+      tableDeleteRow: sharedTool(
-      tableDeleteRow: tool({
+        sharedToolSpecs.tableDeleteRow,
-        description:
+        async ({ pageId, table, index }) =>
-          'Delete a table row at a 0-based index. Reversible via page history.',
+          await client.tableDeleteRow(pageId, table, index),
-        inputSchema: modelFriendlyInput({
+      ),
-          pageId: z.string().describe('The id of the page.'),
-          tableRef: z
-            .string()
-            .describe('"#<index>" from getOutline, or a block id in the table.'),
-          index: z.number().int().describe('0-based row index to delete.'),
-        }),
-        execute: async ({ pageId, tableRef, index }) =>
-          await client.tableDeleteRow(pageId, tableRef, index),
-      }),
 
-      // NOT shared — same `tableRef` (here) vs `table` (MCP) parameter-name
+      // Schema + description now live in @docmost/mcp's SHARED_TOOL_SPECS (#294).
-      // divergence as tableInsertRow.
+      tableUpdateCell: sharedTool(
-      tableUpdateCell: tool({
+        sharedToolSpecs.tableUpdateCell,
-        description:
+        async ({ pageId, table, row, col, text }) =>
-          'Set the plain-text content of a table cell at [row, col] (0-based). ' +
+          await client.tableUpdateCell(pageId, table, row, col, text),
-          'Reversible via page history.',
+      ),
-        inputSchema: modelFriendlyInput({
-          pageId: z.string().describe('The id of the page.'),
-          tableRef: z
-            .string()
-            .describe('"#<index>" from getOutline, or a block id in the table.'),
-          row: z.number().int().describe('0-based row index.'),
-          col: z.number().int().describe('0-based column index.'),
-          text: z.string().describe('The new cell text.'),
-        }),
-        execute: async ({ pageId, tableRef, row, col, text }) =>
-          await client.tableUpdateCell(pageId, tableRef, row, col, text),
-      }),
 
       copyPageContent: sharedTool(
         sharedToolSpecs.copyPageContent,
@@ -918,25 +717,14 @@ export class AiChatToolsService {
           await client.importPageMarkdown(pageId, markdown),
       ),
 
-      // INTENTIONAL per-transport divergence (not shared): adds a security
+      // Schema + description now live in @docmost/mcp's SHARED_TOOL_SPECS (#294).
-      // confirmation framing ("Only share when the user explicitly asked, since
+      // Both layers already carried the security-confirmation framing, so there
-      // this exposes the page to anyone with the link") for the in-app agent; the
+      // was no real divergence to preserve — only wording drift.
-      // standalone MCP `share_page` keeps the plain public-URL wording.
+      sharePage: sharedTool(
-      sharePage: tool({
+        sharedToolSpecs.sharePage,
-        description:
+        async ({ pageId, searchIndexing }) =>
-          'Make a page PUBLICLY accessible and return its public URL. ' +
-          'Reversible via unsharePage. Only share when the user explicitly ' +
-          'asked, since this exposes the page to anyone with the link.',
-        inputSchema: modelFriendlyInput({
-          pageId: z.string().describe('The id of the page to share.'),
-          searchIndexing: z
-            .boolean()
-            .optional()
-            .describe('Allow public search engines to index it (default true).'),
-        }),
-        execute: async ({ pageId, searchIndexing }) =>
           await client.sharePage(pageId, searchIndexing),
-      }),
+      ),
 
       unsharePage: sharedTool(
         sharedToolSpecs.unsharePage,

apps/server/src/core/ai-chat/tools/tool-tiers.ts

@@ -100,54 +100,26 @@ export const INLINE_TOOL_TIERS: Record<
     tier: 'core',
     catalogLine: 'getCurrentPage — the page the user is currently viewing.',
   },
-  getPage: {
+  // NOTE: getPage and listPages moved to @docmost/mcp's SHARED_TOOL_SPECS
-    tier: 'core',
+  // (#294); they carry their own tier ('core') + catalogLine there.
-    catalogLine: 'getPage — fetch a page as Markdown by its id.',
+  // NOTE: createComment, listComments and resolveComment moved to
-  },
+  // @docmost/mcp's SHARED_TOOL_SPECS (#294); they carry their own tier +
-  listPages: {
+  // catalogLine there. getComment stays inline (MCP-only shape divergence is
-    tier: 'core',
+  // n/a — it simply has no shared spec).
-    catalogLine: "listPages — list recent pages, or a space's full page tree.",
-  },
-  listComments: {
-    tier: 'core',
-    catalogLine: 'listComments — list all comments on a page (including resolved).',
-  },
   getComment: {
     tier: 'core',
     catalogLine: 'getComment — fetch a single comment by id.',
   },
-  createComment: {
-    tier: 'core',
-    catalogLine:
-      'createComment — add an inline comment (optionally with a suggested edit).',
-  },
-  resolveComment: {
-    tier: 'core',
-    catalogLine: 'resolveComment — resolve or reopen a comment thread.',
-  },
 
   // --- deferred inline ---
-  createPage: {
+  // NOTE: createPage, renamePage, movePage, deletePage, updatePageJson and
-    tier: 'deferred',
+  // exportPageMarkdown moved to @docmost/mcp's SHARED_TOOL_SPECS (#294); they
-    catalogLine: 'createPage — create a new page with a Markdown body in a space.',
+  // carry their own deferred tier + catalogLine there.
-  },
   updatePageContent: {
     tier: 'deferred',
     catalogLine:
       "updatePageContent — replace a page's body (and optionally title) with new Markdown.",
   },
-  renamePage: {
-    tier: 'deferred',
-    catalogLine: "renamePage — change a page's title only (body untouched).",
-  },
-  movePage: {
-    tier: 'deferred',
-    catalogLine: 'movePage — move a page under a new parent or to the space root.',
-  },
-  deletePage: {
-    tier: 'deferred',
-    catalogLine: 'deletePage — move a page to trash (soft delete, reversible).',
-  },
   listSidebarPages: {
     tier: 'deferred',
     catalogLine:
@@ -157,42 +129,21 @@ export const INLINE_TOOL_TIERS: Record<
     tier: 'deferred',
     catalogLine: 'getTable — read a table as a matrix of cell texts and cell ids.',
   },
-  checkNewComments: {
+  // NOTE: tableInsertRow, tableDeleteRow and tableUpdateCell moved to
-    tier: 'deferred',
+  // @docmost/mcp's SHARED_TOOL_SPECS (#294); they carry their own deferred tier +
-    catalogLine:
+  // catalogLine there. getTable stays inline (its MCP name table_get breaks the
-      'checkNewComments — find comments in a space created after a timestamp.',
+  // snake_case(inAppKey) convention, so it has no shared spec).
-  },
+  // NOTE: checkNewComments moved to @docmost/mcp's SHARED_TOOL_SPECS (#294);
+  // it carries its own deferred tier + catalogLine there.
   getPageHistory: {
     tier: 'deferred',
     catalogLine:
       'getPageHistory — fetch one page-history version with its ProseMirror content.',
   },
-  exportPageMarkdown: {
+  // NOTE: sharePage moved to @docmost/mcp's SHARED_TOOL_SPECS (#294); it carries
-    tier: 'deferred',
+  // its own deferred tier + catalogLine there. transformPage stays inline (its
-    catalogLine:
+  // schema deliberately diverges — it omits the deleteComments field the MCP
-      'exportPageMarkdown — export a page to self-contained Markdown (body + comments).',
+  // docmost_transform exposes, a comment-deletion guardrail).
-  },
-  updatePageJson: {
-    tier: 'deferred',
-    catalogLine:
-      "updatePageJson — overwrite a page's body with a full ProseMirror document.",
-  },
-  tableInsertRow: {
-    tier: 'deferred',
-    catalogLine: 'tableInsertRow — insert a row of plain-text cells into a table.',
-  },
-  tableDeleteRow: {
-    tier: 'deferred',
-    catalogLine: 'tableDeleteRow — delete a table row at a 0-based index.',
-  },
-  tableUpdateCell: {
-    tier: 'deferred',
-    catalogLine: 'tableUpdateCell — set the text of a table cell at [row, col].',
-  },
-  sharePage: {
-    tier: 'deferred',
-    catalogLine: 'sharePage — make a page publicly accessible and return its URL.',
-  },
   transformPage: {
     tier: 'deferred',
     catalogLine: "transformPage — run a sandboxed JS transform over a page's document.",

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

@@ -474,6 +474,19 @@ export class AttachmentController {
     const fileSize = Number(attachment.fileSize);
     const rangeHeader = req.headers.range;
 
+    // Opt this download route out of the global @fastify/compress hook.
+    // Attachment bytes are final and mostly binary, so on-the-fly compression
+    // only burns CPU — and on the 206/Range branch it is actively corrupting:
+    // compress decides purely by Content-Type, so for a compressible mime
+    // (application/octet-stream fallback, image/svg+xml, text/*) it would gzip
+    // the byte slice and drop Content-Length while Content-Range still
+    // describes the RAW offsets and the status stays 206. A resuming client
+    // (`curl -C -`, download managers) then appends the encoded bytes as if
+    // raw and ends up with a broken file. @fastify/compress skips whenever the
+    // request carries `x-no-compression` (see its onSend hook), so setting it
+    // here covers both the 200 (full file) and 206 (range) responses.
+    req.headers['x-no-compression'] = 'true';
+
     res.header('Accept-Ranges', 'bytes');
     res.header(
       'Content-Security-Policy',

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

@@ -52,7 +52,9 @@ import {
   INTERNAL_LINK_REGEX,
   extractPageSlugId,
 } from '../../../integrations/export/utils';
-import { markdownToHtml, canonicalizeFootnotes } from '@docmost/editor-ext';
+import { canonicalizeFootnotes } from '@docmost/editor-ext';
+import { markdownToProseMirror } from '@docmost/prosemirror-markdown';
+import { normalizeForeignMarkdown } from '../../../integrations/import/utils/foreign-markdown';
 import { WatcherService } from '../../watcher/watcher.service';
 import { sql } from 'kysely';
 import { TransclusionService } from '../transclusion/transclusion.service';
@@ -1301,8 +1303,14 @@ export class PageService {
 
     switch (format) {
       case 'markdown': {
-        const html = await markdownToHtml(content as string);
+        // Canonical markdown -> ProseMirror JSON directly via
-        prosemirrorJson = htmlToJson(html as string);
+        // `@docmost/prosemirror-markdown` (issue #345) — no HTML intermediate,
+        // no editor-ext markdown layer. Foreign markdown surfaces the strict
+        // parser rejects (GFM `[^id]` reference footnotes) are normalized to the
+        // canonical inline form first.
+        prosemirrorJson = await markdownToProseMirror(
+          normalizeForeignMarkdown(content as string),
+        );
         break;
       }
       case 'html': {

apps/server/src/database/listeners/page.listener.ts

@@ -4,7 +4,6 @@ import { EventName } from '../../common/events/event.contants';
 import { InjectQueue } from '@nestjs/bullmq';
 import { QueueJob, QueueName } from '../../integrations/queue/constants';
 import { Queue } from 'bullmq';
-import { EnvironmentService } from '../../integrations/environment/environment.service';
 
 /**
  * Thin snapshot of a page node carried inside domain events so the WebSocket
@@ -112,48 +111,24 @@ export class PageListener {
   private readonly logger = new Logger(PageListener.name);
 
   constructor(
-    private readonly environmentService: EnvironmentService,
-    @InjectQueue(QueueName.SEARCH_QUEUE) private searchQueue: Queue,
     @InjectQueue(QueueName.AI_QUEUE) private aiQueue: Queue,
   ) {}
 
   @OnEvent(EventName.PAGE_CREATED)
   async handlePageCreated(event: PageEvent) {
     const { pageIds, workspaceId } = event;
-    if (this.isTypesense()) {
-      await this.searchQueue.add(QueueJob.PAGE_CREATED, {
-        pageIds,
-      });
-    }
-
     await this.aiQueue.add(QueueJob.PAGE_CREATED, { pageIds, workspaceId });
   }
 
-  @OnEvent(EventName.PAGE_UPDATED)
-  async handlePageUpdated(event: PageEvent) {
-    const { pageIds } = event;
-
-    await this.searchQueue.add(QueueJob.PAGE_UPDATED, { pageIds });
-  }
-
   @OnEvent(EventName.PAGE_DELETED)
   async handlePageDeleted(event: PageEvent) {
     const { pageIds, workspaceId } = event;
-    if (this.isTypesense()) {
-      await this.searchQueue.add(QueueJob.PAGE_DELETED, { pageIds });
-    }
-
     await this.aiQueue.add(QueueJob.PAGE_DELETED, { pageIds, workspaceId });
   }
 
   @OnEvent(EventName.PAGE_SOFT_DELETED)
   async handlePageSoftDeleted(event: PageEvent) {
     const { pageIds, workspaceId } = event;
-
-    if (this.isTypesense()) {
-      await this.searchQueue.add(QueueJob.PAGE_SOFT_DELETED, { pageIds });
-    }
-
     await this.aiQueue.add(QueueJob.PAGE_SOFT_DELETED, {
       pageIds,
       workspaceId,
@@ -163,14 +138,6 @@ export class PageListener {
   @OnEvent(EventName.PAGE_RESTORED)
   async handlePageRestored(event: PageEvent) {
     const { pageIds, workspaceId } = event;
-    if (this.isTypesense()) {
-      await this.searchQueue.add(QueueJob.PAGE_RESTORED, { pageIds });
-    }
-
     await this.aiQueue.add(QueueJob.PAGE_RESTORED, { pageIds, workspaceId });
   }
-
-  isTypesense(): boolean {
-    return this.environmentService.getSearchDriver() === 'typesense';
-  }
 }

apps/server/src/database/listeners/space.listener.ts

@@ -4,7 +4,6 @@ import { EventName } from '../../common/events/event.contants';
 import { InjectQueue } from '@nestjs/bullmq';
 import { QueueJob, QueueName } from '../../integrations/queue/constants';
 import { Queue } from 'bullmq';
-import { EnvironmentService } from '../../integrations/environment/environment.service';
 
 export class SpaceEvent {
   spaceId: string;
@@ -15,22 +14,12 @@ export class SpaceListener {
   private readonly logger = new Logger(SpaceListener.name);
 
   constructor(
-    private readonly environmentService: EnvironmentService,
-    @InjectQueue(QueueName.SEARCH_QUEUE) private searchQueue: Queue,
     @InjectQueue(QueueName.AI_QUEUE) private aiQueue: Queue,
   ) {}
 
   @OnEvent(EventName.SPACE_DELETED)
   async handleSpaceDeleted(event: SpaceEvent) {
     const { spaceId } = event;
-    if (this.isTypesense()) {
-      await this.searchQueue.add(QueueJob.SPACE_DELETED, { spaceId });
-    }
-
     await this.aiQueue.add(QueueJob.SPACE_DELETED, { spaceId });
   }
-
-  isTypesense(): boolean {
-    return this.environmentService.getSearchDriver() === 'typesense';
-  }
 }

apps/server/src/database/listeners/workspace.listener.ts

@@ -4,7 +4,6 @@ import { EventName } from '../../common/events/event.contants';
 import { InjectQueue } from '@nestjs/bullmq';
 import { QueueJob, QueueName } from '../../integrations/queue/constants';
 import { Queue } from 'bullmq';
-import { EnvironmentService } from '../../integrations/environment/environment.service';
 
 export class WorkspaceEvent {
   workspaceId: string;
@@ -15,22 +14,12 @@ export class WorkspaceListener {
   private readonly logger = new Logger(WorkspaceListener.name);
 
   constructor(
-    private readonly environmentService: EnvironmentService,
-    @InjectQueue(QueueName.SEARCH_QUEUE) private searchQueue: Queue,
     @InjectQueue(QueueName.AI_QUEUE) private aiQueue: Queue,
   ) {}
 
   @OnEvent(EventName.WORKSPACE_DELETED)
   async handlePageDeleted(event: WorkspaceEvent) {
     const { workspaceId } = event;
-    if (this.isTypesense()) {
-      await this.searchQueue.add(QueueJob.WORKSPACE_DELETED, { workspaceId });
-    }
-
     await this.aiQueue.add(QueueJob.WORKSPACE_DELETED, { workspaceId });
   }
-
-  isTypesense(): boolean {
-    return this.environmentService.getSearchDriver() === 'typesense';
-  }
 }

apps/server/src/database/migrate.ts

@@ -24,6 +24,10 @@ const migrator = new Migrator({
     path,
     migrationFolder,
   }),
+  // Match the startup auto-migrator (migration.service.ts): a back-dated
+  // migration from a long-lived branch must be applied, not rejected as
+  // "corrupted migrations" (incident #361). See that file for the full rationale.
+  allowUnorderedMigrations: true,
 });
 
 run(db, migrator, migrationFolder);

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

@@ -19,6 +19,16 @@ export class MigrationService {
         path,
         migrationFolder: path.join(__dirname, '..', 'migrations'),
       }),
+      // A long-lived branch can add a migration whose timestamped filename sorts
+      // BEFORE migrations already applied in prod (e.g. #234's 20260627 landing
+      // after 20260704 was live). With the default (ordered) setting the startup
+      // migrator then sees "corrupted migrations" — the applied set is no longer a
+      // prefix of the sorted list — throws, and the app crash-loops on boot
+      // (incident #361: 502s for ~11 min). allowUnorderedMigrations runs any
+      // not-yet-applied migration regardless of filename order, so a back-dated
+      // migration is applied instead of bricking startup. A CI order-gate still
+      // discourages back-dating; this is the runtime safety net.
+      allowUnorderedMigrations: true,
     });
 
     const { error, results } = await migrator.migrateToLatest();

apps/server/src/integrations/ai/ai-sdk-partial-output.patch.spec.ts

@@ -0,0 +1,124 @@
+import { readFileSync } from 'fs';
+import { streamText, Output } from 'ai';
+import { MockLanguageModelV3, simulateReadableStream } from 'ai/test';
+
+/**
+ * Regression tests for patches/ai@6.0.134.patch (server heap OOM on long
+ * autonomous agent runs, #184).
+ *
+ * Unpatched ai@6.0.134 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 a
+ * flat snapshot of all text so far as `partialOutput` (O(n^2) memory). Those
+ * snapshots pile up in the never-consumed leftover tee() branch of
+ * DefaultStreamTextResult.baseStream, which is what OOM'd production during a
+ * ~28k-chunk agent turn. The pnpm patch skips partialOutput production
+ * entirely when no output strategy was requested, while keeping per-delta
+ * streaming granularity.
+ */
+describe('ai@6.0.134 pnpm patch: no partialOutput accumulation without an output strategy', () => {
+  const makeModel = () =>
+    new MockLanguageModelV3({
+      doStream: async () => ({
+        stream: simulateReadableStream({
+          chunks: [
+            { type: 'stream-start' as const, warnings: [] },
+            { type: 'text-start' as const, id: '1' },
+            { type: 'text-delta' as const, id: '1', delta: 'Hello' },
+            { type: 'text-delta' as const, id: '1', delta: ', ' },
+            { type: 'text-delta' as const, id: '1', delta: 'world!' },
+            { type: 'text-end' as const, id: '1' },
+            {
+              type: 'finish' as const,
+              finishReason: { unified: 'stop' as const, raw: 'stop' },
+              usage: {
+                inputTokens: {
+                  total: 1,
+                  noCache: undefined,
+                  cacheRead: undefined,
+                  cacheWrite: undefined,
+                },
+                outputTokens: { total: 1, text: 1, reasoning: undefined },
+              },
+            },
+          ],
+        }),
+      }),
+    });
+
+  it('preserves per-delta streaming granularity in textStream', async () => {
+    const result = streamText({ model: makeModel(), prompt: 'hi' });
+
+    const deltas: string[] = [];
+    for await (const delta of result.textStream) {
+      deltas.push(delta);
+    }
+
+    // The patch must NOT coalesce or drop deltas: three model deltas arrive
+    // as three separate textStream chunks.
+    expect(deltas).toEqual(['Hello', ', ', 'world!']);
+  });
+
+  it('emits NO partialOutput values when the caller did not request an output strategy', async () => {
+    const result = streamText({ model: makeModel(), prompt: 'hi' });
+
+    // Fully consume the primary stream first (mirrors production usage).
+    for await (const _ of result.textStream) {
+      // drain
+    }
+
+    const partials: unknown[] = [];
+    for await (const partial of result.experimental_partialOutputStream) {
+      partials.push(partial);
+    }
+
+    // TRIPWIRE: on unpatched ai@6.0.134 the default text() output strategy
+    // yields one cumulative partial per text-delta here (['Hello', 'Hello, ',
+    // 'Hello, world!']). An empty stream proves the patch is applied and no
+    // cumulative snapshots are being produced (and thus none can pile up in
+    // the leftover internal tee branch).
+    expect(partials).toEqual([]);
+  });
+
+  it('preserves cumulative partialOutput when the caller DOES request an output strategy', async () => {
+    // PRESERVE-BRANCH GUARD: the patch only short-circuits partialOutput when
+    // `output == null`. When an output strategy IS set (here Output.text()),
+    // createOutputTransformStream must fall through to the ORIGINAL code path
+    // and keep publishing cumulative snapshots, so object/text-output consumers
+    // behave byte-identically to unpatched ai. A careless re-port that routed
+    // output-set calls into the skip branch would leave partialOutput empty and
+    // silently break those consumers — this test is the tripwire for that.
+    const result = streamText({
+      model: makeModel(),
+      prompt: 'hi',
+      experimental_output: Output.text(),
+    });
+
+    // Drain the primary stream fully and accumulate the complete output text.
+    let fullText = '';
+    for await (const delta of result.textStream) {
+      fullText += delta;
+    }
+
+    const partials: string[] = [];
+    for await (const partial of result.experimental_partialOutputStream) {
+      partials.push(partial);
+    }
+
+    // With a strategy set, partialOutput must be PRESERVED (non-empty) and
+    // cumulative: the last emitted partial equals the full accumulated text.
+    expect(partials.length).toBeGreaterThan(0);
+    expect(partials[partials.length - 1]).toBe(fullText);
+    expect(fullText).toBe('Hello, world!');
+  });
+
+  it('both installed dist builds (CJS and ESM) carry the patch marker', () => {
+    // Secondary guard: pins the patch to BOTH bundles the SDK ships, since
+    // the NestJS server consumes CJS while other tooling may load ESM.
+    const cjsPath = require.resolve('ai');
+    const mjsPath = cjsPath.replace(/index\.js$/, 'index.mjs');
+    expect(cjsPath).toMatch(/index\.js$/);
+    expect(readFileSync(cjsPath, 'utf8')).toContain('PATCH(docmost');
+    expect(readFileSync(mjsPath, 'utf8')).toContain('PATCH(docmost');
+  });
+});

apps/server/src/integrations/export/export-markdown.spec.ts

@@ -0,0 +1,145 @@
+// export.service.ts imports the ESM-only @sindresorhus/slugify (not in jest's
+// transform allowlist). It is irrelevant to the markdown-serialization path under
+// test (only used for page-mention link slugs on the DB path), so it is mocked
+// out to keep the module graph loadable under ts-jest (mirrors the import specs).
+jest.mock('@sindresorhus/slugify', () => ({
+  __esModule: true,
+  default: (input: string) => String(input),
+}));
+
+import { convertProseMirrorToMarkdown } from '@docmost/prosemirror-markdown';
+import { ExportService } from './export.service';
+import { ExportFormat } from './dto/export-dto';
+
+/**
+ * STEP 1 golden test for issue #345: server MARKDOWN export runs DIRECTLY through
+ * the canonical converter (`convertProseMirrorToMarkdown`) — no HTML intermediate
+ * and no `@docmost/editor-ext` markdown layer — so the emitted markdown is in the
+ * canonical package forms and is byte-identical to the git-sync vault body.
+ *
+ * These are the goldens the swap has to satisfy: they assert the CANONICAL
+ * surface (callout `> [!type]`, inline footnote `^[…]`, lossless image
+ * `<!--img …-->`) rather than the old editor-ext forms (`:::type`, `[^id]`,
+ * lossy `![alt](src)`).
+ *
+ * `exportPage(..., singlePage=false)` takes no DB path (no mention rewriting), so
+ * the service is constructed with null collaborators and only the pure
+ * PM -> Markdown path is exercised.
+ */
+
+function makeService(): ExportService {
+  return new ExportService(
+    null as any, // pageRepo
+    null as any, // pagePermissionRepo
+    null as any, // db
+    null as any, // storageService
+    null as any, // environmentService
+    null as any, // domainService
+  );
+}
+
+// A representative page exercising the node types whose canonical markdown form
+// changed with the move off the editor-ext layer: callout, inline footnote, and a
+// lossless image carrying width/align attrs that the old layer dropped.
+const REPRESENTATIVE_DOC = {
+  type: 'doc',
+  content: [
+    {
+      type: 'paragraph',
+      content: [
+        { type: 'text', text: 'Body ' },
+        { type: 'footnoteReference', attrs: { id: 'fn-1' } },
+        { type: 'text', text: ' end.' },
+      ],
+    },
+    {
+      type: 'callout',
+      attrs: { type: 'info', icon: null },
+      content: [
+        {
+          type: 'paragraph',
+          content: [{ type: 'text', text: 'Heads up' }],
+        },
+      ],
+    },
+    {
+      type: 'image',
+      attrs: {
+        src: '/files/pic.png',
+        alt: 'Pic',
+        width: 320,
+        align: 'left',
+      },
+    },
+    {
+      type: 'footnotesList',
+      content: [
+        {
+          type: 'footnoteDefinition',
+          attrs: { id: 'fn-1' },
+          content: [
+            {
+              type: 'paragraph',
+              content: [{ type: 'text', text: 'the note' }],
+            },
+          ],
+        },
+      ],
+    },
+  ],
+};
+
+describe('ExportService — markdown export via the canonical converter (#345)', () => {
+  it('emits canonical callout, inline footnote and lossless image forms', async () => {
+    const service = makeService();
+    const md = (await service.exportPage(ExportFormat.Markdown, {
+      title: '',
+      content: REPRESENTATIVE_DOC,
+    } as any)) as string;
+
+    // Callout: Obsidian `> [!type]`, NOT the legacy `:::type`.
+    expect(md).toContain('> [!info]');
+    expect(md).not.toContain(':::');
+
+    // Inline footnote: `^[…]`, NOT the reference `[^id]` form.
+    expect(md).toContain('^[the note]');
+    expect(md).not.toMatch(/\[\^/);
+
+    // Lossless image: trailing `<!--img …-->` carrying the dropped attrs.
+    expect(md).toContain('![Pic](/files/pic.png)');
+    expect(md).toContain('<!--img');
+    expect(md).toContain('"width":"320"');
+    expect(md).toContain('"align":"left"');
+  });
+
+  it('export body is byte-identical to the git-sync vault serializer (export == vault)', async () => {
+    const service = makeService();
+    // A title-less page: exportPage prepends NO heading, so the whole output is
+    // the page BODY — exactly what git-sync serializes (git-sync stores the title
+    // in frontmatter / the filename, never as an in-body H1).
+    const exported = (await service.exportPage(ExportFormat.Markdown, {
+      title: '',
+      content: REPRESENTATIVE_DOC,
+    } as any)) as string;
+
+    // The git-sync vault writer feeds this SAME converter (git-sync
+    // `stabilizePageBody` = convertProseMirrorToMarkdown(content) at the
+    // fixpoint). For an already-stable doc the single pass IS the fixpoint, so
+    // the two are byte-identical by construction — assert it.
+    const vaultBody = convertProseMirrorToMarkdown(REPRESENTATIVE_DOC);
+    expect(exported).toBe(vaultBody);
+  });
+
+  it('prepends the page title as an H1 heading (the one documented export/vault delta)', async () => {
+    const service = makeService();
+    const md = (await service.exportPage(ExportFormat.Markdown, {
+      title: 'My Page',
+      content: { type: 'doc', content: [] },
+    } as any)) as string;
+
+    // Export makes standalone files, so it prepends the title as an H1. This is
+    // the ONE deliberate difference from the vault body (which carries the title
+    // in frontmatter). The body below the heading still serializes canonically.
+    expect(md.startsWith('# My Page')).toBe(true);
+  });
+});

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

@@ -37,7 +37,7 @@ import {
   getAttachmentIds,
   getProsemirrorContent,
 } from '../../common/helpers/prosemirror/utils';
-import { htmlToMarkdown } from '@docmost/editor-ext';
+import { convertProseMirrorToMarkdown } from '@docmost/prosemirror-markdown';
 
 type AllowedAttachment = { id: string; fileName: string; filePath: string };
 
@@ -79,9 +79,8 @@ export class ExportService {
       prosemirrorJson.content.unshift(titleNode);
     }
 
-    const pageHtml = jsonToHtml(prosemirrorJson);
-
     if (format === ExportFormat.HTML) {
+      const pageHtml = jsonToHtml(prosemirrorJson);
       return `<!DOCTYPE html>
       <html>
         <head>
@@ -92,11 +91,14 @@ export class ExportService {
     }
 
     if (format === ExportFormat.Markdown) {
-      const newPageHtml = pageHtml.replace(
+      // Direct ProseMirror JSON -> Markdown via the canonical converter
-        /<colgroup[^>]*>[\s\S]*?<\/colgroup>/gim,
+      // (`@docmost/prosemirror-markdown`). This is the SAME serializer the
-        '',
+      // git-sync vault writer feeds (see git-sync `stabilizePageBody`), so an
-      );
+      // exported page body is byte-identical to its vault representation — no
-      return htmlToMarkdown(newPageHtml);
+      // HTML intermediate, no second markdown layer, no format drift (issue
+      // #345). The old `<colgroup>` scrub is gone with the HTML step: the
+      // converter emits GFM tables directly and never produces `<colgroup>`.
+      return convertProseMirrorToMarkdown(prosemirrorJson);
     }
 
     return;

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

@@ -17,6 +17,22 @@ jest.mock('image-dimensions', () => ({
   __esModule: true,
   imageDimensionsFromData: () => undefined,
 }));
+// FileImportTaskService -> PageService -> collaboration.gateway ->
+// metrics.registry imports `prom-client`, which is not resolvable in this
+// workspace's node_modules (types-only stub, no runtime entry). Metrics are
+// disabled on this path, so a virtual no-op mock keeps the module graph loadable.
+jest.mock(
+  'prom-client',
+  () => ({
+    collectDefaultMetrics: () => undefined,
+    Registry: class {},
+    Histogram: class {},
+    Gauge: class {},
+    Counter: class {},
+    Summary: class {},
+  }),
+  { virtual: true },
+);
 
 import { promises as fs } from 'fs';
 import * as os from 'os';
@@ -26,14 +42,17 @@ import { ImportService } from './import.service';
 
 /**
  * Binding test for issue #228 / review #5: FileImportTaskService.processGenericImport
- * is a NON-editor write path (markdownToHtml -> processHTML -> JSON, never runs
+ * is a NON-editor write path, so a zip-imported `.md` page ends up with canonical
- * footnoteSyncPlugin), so it canonicalizes footnotes before persisting. This pins
+ * footnotes before persisting: ordered by first reference, reused refs deduped,
- * that binding — the same one import.service has a spec for — which previously had
+ * orphan definitions dropped.
- * NO spec at all.
  *
- * The markdown -> HTML -> ProseMirror conversion is REAL (a real ImportService,
+ * Since #345 the `.md` parse runs `normalizeForeignMarkdown` ->
- * its createYdoc stubbed); the filesystem is a real temp dir with one .md file;
+ * `markdownToProseMirror` -> `jsonToHtml` (feeding the shared HTML attachment /
- * the DB transaction is stubbed to capture the persisted page content.
+ * link pipeline) -> `processHTML` -> `canonicalizeFootnotes`. The parser assigns
+ * fresh `fn-*` ids, so we assert by definition BODY order rather than the source
+ * labels. The conversion is REAL (a real ImportService, its createYdoc stubbed);
+ * the filesystem is a real temp dir with one .md file; the DB transaction is
+ * stubbed to capture the persisted page content.
  */
 
 // Out-of-order references (c, a, b), a REUSED reference ([^a] twice), and an
@@ -49,13 +68,14 @@ const MARKDOWN = [
   '[^z]: orphan note',
 ].join('\n');
 
-function footnoteListIds(content: any): string[] {
+/** Definition body texts of the (single) footnotesList, in list order. */
+function footnoteListBodies(content: any): string[] {
   const list = (content?.content ?? []).find(
     (n: any) => n.type === 'footnotesList',
   );
   return (list?.content ?? [])
     .filter((n: any) => n.type === 'footnoteDefinition')
-    .map((n: any) => n.attrs?.id);
+    .map((n: any) => n.content?.[0]?.content?.[0]?.text);
 }
 
 // A permissive chainable stub for the spaces lookup (selectFrom(...).select(...)
@@ -71,80 +91,127 @@ function chainable(result: any): any {
   return proxy;
 }
 
+/**
+ * Run one markdown file through the REAL zip-import pipeline
+ * (`processGenericImport` -> `markdownToProseMirror` -> `jsonToHtml` ->
+ * `processHTML`/`htmlToJson`) and return the persisted page `content`. This is
+ * the server-specific PM->HTML->PM hop that the package's own PM<->MD tests do
+ * NOT cover.
+ */
+async function runZipImport(markdown: string): Promise<any> {
+  const extractDir = await fs.mkdtemp(path.join(os.tmpdir(), 'fit-canon-'));
+  await fs.writeFile(path.join(extractDir, 'note.md'), markdown, 'utf-8');
+
+  const importService = new ImportService(
+    {} as any,
+    {} as any,
+    {} as any,
+    {} as any,
+  );
+  jest
+    .spyOn(importService as any, 'createYdoc')
+    .mockResolvedValue(Buffer.from([]) as any);
+
+  let captured: any = null;
+  const trx = {
+    insertInto: (table: string) => ({
+      values: (v: any) => {
+        if (table === 'pages') captured = v;
+        return { execute: async () => {} };
+      },
+    }),
+  };
+  const db: any = {
+    selectFrom: () => chainable({ slug: 'space-slug' }),
+    transaction: () => ({ execute: (fn: any) => fn(trx) }),
+  };
+
+  const importAttachmentService = {
+    processAttachments: async ({ html }: any) => html,
+  };
+  const service = new FileImportTaskService(
+    {} as any, // storageService
+    importService as any,
+    { nextPagePosition: async () => 'a0' } as any,
+    { insertBacklink: jest.fn() } as any,
+    db,
+    importAttachmentService as any,
+    { emit: jest.fn() } as any,
+    { logBatchWithContext: jest.fn() } as any,
+  );
+
+  const fileTask: any = {
+    id: 'task-1',
+    source: 'generic',
+    spaceId: 'space-1',
+    workspaceId: 'ws-1',
+    creatorId: 'user-1',
+  };
+
+  try {
+    await service.processGenericImport({ extractDir, fileTask });
+    expect(captured).toBeTruthy();
+    return captured.content;
+  } finally {
+    await fs.rm(extractDir, { recursive: true, force: true });
+  }
+}
+
+/** Find the first node of a given type anywhere in a PM content tree. */
+function findFirst(node: any, type: string): any {
+  if (!node || typeof node !== 'object') return null;
+  if (node.type === type) return node;
+  for (const child of node.content ?? []) {
+    const hit = findFirst(child, type);
+    if (hit) return hit;
+  }
+  return null;
+}
+
 describe('FileImportTaskService.processGenericImport — footnote canonicalization (#228)', () => {
   it('orders footnotes by first reference, dedupes reuse, and drops orphans on zip import', async () => {
-    const extractDir = await fs.mkdtemp(path.join(os.tmpdir(), 'fit-canon-'));
+    const content = await runZipImport(MARKDOWN);
-    await fs.writeFile(path.join(extractDir, 'note.md'), MARKDOWN, 'utf-8');
+    // Definitions ordered by FIRST REFERENCE (C, A, B), NOT the markdown
-
+    // definition order (A, B, C). Ids are the parser's fresh `fn-*`, so pin
-    // Real ImportService for the html -> JSON conversion; stub the yjs encode.
+    // the BODIES.
-    const importService = new ImportService(
+    expect(footnoteListBodies(content)).toEqual(['note C', 'note A', 'note B']);
-      {} as any,
+    // Orphan [^z] dropped; reused [^a] collapses to one definition; one list.
-      {} as any,
+    expect(footnoteListBodies(content)).not.toContain('orphan note');
-      {} as any,
+    const lists = (content.content ?? []).filter(
-      {} as any,
+      (n: any) => n.type === 'footnotesList',
     );
-    jest
+    expect(lists).toHaveLength(1);
-      .spyOn(importService as any, 'createYdoc')
+    expect(
-      .mockResolvedValue(Buffer.from([]) as any);
+      footnoteListBodies(content).filter((b) => b === 'note A'),
+    ).toHaveLength(1);
+  });
 
-    let captured: any = null;
+  // #345 F4: the zip path routes markdown through jsonToHtml -> processHTML ->
-    const trx = {
+  // htmlToJson (the shared HTML attachment pipeline). #345's headline is LOSSLESS
-      insertInto: (table: string) => ({
+  // image width/align via the `<!--img {...}-->` comment; a callout carries its
-        values: (v: any) => {
+  // `type`. This asserts those survive the PM->HTML->PM hop — the one hop the
-          if (table === 'pages') captured = v;
+  // package's PM<->MD suite does not exercise.
-          return { execute: async () => {} };
+  it('preserves image width/align and callout type through the PM->HTML->PM hop', async () => {
-        },
+    const md = [
-      }),
+      '# Doc',
-    };
+      '',
-    const db: any = {
+      '![a picture](https://example.com/i.png) <!--img {"width":"320","align":"left"}-->',
-      selectFrom: () => chainable({ slug: 'space-slug' }),
+      '',
-      transaction: () => ({ execute: (fn: any) => fn(trx) }),
+      ':::warning',
-    };
+      'Careful now.',
+      ':::',
+    ].join('\n');
 
-    const importAttachmentService = {
+    const content = await runZipImport(md);
-      processAttachments: async ({ html }: any) => html,
-    };
-    const backlinkRepo = { insertBacklink: jest.fn() };
-    const eventEmitter = { emit: jest.fn() };
-    const auditService = { logBatchWithContext: jest.fn() };
 
-    const pageService = { nextPagePosition: async () => 'a0' };
+    const image = findFirst(content, 'image');
+    expect(image).toBeTruthy();
+    // The lossless sizing/alignment must survive the HTML hop.
+    expect(String(image.attrs?.width)).toBe('320');
+    expect(image.attrs?.align).toBe('left');
 
-    const service = new FileImportTaskService(
+    const callout = findFirst(content, 'callout');
-      {} as any, // storageService
+    expect(callout).toBeTruthy();
-      importService as any,
+    expect(callout.attrs?.type).toBe('warning');
-      pageService as any,
-      backlinkRepo as any,
-      db,
-      importAttachmentService as any,
-      eventEmitter as any,
-      auditService as any,
-    );
-
-    const fileTask: any = {
-      id: 'task-1',
-      source: 'generic',
-      spaceId: 'space-1',
-      workspaceId: 'ws-1',
-      creatorId: 'user-1',
-    };
-
-    try {
-      await service.processGenericImport({ extractDir, fileTask });
-
-      expect(captured).toBeTruthy();
-      const content = captured.content;
-      // Reference order is c, a, b (NOT the markdown definition order a, b, c).
-      expect(footnoteListIds(content)).toEqual(['c', 'a', 'b']);
-      // Orphan [^z] dropped; reused [^a] collapses to one definition; one list.
-      expect(footnoteListIds(content)).not.toContain('z');
-      const lists = (content.content ?? []).filter(
-        (n: any) => n.type === 'footnotesList',
-      );
-      expect(lists).toHaveLength(1);
-      expect(footnoteListIds(content).filter((id) => id === 'a')).toHaveLength(1);
-    } finally {
-      await fs.rm(extractDir, { recursive: true, force: true });
-    }
   });
 });

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

@@ -1,6 +1,9 @@
 import { Inject, Injectable, Logger } from '@nestjs/common';
 import * as path from 'path';
-import { jsonToText } from '../../../collaboration/collaboration.util';
+import {
+  jsonToHtml,
+  jsonToText,
+} from '../../../collaboration/collaboration.util';
 import { InjectKysely } from 'nestjs-kysely';
 import { KyselyDB } from '@docmost/db/types/kysely.types';
 import {
@@ -18,9 +21,11 @@ import { generateSlugId } from '../../../common/helpers';
 import { v7 } from 'uuid';
 import { generateJitteredKeyBetween } from 'fractional-indexing-jittered';
 import { FileTask, InsertablePage } from '@docmost/db/types/entity.types';
-import { markdownToHtml, canonicalizeFootnotes } from '@docmost/editor-ext';
+import { canonicalizeFootnotes } from '@docmost/editor-ext';
+import { markdownToProseMirror } 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,
@@ -461,7 +466,18 @@ export class FileImportTaskService {
               content = await fs.readFile(absPath, 'utf-8');
 
               if (page.fileExtension.toLowerCase() === '.md') {
-                content = await markdownToHtml(content);
+                // Parse markdown with the single canonical converter
+                // (`@docmost/prosemirror-markdown`), after normalizing foreign
+                // reference footnotes, then serialize to HTML so the shared HTML
+                // pipeline below (processAttachments + formatImportHtml +
+                // processHTML) keeps handling `.md` and `.html` imports
+                // uniformly. The markdown PARSE no longer goes through the
+                // editor-ext markdown layer (issue #345) — the drift source is
+                // gone. The PM -> HTML -> PM hop that follows is lossless
+                // plumbing for attachment/link resolution, NOT a second parse.
+                content = jsonToHtml(
+                  await markdownToProseMirror(normalizeForeignMarkdown(content)),
+                );
               }
             } catch (err: any) {
               if (err?.code === 'ENOENT') {
@@ -500,10 +516,12 @@ export class FileImportTaskService {
               this.importService.extractTitleAndRemoveHeading(pmState);
 
             // Canonicalize footnote topology on this non-editor write path
-            // (markdownToHtml/processHTML never runs footnoteSyncPlugin), so a
+            // (the HTML pipeline's processHTML never runs footnoteSyncPlugin), so
-            // zip-imported page's footnotes are reference-ordered, deduped, and
+            // a zip-imported page's footnotes are reference-ordered, deduped, and
             // orphan-free like the editor's invariant (issue #228). Pure +
-            // idempotent + shape-safe; a footnote-free doc is unchanged.
+            // idempotent + shape-safe; a footnote-free doc is unchanged. (For a
+            // `.md` file the package parser already yields canonical footnotes,
+            // so this is a no-op there.)
             // (Future consolidation, architecture B: like import.service, this
             // path persists directly rather than via PageService — a shared
             // "prepare JSON for persist" helper would centralize this call.)

apps/server/src/integrations/import/services/import.service.footnote-canonicalize.spec.ts

@@ -12,13 +12,19 @@ import { canonicalizeFootnotes } from '@docmost/editor-ext';
 
 /**
  * Integration-ish test for the USER-FACING markdown import path
- * (`ImportService.importPage`). It exercises the REAL markdown -> HTML -> JSON
+ * (`ImportService.importPage`). It exercises the REAL markdown -> ProseMirror
- * conversion and asserts that the stored page content has its footnotes
+ * conversion and asserts the stored page's footnotes are canonical: ordered by
- * canonicalized — the gap that issue #228 fixes: the import path builds
+ * FIRST REFERENCE (not markdown definition order), reused references deduped to a
- * ProseMirror JSON directly (never running the editor's footnoteSyncPlugin), so
+ * single definition, and orphan definitions dropped.
- * before this wiring the stored footnotes kept the markdown's physical
+ *
- * definition order (out of order vs. references), retained orphan definitions,
+ * Since #345 the markdown parse runs through the canonical package
- * and did not collapse reused references.
+ * (`normalizeForeignMarkdown` -> `markdownToProseMirror`), which owns this
+ * canonicalization: the input's GFM `[^id]` reference footnotes are normalized to
+ * inline `^[…]`, and the parser assigns fresh sequential ids (`fn-*`) in
+ * reference order while merging identical bodies — so we assert by definition
+ * BODY order, not by the source labels. `canonicalizeFootnotes` remains wired as
+ * an idempotent safety net (issue #228) and is a no-op on this already-canonical
+ * output.
  *
  * The DB/ydoc side-effects are stubbed: `getNewPagePosition` (DB query) and
  * `createYdoc` (Yjs encode) are spied, and `pageRepo.insertPage` captures the
@@ -67,24 +73,14 @@ function makeService() {
 }
 
 /** List the footnote-definition ids of the (single) footnotesList, in order. */
-function footnoteListIds(content: any): string[] {
+/** Definition body texts of the (single) footnotesList, in list order. */
+function footnoteListBodies(content: any): string[] {
   const list = (content.content ?? []).find(
     (n: any) => n.type === 'footnotesList',
   );
-  if (!list) return [];
+  return (list?.content ?? [])
-  return (list.content ?? [])
     .filter((n: any) => n.type === 'footnoteDefinition')
-    .map((n: any) => n.attrs?.id);
+    .map((n: any) => n.content?.[0]?.content?.[0]?.text);
-}
-
-function definitionText(content: any, id: string): string | undefined {
-  const list = (content.content ?? []).find(
-    (n: any) => n.type === 'footnotesList',
-  );
-  const def = (list?.content ?? []).find(
-    (n: any) => n.type === 'footnoteDefinition' && n.attrs?.id === id,
-  );
-  return def?.content?.[0]?.content?.[0]?.text;
 }
 
 describe('ImportService.importPage — footnote canonicalization (#228)', () => {
@@ -101,23 +97,23 @@ describe('ImportService.importPage — footnote canonicalization (#228)', () =>
     const content = getCaptured().content;
     expect(content).toBeTruthy();
 
-    // Reference order is c, a, b (NOT the markdown definition order a, b, c).
+    // Definitions ordered by FIRST REFERENCE (C, A, B) — NOT the markdown
-    expect(footnoteListIds(content)).toEqual(['c', 'a', 'b']);
+    // definition order (A, B, C) — with the orphan [^z] dropped and the reused
-
+    // [^a] collapsed to a single definition. (Ids are the parser's fresh `fn-*`,
-    // Definitions preserved and attached to the right ids.
+    // so we pin the BODIES.)
-    expect(definitionText(content, 'c')).toBe('note C');
+    expect(footnoteListBodies(content)).toEqual(['note C', 'note A', 'note B']);
-    expect(definitionText(content, 'a')).toBe('note A');
-    expect(definitionText(content, 'b')).toBe('note B');
 
     // Orphan definition [^z] is dropped.
-    expect(footnoteListIds(content)).not.toContain('z');
+    expect(footnoteListBodies(content)).not.toContain('orphan note');
 
     // Reused [^a] yields exactly ONE definition, and exactly one list.
     const lists = (content.content ?? []).filter(
       (n: any) => n.type === 'footnotesList',
     );
     expect(lists).toHaveLength(1);
-    expect(footnoteListIds(content).filter((id) => id === 'a')).toHaveLength(1);
+    expect(
+      footnoteListBodies(content).filter((b) => b === 'note A'),
+    ).toHaveLength(1);
   });
 
   it('is idempotent: canonicalizing the stored output again is a no-op', async () => {
@@ -134,6 +130,6 @@ describe('ImportService.importPage — footnote canonicalization (#228)', () =>
     // time must not change it (safe to wire into every write path).
     const second = canonicalizeFootnotes(stored);
     expect(second).toEqual(stored);
-    expect(footnoteListIds(second)).toEqual(['c', 'a', 'b']);
+    expect(footnoteListBodies(second)).toEqual(['note C', 'note A', 'note B']);
   });
 });

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

@@ -17,7 +17,9 @@ import {
 import { generateJitteredKeyBetween } from 'fractional-indexing-jittered';
 import { TiptapTransformer } from '@hocuspocus/transformer';
 import * as Y from 'yjs';
-import { markdownToHtml, canonicalizeFootnotes } from '@docmost/editor-ext';
+import { canonicalizeFootnotes } from '@docmost/editor-ext';
+import { markdownToProseMirror } from '@docmost/prosemirror-markdown';
+import { normalizeForeignMarkdown } from '../utils/foreign-markdown';
 import {
   FileTaskStatus,
   FileTaskType,
@@ -85,11 +87,13 @@ export class ImportService {
 
     const extracted = this.extractTitleAndRemoveHeading(prosemirrorState);
     const title = extracted.title;
-    // Imported markdown/HTML is built via markdownToHtml -> htmlToJson, which
+    // The markdown path now canonicalizes footnotes itself (the package parser),
-    // never runs the editor's footnoteSyncPlugin, so the footnote topology keeps
+    // but the HTML path (processHTML -> htmlToJson) does NOT run the editor's
-    // the source's PHYSICAL definition order (out of order vs. references),
+    // footnoteSyncPlugin, so an imported HTML doc can keep its source's PHYSICAL
-    // retains orphan definitions, and is not deduped. Canonicalize before
+    // definition order (out of order vs. references), retain orphan definitions,
-    // persisting so the stored page matches the editor's invariant (issue #228).
+    // and not be deduped. Canonicalize before persisting so the stored page
+    // matches the editor's invariant (issue #228); it is an idempotent no-op on
+    // the already-canonical markdown output.
     // Pure + idempotent + shape-safe: a doc with no footnotes is unchanged.
     // (Future consolidation, architecture B: this import path persists directly
     // via pageRepo.insertPage rather than through PageService.createPage, so the
@@ -133,12 +137,15 @@ export class ImportService {
   }
 
   async processMarkdown(markdownInput: string): Promise<any> {
-    try {
+    // Canonical markdown -> ProseMirror JSON directly via
-      const html = await markdownToHtml(markdownInput);
+    // `@docmost/prosemirror-markdown` (issue #345) — no HTML intermediate and no
-      return this.processHTML(html);
+    // second editor-ext markdown layer. Foreign markdown surfaces the strict
-    } catch (err) {
+    // canonical parser does not accept (GFM `[^id]` reference footnotes) are
-      throw err;
+    // rewritten to the canonical inline form by `normalizeForeignMarkdown` first.
-    }
+    // The HTML-cleanup pass (`normalizeImportHtml`) is intentionally skipped here:
+    // it targets foreign *HTML* (Notion/XWiki), which only ever arrives on the
+    // `.html` path (`processHTML`), never as canonical markdown.
+    return markdownToProseMirror(normalizeForeignMarkdown(markdownInput));
   }
 
   async processHTML(htmlInput: string): Promise<any> {

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

@@ -0,0 +1,218 @@
+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

@@ -0,0 +1,265 @@
+/**
+ * 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 `markdownToHtml`
+ * strip 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/metrics/http-metrics.hook.ts

@@ -2,15 +2,36 @@ import { FastifyReply, FastifyRequest } from 'fastify';
 import { isStreamingResponse } from './metrics.constants';
 import { observeHttp } from './metrics.registry';
 
+// URL path prefixes served by @fastify/static (client build output under
+// client/dist). `/assets/` holds the content-hashed bundle (index-*.js,
+// chunk-*.js) — a NEW set of names every deploy, i.e. an UNBOUNDED label set
+// (#362); the others (/vad/, /brand/, /locales/, /icons/ — copied verbatim from
+// public/) have stable names, so they are merely repetitive per-file labels
+// rather than unbounded. Either way none of these belong in the API-route
+// histogram: collapse them all to one bounded `static` label. (Edge latency for
+// static is already measured by Traefik's traefik_router_request_duration_*.)
+const STATIC_PATH_PREFIXES = [
+  '/assets/',
+  '/vad/',
+  '/brand/',
+  '/locales/',
+  '/icons/',
+];
+
 /**
  * Resolve the BOUNDED route label for an HTTP response.
  *
- * HARD REQUIREMENT (#355): use the ROUTE TEMPLATE (`/pages/:id`), NEVER the raw
+ * HARD REQUIREMENT (#355): use the ROUTE TEMPLATE (`/pages/:id`), NEVER a raw
- * URL (`/pages/abc-123`), so label cardinality stays finite. Fastify exposes the
+ * URL (`/pages/abc-123` or `/assets/index-CAbxDtto.js`), so label cardinality
- * matched template on `req.routeOptions.url`. On 404s (no route matched) that is
+ * stays finite. Fastify exposes the matched template on `req.routeOptions.url`,
- * missing → collapse to the literal `unknown`.
+ * BUT @fastify/static serves each file through a route whose matched url is the
+ * raw (hashed) file path — so for static assets that value is itself unbounded.
+ * Detect static requests by their path prefix FIRST and collapse to `static`;
+ * otherwise use the route template; on a 404 (no route matched) → `unknown`.
  */
 export function resolveRouteLabel(req: FastifyRequest): string {
+  const path = (req.url ?? '').split('?', 1)[0];
+  if (STATIC_PATH_PREFIXES.some((p) => path.startsWith(p))) return 'static';
   const url = req.routeOptions?.url;
   return typeof url === 'string' && url.length > 0 ? url : 'unknown';
 }

apps/server/src/integrations/metrics/metrics-bull.service.ts

@@ -46,7 +46,6 @@ export class MetricsBullService implements OnModuleInit, OnModuleDestroy {
     @InjectQueue(QueueName.GENERAL_QUEUE) generalQueue: Queue,
     @InjectQueue(QueueName.BILLING_QUEUE) billingQueue: Queue,
     @InjectQueue(QueueName.FILE_TASK_QUEUE) fileTaskQueue: Queue,
-    @InjectQueue(QueueName.SEARCH_QUEUE) searchQueue: Queue,
     @InjectQueue(QueueName.AI_QUEUE) aiQueue: Queue,
     @InjectQueue(QueueName.HISTORY_QUEUE) historyQueue: Queue,
     @InjectQueue(QueueName.NOTIFICATION_QUEUE) notificationQueue: Queue,
@@ -58,7 +57,6 @@ export class MetricsBullService implements OnModuleInit, OnModuleDestroy {
       { label: 'general', queue: generalQueue },
       { label: 'billing', queue: billingQueue },
       { label: 'file-task', queue: fileTaskQueue },
-      { label: 'search', queue: searchQueue },
       { label: 'ai', queue: aiQueue },
       { label: 'history', queue: historyQueue },
       { label: 'notification', queue: notificationQueue },

apps/server/src/integrations/metrics/metrics.spec.ts

@@ -25,6 +25,61 @@ describe('resolveRouteLabel (histogram route label)', () => {
     const req = { url: '/x' } as unknown as FastifyRequest;
     expect(resolveRouteLabel(req)).toBe('unknown');
   });
+
+  it.each([
+    '/assets/index-CAbxDtto.js',
+    '/assets/chunk-3OPIFGDE-CJOt9nr5.js',
+    '/assets/excalidraw-menu-DpsI0kFW.js',
+    '/vad/silero_vad_v5.onnx',
+    '/brand/logo.svg',
+    '/locales/en.json',
+    '/icons/app-icon-192x192.png',
+  ])('collapses hashed/static asset %p to "static" (#362 cardinality)', (url) => {
+    // @fastify/static serves each file through a route whose matched url is the
+    // raw (hashed) file path, so routeOptions.url is itself unbounded here.
+    const req = {
+      url,
+      routeOptions: { url },
+    } as unknown as FastifyRequest;
+    const label = resolveRouteLabel(req);
+    expect(label).toBe('static');
+    expect(label).not.toContain('.js');
+    expect(label).not.toContain('index-');
+  });
+
+  it('strips the query string before the static-prefix check', () => {
+    const req = {
+      url: '/assets/index-CAbxDtto.js?v=2',
+      routeOptions: { url: '/assets/index-CAbxDtto.js' },
+    } as unknown as FastifyRequest;
+    expect(resolveRouteLabel(req)).toBe('static');
+  });
+
+  it('does NOT collapse a real API route that merely mentions assets', () => {
+    // A templated API route is kept as-is; only the static path PREFIXES collapse.
+    const req = {
+      url: '/api/pages/assets-guide',
+      routeOptions: { url: '/api/pages/:id' },
+    } as unknown as FastifyRequest;
+    expect(resolveRouteLabel(req)).toBe('/api/pages/:id');
+  });
+
+  it.each([
+    // The TRAILING SLASH on the prefix is the anti-false-collapse guard: a path
+    // that is the prefix WITHOUT its slash, or merely shares the prefix as a
+    // substring of a longer segment, must NOT collapse. These would collapse
+    // under a buggy `includes('/assets/')` / slashless-prefix impl.
+    '/assets',
+    '/assetsx/foo.js',
+    '/iconset/x.png',
+  ])('does NOT collapse the prefix-boundary case %p', (url) => {
+    const req = {
+      url,
+      routeOptions: { url: '/some/:route' },
+    } as unknown as FastifyRequest;
+    expect(resolveRouteLabel(req)).not.toBe('static');
+    expect(resolveRouteLabel(req)).toBe('/some/:route');
+  });
 });
 
 describe('isStreamingResponse (SSE exclusion)', () => {

apps/server/src/integrations/queue/constants/queue.constants.ts

@@ -4,7 +4,6 @@ export enum QueueName {
   GENERAL_QUEUE = '{general-queue}',
   BILLING_QUEUE = '{billing-queue}',
   FILE_TASK_QUEUE = '{file-task-queue}',
-  SEARCH_QUEUE = '{search-queue}',
   AI_QUEUE = '{ai-queue}',
   HISTORY_QUEUE = '{history-queue}',
   NOTIFICATION_QUEUE = '{notification-queue}',
@@ -32,12 +31,6 @@ export enum QueueJob {
   IMPORT_TASK = 'import-task',
   EXPORT_TASK = 'export-task',
 
-  SEARCH_INDEX_PAGE = 'search-index-page',
-  SEARCH_INDEX_PAGES = 'search-index-pages',
-  SEARCH_INDEX_COMMENT = 'search-index-comment',
-  SEARCH_INDEX_COMMENTS = 'search-index-comments',
-  SEARCH_INDEX_ATTACHMENT = 'search-index-attachment',
-  SEARCH_INDEX_ATTACHMENTS = 'search-index-attachments',
   SEARCH_REMOVE_PAGE = 'search-remove-page',
   SEARCH_REMOVE_ASSET = 'search-remove-attachment',
   SEARCH_REMOVE_FACE = 'search-remove-comment',

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

@@ -57,14 +57,6 @@ import { GeneralQueueProcessor } from './processors/general-queue.processor';
         attempts: 1,
       },
     }),
-    BullModule.registerQueue({
-      name: QueueName.SEARCH_QUEUE,
-      defaultJobOptions: {
-        removeOnComplete: true,
-        removeOnFail: true,
-        attempts: 2,
-      },
-    }),
     BullModule.registerQueue({
       name: QueueName.AI_QUEUE,
       defaultJobOptions: {

apps/server/src/integrations/static/static.module.spec.ts

@@ -0,0 +1,35 @@
+import { resolveStaticAssetHeaders } from './static.module';
+
+// Unit tests for the static-asset cache classifier extracted from the
+// @fastify/static setHeaders callback (precedent: sandbox.controller.spec.ts).
+describe('resolveStaticAssetHeaders', () => {
+  it('marks a content-hashed /assets/ file immutable and sets Vary', () => {
+    const headers = resolveStaticAssetHeaders(
+      '/app/apps/client/dist/assets/index-a1b2c3.js',
+    );
+    expect(headers['cache-control']).toBe(
+      'public, max-age=31536000, immutable',
+    );
+    expect(headers['vary']).toBe('Accept-Encoding');
+  });
+
+  it('makes index.html always revalidate (never immutable)', () => {
+    const headers = resolveStaticAssetHeaders(
+      '/app/apps/client/dist/index.html',
+    );
+    expect(headers['cache-control']).toBe(
+      'no-cache, no-store, must-revalidate',
+    );
+    expect(headers['vary']).toBe('Accept-Encoding');
+  });
+
+  it('does NOT mark a non-hashed asset immutable but still sets Vary', () => {
+    const headers = resolveStaticAssetHeaders(
+      '/app/apps/client/dist/locales/en.json',
+    );
+    // No immutable cache-control — this path keeps @fastify/static's default
+    // etag/last-modified revalidation.
+    expect(headers['cache-control']).toBeUndefined();
+    expect(headers['vary']).toBe('Accept-Encoding');
+  });
+});

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

@@ -5,6 +5,46 @@ import * as fs from 'node:fs';
 import fastifyStatic from '@fastify/static';
 import { EnvironmentService } from '../environment/environment.service';
 
+/**
+ * Resolve the response headers for a statically served client asset.
+ *
+ * Extracted from the @fastify/static `setHeaders` callback so the cache
+ * classification stays a pure, unit-testable function (see
+ * static.module.spec.ts).
+ *
+ * `Vary: Accept-Encoding` is emitted for every static response because
+ * @fastify/static negotiates a precompressed .br/.gz neighbour by the client's
+ * Accept-Encoding but does NOT set Vary itself. Without it a shared/proxy cache
+ * keyed on the URL alone could store the brotli variant and later serve it to a
+ * client that only sent `Accept-Encoding: identity`/gzip → an undecodable body.
+ * This matters most for the immutable /assets/ files, which proxies may keep
+ * for a year.
+ */
+export function resolveStaticAssetHeaders(
+  filePath: string,
+): Record<string, string> {
+  const headers: Record<string, string> = { vary: 'Accept-Encoding' };
+
+  // Content-hashed files under /assets/ never change for a given URL, so they
+  // can be cached forever and skip revalidation entirely.
+  if (filePath.includes('/assets/')) {
+    headers['cache-control'] = 'public, max-age=31536000, immutable';
+    return headers;
+  }
+
+  // index.html is rewritten at boot (window.CONFIG injection) and on every
+  // deploy — it must be revalidated on every load.
+  if (filePath.endsWith('index.html')) {
+    headers['cache-control'] = 'no-cache, no-store, must-revalidate';
+    return headers;
+  }
+
+  // Everything else (locales, vad, icons, manifest) is NOT content-hashed and
+  // changes between deploys, so it keeps @fastify/static's default
+  // etag/last-modified revalidation — do NOT mark it immutable.
+  return headers;
+}
+
 @Module({})
 export class StaticModule implements OnModuleInit {
   constructor(
@@ -72,6 +112,16 @@ export class StaticModule implements OnModuleInit {
       await app.register(fastifyStatic, {
         root: clientDistPath,
         wildcard: false,
+        // Serve the build-time .br/.gz neighbour when the client accepts it
+        // (see vite-plugin-compression2 in apps/client/vite.config.ts).
+        preCompressed: true,
+        setHeaders: (res, filePath) => {
+          for (const [name, value] of Object.entries(
+            resolveStaticAssetHeaders(filePath),
+          )) {
+            res.setHeader(name, value);
+          }
+        },
       });
 
       app.get(RENDER_PATH, (req: any, res: any) => {

apps/server/src/main.ts

@@ -10,6 +10,7 @@ import { TransformHttpResponseInterceptor } from './common/interceptors/http-res
 import { WsRedisIoAdapter } from './ws/adapter/ws-redis.adapter';
 import fastifyMultipart from '@fastify/multipart';
 import fastifyCookie from '@fastify/cookie';
+import fastifyCompress from '@fastify/compress';
 import fastifyIp from 'fastify-ip';
 import { InternalLogFilter } from './common/logger/internal-log-filter';
 import { EnvironmentService } from './integrations/environment/environment.service';
@@ -63,6 +64,17 @@ async function bootstrap() {
   await app.register(fastifyIp);
   await app.register(fastifyMultipart);
   await app.register(fastifyCookie);
+  // Compress dynamic responses (API JSON, the rewritten share-SEO HTML) when the
+  // client accepts br/gzip. @fastify/compress only compresses content-types that
+  // mime-db flags `compressible` (application/json, text/html, …); `text/event-stream`
+  // is not in mime-db, so SSE is never compressed by the allowlist. The AI-chat
+  // stream additionally hijacks the raw socket (pipeUIMessageStreamToResponse ->
+  // res.raw in ai-chat.service.ts), bypassing Fastify's reply/onSend lifecycle
+  // entirely, so this hook can never buffer that stream.
+  await app.register(fastifyCompress, {
+    // Skip tiny payloads where compression overhead outweighs the savings.
+    threshold: 1024,
+  });
 
   const environmentService = app.get(EnvironmentService);
   const frameHeader = resolveFrameHeader(

apps/server/test/integration/db.ts

@@ -38,6 +38,24 @@ export const TEST_DATABASE_URL =
   process.env.TEST_DATABASE_URL ??
   'postgresql://docmost:docmost_dev_pw@localhost:5432/docmost_test';
 
+// Build the raw postgres.js client (mirrors database.module.ts: max pool,
+// silenced notices, bigint-as-number parsing). Kept separate so the singleton
+// can hold a reference to bound its shutdown in destroyTestDb.
+function buildTestSql(url: string = TEST_DATABASE_URL) {
+  return postgres(url, {
+    max: 5,
+    onnotice: () => {},
+    types: {
+      bigint: {
+        to: 20,
+        from: [20, 1700],
+        serialize: (value: number) => value.toString(),
+        parse: (value: string) => Number.parseInt(value),
+      },
+    },
+  });
+}
+
 /**
  * Build a Kysely instance that MIRRORS the app's setup in database.module.ts:
  * PostgresJSDialect over postgres(), CamelCasePlugin, and the bigint type
@@ -47,38 +65,40 @@ export const TEST_DATABASE_URL =
  */
 export function buildTestDb(url: string = TEST_DATABASE_URL): Kysely<any> {
   return new Kysely<any>({
-    dialect: new PostgresJSDialect({
+    dialect: new PostgresJSDialect({ postgres: buildTestSql(url) }),
-      postgres: postgres(url, {
-        max: 5,
-        onnotice: () => {},
-        types: {
-          bigint: {
-            to: 20,
-            from: [20, 1700],
-            serialize: (value: number) => value.toString(),
-            parse: (value: string) => Number.parseInt(value),
-          },
-        },
-      }),
-    }),
     plugins: [new CamelCasePlugin()],
   });
 }
 
 let singleton: Kysely<any> | undefined;
+let singletonSql: ReturnType<typeof buildTestSql> | undefined;
 
 /** Lazily-built shared Kysely for the test suite (one per worker; maxWorkers=1). */
 export function getTestDb(): Kysely<any> {
   if (!singleton) {
-    singleton = buildTestDb();
+    singletonSql = buildTestSql();
+    singleton = new Kysely<any>({
+      dialect: new PostgresJSDialect({ postgres: singletonSql }),
+      plugins: [new CamelCasePlugin()],
+    });
   }
   return singleton;
 }
 
 export async function destroyTestDb(): Promise<void> {
-  if (singleton) {
+  if (!singleton) return;
-    await singleton.destroy();
+  const sql = singletonSql;
-    singleton = undefined;
+  // Clear the refs first so a hung end() cannot leave a half-closed singleton.
+  singleton = undefined;
+  singletonSql = undefined;
+  // postgres.js .end() waits indefinitely for in-flight queries by default; a
+  // leaked/stuck pooled connection would hang the afterAll hook (a 60s hook
+  // timeout in CI). Bound the shutdown: the { timeout } grace period lets
+  // active queries drain, then force-closes lingering sockets so teardown
+  // always completes. We close the pool directly instead of Kysely.destroy()
+  // (which would call sql.end() again with no timeout).
+  if (sql) {
+    await sql.end({ timeout: 5 });
   }
 }
 

apps/server/test/jest-e2e.json

@@ -4,10 +4,14 @@
   "testEnvironment": "node",
   "testRegex": ".e2e-spec.ts$",
   "transform": {
+    "prosemirror-markdown/build/.+\\.js$": [
+      "babel-jest",
+      { "presets": [["@babel/preset-env", { "targets": { "node": "current" } }]] }
+    ],
     "^.+\\.(t|j)sx?$": ["ts-jest", { "tsconfig": { "allowJs": true } }]
   },
   "transformIgnorePatterns": [
-    "/node_modules/(?!(\\.pnpm/)?(nanoid|uuid|image-dimensions|marked|happy-dom|lib0|@sindresorhus[+/][a-z0-9-]+|escape-string-regexp|p-limit|yocto-queue)(@|/))"
+    "/node_modules/(?!(\\.pnpm/)?(nanoid|uuid|image-dimensions|marked|happy-dom|lib0|@sindresorhus[+/][a-z0-9-]+|escape-string-regexp|p-limit|yocto-queue|@docmost/prosemirror-markdown)(@|/))"
   ],
   "moduleNameMapper": {
     "^@docmost/db/(.*)$": "<rootDir>/../src/database/$1",

apps/server/test/jest-integration.json

@@ -4,14 +4,19 @@
   "testRegex": ".*\\.int-spec\\.ts$",
   "testPathIgnorePatterns": ["/node_modules/"],
   "transform": {
+    "prosemirror-markdown/build/.+\\.js$": [
+      "babel-jest",
+      { "presets": [["@babel/preset-env", { "targets": { "node": "current" } }]] }
+    ],
     "^.+\\.(t|j)sx?$": "ts-jest"
   },
   "transformIgnorePatterns": [
-    "/node_modules/(?!(\\.pnpm/)?(nanoid|uuid|image-dimensions|marked|happy-dom|lib0)(@|/))"
+    "/node_modules/(?!(\\.pnpm/)?(nanoid|uuid|image-dimensions|marked|happy-dom|lib0|@docmost/prosemirror-markdown)(@|/))"
   ],
   "testEnvironment": "node",
   "testTimeout": 60000,
   "maxWorkers": 1,
+  "forceExit": true,
   "globalSetup": "<rootDir>/test/integration/global-setup.ts",
   "globalTeardown": "<rootDir>/test/integration/global-teardown.ts",
   "moduleNameMapper": {

apps/server/test/stubs/tiptap-react.js

@@ -0,0 +1,23 @@
+// Jest stub for @tiptap/react.
+//
+// The server export/import code paths transitively import editor-ext, whose node
+// extensions import from `@tiptap/react`. The real module re-exports all of
+// `@tiptap/core` (headless, safe under node) AND adds React view helpers
+// (`ReactNodeViewRenderer`, …) that eagerly pull in react-dom — which throws
+// `navigator is not defined` under jest's node environment.
+//
+// So this stub DELEGATES to the real `@tiptap/core` (keeping `mergeAttributes`,
+// `Node`, `Mark`, `nodeInputRule`, … working — they are used by
+// `jsonToHtml`/`htmlToJson` on the server) and overrides ONLY the React view
+// helpers with no-ops. Those helpers are referenced solely inside `addNodeView()`
+// — code that runs only in a live browser editor, never on the server; if any
+// were actually invoked here it would (correctly) surface as a test failure.
+const core = require('@tiptap/core');
+
+module.exports = {
+  ...core,
+  ReactNodeViewRenderer: () => () => ({}),
+  NodeViewWrapper: () => null,
+  NodeViewContent: () => null,
+  ReactRenderer: class {},
+};

docker-compose.yml

@@ -12,6 +12,11 @@ services:
     ports:
       - "3000:3000"
     restart: unless-stopped
+    # The app already serves precompressed (brotli/gzip) static assets with
+    # long-lived cache headers and gzips dynamic API responses. For the best
+    # cold-load latency you can OPTIONALLY put a reverse proxy (caddy / nginx /
+    # traefik) in front with HTTP/2 (or HTTP/3) and brotli enabled — none is
+    # required for compression to work.
     volumes:
       - docmost:/app/data/storage
 

docs/dev-stand.md

@@ -131,5 +131,14 @@ const { Client } = require("pg");
 7. **Migrations don't auto-run in dev** — run `migration:latest` after every pull
    or branch switch.
 
+8. **Automation (Playwright): type into the BODY editor, not the title.** A page has
+   two `.ProseMirror` editors — `[aria-label='Page title']` (non-collab) and
+   `[aria-label='Page content']` (the collab body). `document.querySelector('.ProseMirror')`
+   returns the TITLE editor, so typing there never changes body content and `mod+S`
+   versions nothing. Target `[aria-label='Page content']`, confirm it's collab-bound
+   (`el.editor.extensionManager.extensions.some(e=>e.name==='collaboration')`), and
+   wait ~10-12s for the store debounce before asserting `pages.content` changed. Full
+   testing methodology + traps: **[how-to-test.md](how-to-test.md)**.
+
 See also the **Commands** and **Architecture → Two server processes** sections in
 [`AGENTS.md`](../AGENTS.md).

docs/how-to-test.md

@@ -0,0 +1,108 @@
+# How to test the application (browser E2E + out-of-band)
+
+How to actually verify a feature end-to-end against a running stand — driving the
+**real app in a browser** and confirming results **out-of-band** in the DB/git, not
+through the same API you're supposed to be testing. Written from real false-positives
+that wasted hours (see **Traps** — read them before you write a test).
+
+Prereq: a running stand — see **[dev-stand.md](dev-stand.md)**. Automation uses
+Playwright (`pip install playwright && python -m playwright install chromium`).
+
+## Principles
+
+1. **Drive the behaviour under test through the browser.** The stand exists so you
+   exercise the real UI + realtime-collab + server path. Using `POST /api/pages/*` to
+   perform the action you're validating tests the API, not the app — an e2e suite can
+   do that. API calls are fine ONLY for one-time setup/fixtures, never for the
+   interaction you're asserting on.
+2. **Evidence before claim.** Nothing "passes" without an artifact: a DB row, a git
+   diff, a screenshot looked at as an image. If you can't show it, you didn't verify it.
+3. **Verify out-of-band.** Judge results from a source independent of the UI: `psql`
+   against the DB, a fresh `git clone` of a synced repo, a hard reload. Optimistic UI
+   lies about persistence.
+4. **Disconfirm by default.** For each feature, actively try to prove it's broken
+   before concluding it works. Reload after every create/edit/save.
+5. **Recon actuatability FIRST.** Before building editor tests, confirm the
+   interaction even works in your harness (does a typed edit reach the DB?). Skipping
+   this is how you ship a pile of tests that all silently exercised the wrong thing.
+
+## The editor: two ProseMirror instances (READ THIS)
+
+A page has **two** `.ProseMirror` editors:
+
+| index | selector | role | collab? |
+|---|---|---|---|
+| 0 | `[aria-label='Page title']` | title field | **NO** (16 exts, no `collaboration`) |
+| 1 | `[aria-label='Page content']` | body | **YES** (95 exts, has `collaboration`) |
+
+`document.querySelector('.ProseMirror')` returns the **title** editor (first match).
+Type there and you edit the title only — body page content never changes, so `mod+S`
+"versions" unchanged content and every content test silently no-ops.
+
+**Always target the body editor** and confirm it's collab-bound before typing:
+
+```js
+const el = document.querySelector("[aria-label='Page content']");
+el.editor.extensionManager.extensions.some(e => e.name === 'collaboration'); // must be true
+```
+
+Body edits emit ~20 `/collab` websocket frames while typing and land in
+`pages.content` after the **hocuspocus store debounce (~10s)** — so **wait ~12s**
+before asserting persistence (checking at 6–8s is a false negative). `mod+S` (the
+`save-version` stateless message) flushes immediately, so a version created right
+after a settled body edit holds the typed text.
+
+## A known-good browser flow
+
+```
+1. goto /s/<space-slug>            # the "Create page" button lives in the space sidebar, not /home
+2. click button[aria-label='Create page']   # fully UI-driven page creation
+3. type into [aria-label='Page title']       # optional title
+4. click [aria-label='Page content'] → type body text
+5. wait ~12s (store debounce)
+6. assert pages.content changed (psql)       # out-of-band
+7. mod+S / menu Save → assert page_history row (psql)
+8. reload / fresh context → re-assert (persistence round-trip)
+```
+
+Auth: log in ONCE, save `storage_state.json`, reuse it across pages/agents (re-login
+per run trips shared rate-limits). Cookie-based session authorizes both REST and the
+collab websocket.
+
+## Judging out-of-band
+
+```bash
+# page content / history
+docker exec <db> psql -U docmost -d docmost -tAc \
+  "select coalesce(kind,'null'), content::text from page_history where page_id='<id>' order by created_at;"
+# git-sync round-trip: clone the space repo and diff against what you pushed
+git clone http://<user>:<pass>@127.0.0.1:3000/git/<spaceId>.git /tmp/x
+```
+
+`page_history.content` is full JSON — parse it, don't truncate the snippet, or a
+marker check misses. For sync/async features (autosave, git-sync, idle-flush) use an
+active probe: write a unique marker, wait past the debounce/poll window, re-read
+out-of-band, ≥2 iterations — never conclude "broken" from a single snapshot.
+
+## Traps (each of these produced a false result in a real run)
+
+- **Wrong editor.** Typed into `.ProseMirror` (= title). Edits never touched body
+  content. → target `[aria-label='Page content']`.
+- **Checked persistence too early.** Store debounce ~10s; a 6–8s check reads stale.
+- **Truncated the DB snapshot** below where the test marker sits → false "content
+  missing".
+- **API-seeded the content under test**, then "verified" the feature — that validated
+  the API, not the app.
+- **Reused a fixed marker on a non-rebooted stand** → title/row collisions inflate
+  counts (`count==2`). Use a unique per-run marker (timestamp).
+- **Idle/async read once** and called it "permanently broken" — it was mid-debounce.
+- **Concluded env-limitation without a cross-build control.** If unsure whether a
+  failure is your harness or the product, run the SAME harness against a known-good
+  build; a divergence localizes it.
+
+## Scope note
+
+Some paths genuinely need a human in a real browser (rich drag-drop, native file
+pickers, clipboard, and anything the harness can't actuate). Label those UNTESTED in
+the report — "handled gracefully" is not "works". Keep four states distinct:
+verified-working, defect, untested, env-limitation.

package.json

@@ -96,7 +96,8 @@
   "pnpm": {
     "patchedDependencies": {
       "scimmy@1.3.5": "patches/scimmy@1.3.5.patch",
-      "yjs@13.6.30": "patches/yjs@13.6.30.patch"
+      "yjs@13.6.30": "patches/yjs@13.6.30.patch",
+      "ai@6.0.134": "patches/ai@6.0.134.patch"
     },
     "overrides": {
       "prosemirror-changeset": "2.4.0",

packages/mcp/src/index.ts

@@ -118,56 +118,19 @@ export function createDocmostMcpServer(config: DocmostMcpConfig): McpServer {
 // transport exposes a `tree:true` mode that returns the full nested hierarchy;
 // the in-app copy keeps the same tree option but is worded for the in-app agent.
 // Kept per-layer so each side can tune its own guidance.
-server.registerTool(
+// Schema + description now live in @docmost/mcp's SHARED_TOOL_SPECS (#294). This
-  "list_pages",
+// transport keeps applying its own defaults (limit=50, tree=false) in execute.
-  {
+registerShared(SHARED_TOOL_SPECS.listPages, async ({ spaceId, limit, tree }) => {
-    description:
+  const result = await docmostClient.listPages(spaceId, limit ?? 50, tree ?? false);
-      "List most recent pages in a space ordered by updatedAt (descending). " +
+  return jsonContent(result);
-      "Returns a bounded list (default 50, max 100) — use search for lookups " +
+});
-      "in large spaces. Pass tree:true (with spaceId) to instead get the " +
-      "space's full page hierarchy as a nested tree.",
-    inputSchema: {
-      spaceId: z.string().optional(),
-      limit: z
-        .number()
-        .int()
-        .min(1)
-        .max(100)
-        .optional()
-        .describe("Max pages to return (default 50, max 100)"),
-      tree: z
-        .boolean()
-        .optional()
-        .describe(
-          "When true, return the space's full page hierarchy as a nested tree (each node has a children array) instead of the recent-by-updatedAt flat list. Requires spaceId; ignores limit.",
-        ),
-    },
-  },
-  async ({ spaceId, limit, tree }) => {
-    const result = await docmostClient.listPages(spaceId, limit ?? 50, tree ?? false);
-    return jsonContent(result);
-  },
-);
 
 // Tool: get_page
-server.registerTool(
+// Schema + description now live in the shared registry (#294).
-  "get_page",
+registerShared(SHARED_TOOL_SPECS.getPage, async ({ pageId }) => {
-  {
+  const page = await docmostClient.getPage(pageId);
-    description:
+  return jsonContent(page);
-      "Get page details with content converted to Markdown. The conversion is " +
+});
-      "LOSSY (block ids, exact table/callout structure are approximated); for a " +
-      "lossless representation use get_page_json. Inline <span data-comment-id> " +
-      "tags in the markdown are comment highlight anchors (also present for " +
-      "RESOLVED threads) — treat them as markup, not page text.",
-    inputSchema: {
-      pageId: z.string().min(1),
-    },
-  },
-  async ({ pageId }) => {
-    const page = await docmostClient.getPage(pageId);
-    return jsonContent(page);
-  },
-);
 
 // Tool: get_page_json
 registerShared(SHARED_TOOL_SPECS.getPageJson, async ({ pageId }) => {
@@ -201,6 +164,10 @@ registerShared(
 );
 
 // Tool: table_get
+// NOT in the shared registry: the MCP tool name `table_get` is noun-first while
+// the in-app key is `getTable` (verb-first), breaking the snake_case(inAppKey)
+// convention the shared registry enforces (shared-tool-specs.contract.spec.ts).
+// Renaming the public MCP tool would break external clients, so it stays inline.
 server.registerTool(
   "table_get",
   {
@@ -223,25 +190,10 @@ server.registerTool(
 );
 
 // Tool: table_insert_row
-// NOT in the shared registry: this transport names the table argument `table`,
+// Schema + description now live in the shared registry (#294); the `table`
-// while the in-app tool names it `tableRef` (ai-chat-tools.service.ts). Sharing
+// parameter name is the canonical one (the in-app layer was unified to it).
-// one buildShape would rename a public MCP parameter, so the table row/cell
+registerShared(
-// tools stay per-transport by design.
+  SHARED_TOOL_SPECS.tableInsertRow,
-server.registerTool(
-  "table_insert_row",
-  {
-    description:
-      "Insert a row of plain-text cells into a table. `table` = `#<index>` or " +
-      "a block id inside it. `cells` = text per column (padded to the table's " +
-      "column count; error if more cells than columns). `index` = 0-based " +
-      "insert position (0 inserts before the header); omit to append at the end.",
-    inputSchema: {
-      pageId: z.string().min(1),
-      table: z.string().min(1),
-      cells: z.array(z.string()),
-      index: z.number().int().optional(),
-    },
-  },
   async ({ pageId, table, cells, index }) => {
     const result = await docmostClient.tableInsertRow(
       pageId,
@@ -254,22 +206,9 @@ server.registerTool(
 );
 
 // Tool: table_delete_row
-// NOT shared — same `table` (here) vs `tableRef` (in-app) parameter-name
+// Schema + description now live in the shared registry (#294).
-// divergence as table_insert_row.
+registerShared(
-server.registerTool(
+  SHARED_TOOL_SPECS.tableDeleteRow,
-  "table_delete_row",
-  {
-    description:
-      "Delete the row at 0-based `index` from a table (`table` = `#<index>` or " +
-      "a block id inside it). Refuses to delete the table's only row. An " +
-      "out-of-range `index` throws. Deleting `index` 0 removes the header row, " +
-      "and the next row becomes the new header.",
-    inputSchema: {
-      pageId: z.string().min(1),
-      table: z.string().min(1),
-      index: z.number().int(),
-    },
-  },
   async ({ pageId, table, index }) => {
     const result = await docmostClient.tableDeleteRow(pageId, table, index);
     return jsonContent(result);
@@ -277,24 +216,9 @@ server.registerTool(
 );
 
 // Tool: table_update_cell
-// NOT shared — same `table` (here) vs `tableRef` (in-app) parameter-name
+// Schema + description now live in the shared registry (#294).
-// divergence as table_insert_row.
+registerShared(
-server.registerTool(
+  SHARED_TOOL_SPECS.tableUpdateCell,
-  "table_update_cell",
-  {
-    description:
-      "Set the plain-text content of cell [row,col] (0-based) in a table " +
-      "(`table` = `#<index>` or a block id inside it). Replaces the cell's " +
-      "content with a single text paragraph; for rich formatting use patch_node " +
-      "on the cell's paragraph id from table_get.",
-    inputSchema: {
-      pageId: z.string().min(1),
-      table: z.string().min(1),
-      row: z.number().int(),
-      col: z.number().int(),
-      text: z.string(),
-    },
-  },
   async ({ pageId, table, row, col, text }) => {
     const result = await docmostClient.tableUpdateCell(
       pageId,
@@ -308,22 +232,9 @@ server.registerTool(
 );
 
 // Tool: create_page
-server.registerTool(
+// Schema + description now live in the shared registry (#294).
-  "create_page",
+registerShared(
-  {
+  SHARED_TOOL_SPECS.createPage,
-    description:
-      "Create a new page from Markdown in a space. Pass parentPageId to nest " +
-      "it under a parent; omit it to create at the space root.",
-    inputSchema: {
-      title: z.string().min(1).describe("Title of the page"),
-      content: z.string().min(1).describe("Markdown content"),
-      spaceId: z.string().min(1),
-      parentPageId: z
-        .string()
-        .optional()
-        .describe("Optional parent page ID to nest under"),
-    },
-  },
   async ({ title, content, spaceId, parentPageId }) => {
     const result = await docmostClient.createPage(
       title,
@@ -336,32 +247,11 @@ server.registerTool(
 );
 
 // Tool: update_page_json
-server.registerTool(
+// Schema + description now live in the shared registry (#294). The execute body
-  "update_page_json",
+// keeps this transport's content normalization (parse a JSON-string content,
-  {
+// pass undefined/null through for a title-only/no-op update).
-    description:
+registerShared(
-      "Replace a page's content with a raw ProseMirror JSON document " +
+  SHARED_TOOL_SPECS.updatePageJson,
-      "(lossless write: preserves the block ids, callouts, tables and " +
-      "attributes you pass in). Typical flow: get_page_json -> modify the " +
-      "JSON -> update_page_json. Keep existing node ids intact so heading " +
-      "anchors and history stay stable. Minimal full-doc example: " +
-      '{"type":"doc","content":[{"type":"paragraph","content":' +
-      '[{"type":"text","text":"Hi"}]}]}. `content` may be a JSON object or a ' +
-      "JSON string (both accepted), and is OPTIONAL: omit it to update only " +
-      "the title (though prefer rename_page for a title-only change). " +
-      "Supplying neither content nor title is an error.",
-    inputSchema: {
-      pageId: z.string().min(1).describe("ID of the page to update"),
-      content: z
-        .any()
-        .optional()
-        .describe(
-          'ProseMirror document {"type":"doc","content":[...]} (JSON object or ' +
-            "JSON string). Omit to rename only.",
-        ),
-      title: z.string().optional().describe("Optional new title"),
-    },
-  },
   async ({ pageId, content, title }) => {
     // Only parse/validate the document when it was actually supplied; when it
     // is omitted, pass it straight through so the client performs a title-only
@@ -379,26 +269,11 @@ server.registerTool(
 );
 
 // Tool: export_page_markdown
-server.registerTool(
+// Schema + description now live in the shared registry (#294).
-  "export_page_markdown",
+registerShared(SHARED_TOOL_SPECS.exportPageMarkdown, async ({ pageId }) => {
-  {
+  const md = await docmostClient.exportPageMarkdown(pageId);
-    description:
+  return { content: [{ type: "text" as const, text: md }] };
-      "Export a page to a single self-contained, lossless Docmost-flavoured " +
+});
-      "Markdown file (custom extensions): YAML-free meta header, body with " +
-      "inline comment anchors and diagrams, and a trailing comments-thread " +
-      "block. Designed for a download -> edit body -> import_page_markdown " +
-      "round-trip that preserves everything, including comment highlights. " +
-      "Comment THREADS are preserved in the file but are not re-pushed to the " +
-      "server on import.",
-    inputSchema: {
-      pageId: z.string().min(1),
-    },
-  },
-  async ({ pageId }) => {
-    const md = await docmostClient.exportPageMarkdown(pageId);
-    return { content: [{ type: "text" as const, text: md }] };
-  },
-);
 
 // Tool: import_page_markdown
 registerShared(
@@ -422,22 +297,11 @@ registerShared(
 );
 
 // Tool: rename_page
-server.registerTool(
+// Schema + description now live in the shared registry (#294).
-  "rename_page",
+registerShared(SHARED_TOOL_SPECS.renamePage, async ({ pageId, title }) => {
-  {
+  const result = await docmostClient.renamePage(pageId, title);
-    description:
+  return jsonContent(result);
-      "Rename a page (change its title only) without touching or resending " +
+});
-      "its content.",
-    inputSchema: {
-      pageId: z.string().min(1).describe("ID of the page to rename"),
-      title: z.string().min(1).describe("New title"),
-    },
-  },
-  async ({ pageId, title }) => {
-    const result = await docmostClient.renamePage(pageId, title);
-    return jsonContent(result);
-  },
-);
 
 // Tool: edit_page_text
 registerShared(SHARED_TOOL_SPECS.editPageText, async ({ pageId, edits }) => {
@@ -516,6 +380,10 @@ registerShared(SHARED_TOOL_SPECS.deleteNode, async ({ pageId, nodeId }) => {
 });
 
 // Tool: insert_image
+// MCP-only by design (NOT in the shared registry): the in-app AI-chat agent
+// exposes no image tools (insert/replace), so there is no second layer to unify
+// — a SHARED_TOOL_SPECS entry's tier/catalogLine are in-app metadata and the
+// catalog-partition test forbids a spec without a live in-app tool (#294).
 server.registerTool(
   "insert_image",
   {
@@ -561,6 +429,7 @@ server.registerTool(
 );
 
 // Tool: replace_image
+// MCP-only by design (see insert_image): no in-app equivalent, stays inline.
 server.registerTool(
   "replace_image",
   {
@@ -603,25 +472,10 @@ server.registerTool(
 );
 
 // Tool: share_page
-// INTENTIONAL per-transport divergence (not shared): the in-app copy adds a
+// Schema + description now live in the shared registry (#294). The execute body
-// security-confirmation framing ("only share when the user explicitly asked,
+// keeps this transport's own `searchIndexing ?? true` default.
-// since this exposes the page to anyone with the link") tuned for the in-app
+registerShared(
-// agent; this transport keeps the plain public-URL wording.
+  SHARED_TOOL_SPECS.sharePage,
-server.registerTool(
-  "share_page",
-  {
-    description:
-      "Make a page publicly accessible (idempotent) and return its public " +
-      "URL. The URL format is <app>/share/<key>/p/<slugId>. This exposes the " +
-      "page content to ANYONE with the URL — do it only when explicitly asked.",
-    inputSchema: {
-      pageId: z.string().min(1).describe("ID of the page to share"),
-      searchIndexing: z
-        .boolean()
-        .optional()
-        .describe("Allow search engines to index the page (default true)"),
-    },
-  },
   async ({ pageId, searchIndexing }) => {
     const result = await docmostClient.sharePage(pageId, searchIndexing ?? true);
     return jsonContent(result);
@@ -641,29 +495,11 @@ registerShared(SHARED_TOOL_SPECS.listShares, async () => {
 });
 
 // Tool: move_page
-server.registerTool(
+// Schema + description now live in the shared registry (#294). The execute body
-  "move_page",
+// keeps this transport's cycle guard, its 'null'/'' -> null string coercion, and
-  {
+// its positive-confirmation check on the move response.
-    description:
+registerShared(
-      "Move a page under a new parent (nesting) or to the space root.",
+  SHARED_TOOL_SPECS.movePage,
-    inputSchema: {
-      pageId: z.string().min(1),
-      parentPageId: z
-        .string()
-        .nullable()
-        .optional()
-        .describe(
-          "Target parent page ID. Pass 'null' or empty string to move to root.",
-        ),
-      position: z
-        .string()
-        .min(5)
-        .optional()
-        .describe(
-          "fractional-index position key; min 5 chars; omit to append at the end.",
-        ),
-    },
-  },
   async ({ pageId, parentPageId, position }) => {
     const finalParentId =
       parentPageId === "" || parentPageId === "null" ? null : parentPageId;
@@ -698,49 +534,22 @@ server.registerTool(
 );
 
 // Tool: delete_page
-server.registerTool(
+// Schema + description now live in the shared registry (#294). The shared schema
-  "delete_page",
+// exposes ONLY pageId, so no permanent/force-delete flag can reach the client.
-  {
+registerShared(SHARED_TOOL_SPECS.deletePage, async ({ pageId }) => {
-    description:
+  await docmostClient.deletePage(pageId);
-      "Delete a single page by ID. SOFT delete only: the page is moved to " +
+  return {
-      "trash and can be restored; nothing is permanently deleted.",
+    content: [
-    inputSchema: {
+      { type: "text" as const, text: `Successfully deleted page ${pageId}` },
-      pageId: z.string().min(1),
+    ],
-    },
+  };
-  },
+});
-  async ({ pageId }) => {
-    await docmostClient.deletePage(pageId);
-    return {
-      content: [
-        { type: "text" as const, text: `Successfully deleted page ${pageId}` },
-      ],
-    };
-  },
-);
 
 // --- Comment tools (ported from upstream PR #3 by Max Nikitin) ---
 
 // Tool: list_comments
-server.registerTool(
+registerShared(
-  "list_comments",
+  SHARED_TOOL_SPECS.listComments,
-  {
-    description:
-      "List comments on a page in one call (pagination is handled " +
-      "internally). By DEFAULT only ACTIVE threads are returned; resolved " +
-      "threads (a resolved top-level comment and all its replies) are hidden " +
-      "and their count reported as `resolvedThreadsHidden` so you can re-query " +
-      "with `includeResolved: true` to see everything. Returns " +
-      "`{ items, resolvedThreadsHidden }`. Content is returned as Markdown.",
-    inputSchema: {
-      pageId: z.string().describe("ID of the page"),
-      includeResolved: z
-        .boolean()
-        .optional()
-        .describe(
-          "default only active threads; true — include resolved",
-        ),
-    },
-  },
   async ({ pageId, includeResolved }) => {
     const comments = await docmostClient.listComments(pageId, includeResolved);
     return jsonContent(comments);
@@ -748,55 +557,11 @@ server.registerTool(
 );
 
 // Tool: create_comment
-// INTENTIONAL per-transport divergence (not shared): the in-app copy tunes the
+// Schema + description now live in the shared registry (#294). The execute body
-// guidance for the in-app agent (e.g. "retry with a corrected EXACT selection"
+// keeps this transport's own guards (require a selection for a top-level
-// and "Reversible via the comment UI"); this transport keeps its own wording.
+// comment; reject suggestedText on a reply / without a selection).
-server.registerTool(
+registerShared(
-  "create_comment",
+  SHARED_TOOL_SPECS.createComment,
-  {
-    description:
-      "Create a new comment on a page. The comment is ALWAYS inline and is " +
-      "anchored to (highlights) its `selection` text — there are no page-level " +
-      "comments. Content is provided as Markdown and automatically converted. " +
-      "A top-level comment REQUIRES an exact `selection`; if the selection " +
-      "cannot be found in the page the call fails (no orphan comment is left). " +
-      "Replies (parentCommentId set) inherit the parent's anchor and take no " +
-      "selection. You may also attach a `suggestedText` proposing a replacement " +
-      "for the `selection`; a human applies (or rejects) it from the UI. When " +
-      "`suggestedText` is set the `selection` MUST occur exactly once in the " +
-      "page — expand it with surrounding context if it is ambiguous.",
-    inputSchema: {
-      pageId: z.string().describe("ID of the page to comment on"),
-      content: z.string().min(1).describe("Comment content in Markdown format"),
-      selection: z
-        .string()
-        .min(1)
-        // Enforce the documented 250-char cap to match the description above.
-        .max(250)
-        .optional()
-        .describe(
-          "EXACT contiguous text from a single paragraph/block to anchor the " +
-            "comment on (<=250 chars). Required for a top-level comment; omit " +
-            "only when replying via parentCommentId.",
-        ),
-      parentCommentId: z
-        .string()
-        .optional()
-        .describe("Parent comment ID to create a reply (max 2 nesting levels)"),
-      suggestedText: z
-        .string()
-        .min(1)
-        .max(2000)
-        .optional()
-        .describe(
-          "Optional proposed replacement (PLAIN TEXT) for the `selection`, " +
-            "applied by a human via the UI (never auto-applied). REQUIRES a " +
-            "`selection`; NOT allowed on a reply. When set, the `selection` must " +
-            "be UNIQUE in the page — expand it with surrounding context (still " +
-            "<=250 chars) if it occurs more than once, or the call is refused.",
-        ),
-    },
-  },
   async ({ pageId, content, selection, parentCommentId, suggestedText }) => {
     if (!parentCommentId && (!selection || !selection.trim())) {
       throw new Error(
@@ -872,28 +637,9 @@ server.registerTool(
 );
 
 // Tool: resolve_comment
-server.registerTool(
+// Schema + description now live in the shared registry (#294).
-  "resolve_comment",
+registerShared(
-  {
+  SHARED_TOOL_SPECS.resolveComment,
-    description:
-      "Resolve (close) or reopen a comment thread. Only top-level comments can " +
-      "be resolved — the server rejects resolving a reply. Reversible: pass " +
-      "resolved=false to reopen. Resolving keeps the thread and its replies " +
-      "(unlike delete_comment, which permanently removes them).",
-    inputSchema: {
-      commentId: z
-        .string()
-        .min(1)
-        .describe("ID of the top-level comment thread to resolve or reopen"),
-      resolved: z
-        .boolean()
-        .optional()
-        .default(true)
-        .describe(
-          "true (default) marks the thread resolved/closed; false reopens it",
-        ),
-    },
-  },
   async ({ commentId, resolved }) => {
     const result = await docmostClient.resolveComment(commentId, resolved);
     return jsonContent(result);
@@ -901,30 +647,10 @@ server.registerTool(
 );
 
 // Tool: check_new_comments
-server.registerTool(
+// Schema + description now live in the shared registry (#294). The execute body
-  "check_new_comments",
+// keeps this transport's own guard rejecting an unparseable `since` timestamp.
-  {
+registerShared(
-    description:
+  SHARED_TOOL_SPECS.checkNewComments,
-      "Check for new comments across pages in a space since a given timestamp. " +
-      "Optionally scope to a page subtree (folder). Returns only comments " +
-      "created after the specified time.",
-    inputSchema: {
-      spaceId: z.string().describe("Space ID to check for new comments"),
-      since: z
-        .string()
-        .min(1)
-        .describe(
-          "ISO 8601 timestamp — only return comments created after this time (e.g. '2026-03-10T00:00:00Z')",
-        ),
-      parentPageId: z
-        .string()
-        .optional()
-        .describe(
-          "Optional root page ID to scope the check to a subtree (folder). " +
-            "Only pages under this parent will be checked.",
-        ),
-    },
-  },
   async ({ spaceId, since, parentPageId }) => {
     // Reject an unparseable timestamp up front: otherwise the comparison
     // against NaN silently treats every comment as "not new" and the tool
@@ -1053,6 +779,8 @@ server.registerTool(
 );
 
 // Tool: insert_footnote
+// MCP-only by design (see insert_image): the in-app AI-chat agent exposes no
+// footnote tool, so there is no second layer to unify — stays inline (#294).
 server.registerTool(
   "insert_footnote",
   {

packages/mcp/src/tool-specs.ts

@@ -316,6 +316,34 @@ export const SHARED_TOOL_SPECS = {
 
   // --- share management ---
 
+  // Unified from the per-layer inline definitions (#294). Both layers already
+  // carried the "only share when explicitly asked" security framing (the
+  // "per-transport divergence" note on the old inline copies was stale), so
+  // there was no real behavioral divergence to preserve — only wording drift.
+  sharePage: {
+    mcpName: 'share_page',
+    inAppKey: 'sharePage',
+    // CANONICAL: merges the MCP copy's URL-format + idempotency detail with the
+    // in-app copy's reversibility note; keeps the security framing both had.
+    description:
+      'Make a page PUBLICLY accessible (idempotent) and return its public URL ' +
+      '(format: <app>/share/<key>/p/<slugId>). This exposes the page content ' +
+      'to ANYONE with the URL — only share when the user explicitly asked. ' +
+      'Reversible: unshare it later to revoke the public URL.',
+    tier: 'deferred',
+    catalogLine: 'sharePage — make a page publicly accessible and return its URL.',
+    // Reconciled: MCP's stricter .min(1) on pageId kept; field descriptions from
+    // the in-app copy. The MCP execute keeps its own `searchIndexing ?? true`
+    // default (a per-layer concern, not part of the shared schema).
+    buildShape: (z) => ({
+      pageId: z.string().min(1).describe('The id of the page to share.'),
+      searchIndexing: z
+        .boolean()
+        .optional()
+        .describe('Allow public search engines to index it (default true).'),
+    }),
+  },
+
   unsharePage: {
     mcpName: 'unshare_page',
     inAppKey: 'unsharePage',
@@ -509,4 +537,470 @@ export const SHARED_TOOL_SPECS = {
       pageId: z.string().min(1),
     }),
   },
+
+  // --- page tools (unified from the per-layer inline definitions, #294) ---
+  //
+  // Descriptions merge both layers (the MCP copy's richer structural notes + the
+  // in-app copy's "Reversible via history/trash" framing where it added one).
+  // Field constraints keep the MCP copy's stricter .min(1) EXCEPT where the
+  // in-app layer deliberately allowed a looser value (documented per field).
+
+  getPage: {
+    mcpName: 'get_page',
+    inAppKey: 'getPage',
+    description:
+      'Fetch a single page as Markdown by its id. Returns the page title and ' +
+      'its Markdown content. The Markdown conversion is LOSSY (block ids, exact ' +
+      'table/callout structure are approximated); for a lossless representation ' +
+      'use the lossless page-JSON read tool. Inline <span data-comment-id> tags in the markdown ' +
+      'are comment highlight anchors (also present for RESOLVED threads) — ' +
+      'treat them as markup, not page text.',
+    tier: 'core',
+    catalogLine: 'getPage — fetch a page as Markdown by its id.',
+    // Reconciled: MCP's stricter .min(1) kept; in-app's more-informative
+    // "(or slugId)" describe kept.
+    buildShape: (z) => ({
+      pageId: z.string().min(1).describe('The id (or slugId) of the page.'),
+    }),
+  },
+
+  listPages: {
+    mcpName: 'list_pages',
+    inAppKey: 'listPages',
+    description:
+      'List the most recent pages (ordered by updatedAt, descending), ' +
+      'optionally scoped to a single space. Returns a bounded list (default ' +
+      '50, max 100) — use search for lookups in large spaces. Pass tree:true ' +
+      "(with spaceId) to instead get the space's full page hierarchy as a " +
+      'nested tree.',
+    tier: 'core',
+    catalogLine: "listPages — list recent pages, or a space's full page tree.",
+    buildShape: (z) => ({
+      spaceId: z
+        .string()
+        .optional()
+        .describe('Optional space id to scope the listing to.'),
+      limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(100)
+        .optional()
+        .describe('Maximum number of pages (default 50, max 100).'),
+      tree: z
+        .boolean()
+        .optional()
+        .describe(
+          "When true, return the space's full page hierarchy as a nested tree " +
+            '(children arrays) instead of the recent-by-updatedAt flat list. ' +
+            'Requires spaceId; ignores limit.',
+        ),
+    }),
+  },
+
+  createPage: {
+    mcpName: 'create_page',
+    inAppKey: 'createPage',
+    description:
+      'Create a new page with a Markdown body in a space, optionally under a ' +
+      'parent page (omit parentPageId to create at the space root). Returns ' +
+      'the new page id and title. Reversible: a page can be moved to trash ' +
+      'later.',
+    tier: 'deferred',
+    catalogLine: 'createPage — create a new page with a Markdown body in a space.',
+    // Reconciled schema DRIFT: the MCP copy pinned `content` to .min(1) while
+    // the in-app copy left it unbounded and DOCUMENTS an empty body as valid
+    // ("may be empty") — creating an empty page to fill in later is a real use
+    // case. The looser (no-min) form is kept, so create_page now also accepts an
+    // empty body (harmless — it creates an empty page) and no previously-valid
+    // in-app input is ever rejected. `title`/`spaceId` keep the MCP .min(1)
+    // (an empty title or space is never valid).
+    buildShape: (z) => ({
+      title: z.string().min(1).describe('The title of the new page.'),
+      content: z.string().describe('The page body as Markdown (may be empty).'),
+      spaceId: z.string().min(1).describe('The id of the space to create the page in.'),
+      parentPageId: z
+        .string()
+        .optional()
+        .describe('Optional parent page id to nest the new page under.'),
+    }),
+  },
+
+  movePage: {
+    mcpName: 'move_page',
+    inAppKey: 'movePage',
+    description:
+      'Move a page under a new parent page, or to the space root when no ' +
+      'parent is given. Reversible: move it back at any time.',
+    tier: 'deferred',
+    catalogLine: 'movePage — move a page under a new parent or to the space root.',
+    // Reconciled schema DRIFT: the MCP copy exposed a `position` field
+    // (fractional-index ordering) that 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) — it is optional, so no
+    // previously-valid in-app call is rejected. `parentPageId` is `.nullable()`
+    // on both, so a real JSON null moves to root on either transport; the MCP
+    // execute additionally coerces the strings 'null'/'' to null as a robustness
+    // fallback (kept in its execute body, not in the shared schema).
+    buildShape: (z) => ({
+      pageId: z.string().min(1).describe('The id of the page to move.'),
+      parentPageId: z
+        .string()
+        .nullable()
+        .optional()
+        .describe(
+          'Target parent page id. Null or omitted moves the page to the space ' +
+            'root.',
+        ),
+      position: z
+        .string()
+        .min(5)
+        .optional()
+        .describe(
+          'Optional fractional-index position key (min 5 chars); omit to ' +
+            'append at the end.',
+        ),
+    }),
+  },
+
+  renamePage: {
+    mcpName: 'rename_page',
+    inAppKey: 'renamePage',
+    description:
+      'Rename a page (change its title only; the body is untouched, never ' +
+      'resent). Reversible: rename back at any time.',
+    tier: 'deferred',
+    catalogLine: "renamePage — change a page's title only (body untouched).",
+    buildShape: (z) => ({
+      pageId: z.string().min(1).describe('The id of the page to rename.'),
+      title: z.string().min(1).describe('The new title.'),
+    }),
+  },
+
+  deletePage: {
+    mcpName: 'delete_page',
+    inAppKey: 'deletePage',
+    description:
+      'Move a page to the trash — SOFT delete only: the page can be restored ' +
+      'from trash and nothing is ever permanently deleted.',
+    tier: 'deferred',
+    catalogLine: 'deletePage — move a page to trash (soft delete, reversible).',
+    // GUARDRAIL preserved (§14 H4): the schema exposes ONLY pageId, so a
+    // permanentlyDelete/forceDelete flag can never reach the client through this
+    // tool (asserted by ai-chat-tools.service.spec.ts).
+    buildShape: (z) => ({
+      pageId: z.string().min(1).describe('The id of the page to move to trash.'),
+    }),
+  },
+
+  updatePageJson: {
+    mcpName: 'update_page_json',
+    inAppKey: 'updatePageJson',
+    description:
+      "Replace a page's content with a raw ProseMirror JSON document (lossless " +
+      'write: preserves the block ids, callouts, tables and attributes you pass ' +
+      'in). Typical flow: read the page-JSON view -> modify the JSON -> write it back. ' +
+      'Keep existing node ids intact so heading anchors and history stay ' +
+      'stable. Minimal full-doc example: {"type":"doc","content":[{"type":' +
+      '"paragraph","content":[{"type":"text","text":"Hi"}]}]}. `content` may be ' +
+      'a JSON object or a JSON string (both accepted), and is OPTIONAL: omit it ' +
+      'to update only the title (though prefer the rename-page tool for a title-only ' +
+      'change). Supplying neither content nor title is an error. Reversible: ' +
+      'the previous version is kept in page history.',
+    tier: 'deferred',
+    catalogLine:
+      "updatePageJson — overwrite a page's body with a full ProseMirror document.",
+    buildShape: (z) => ({
+      pageId: z.string().min(1).describe('ID of the page to update'),
+      content: z
+        .any()
+        .optional()
+        .describe(
+          'ProseMirror document {"type":"doc","content":[...]} (JSON object or ' +
+            'JSON string). Omit to update only the title.',
+        ),
+      title: z.string().optional().describe('Optional new title'),
+    }),
+  },
+
+  exportPageMarkdown: {
+    mcpName: 'export_page_markdown',
+    inAppKey: 'exportPageMarkdown',
+    // CANONICAL: the MCP copy (a strict superset of the terse in-app wording).
+    description:
+      'Export a page to a single self-contained, lossless Docmost-flavoured ' +
+      'Markdown file (custom extensions): YAML-free meta header, body with ' +
+      'inline comment anchors and diagrams, and a trailing comments-thread ' +
+      'block. Designed for a download -> edit body -> page-Markdown import ' +
+      'round-trip that preserves everything, including comment highlights. ' +
+      'Comment THREADS are preserved in the file but are not re-pushed to the ' +
+      'server on import.',
+    tier: 'deferred',
+    catalogLine:
+      'exportPageMarkdown — export a page to self-contained Markdown (body + comments).',
+    buildShape: (z) => ({
+      pageId: z.string().min(1).describe('The id of the page to export.'),
+    }),
+  },
+
+  // --- comment tools (unified from the per-layer inline definitions, #294) ---
+  //
+  // create_comment and resolve_comment previously carried a "per-transport
+  // divergence" note in BOTH layers; #294 unifies their schema + description
+  // here. Only the four tools that genuinely exist in BOTH layers live in the
+  // registry: create/list/resolve comment and check_new_comments.
+  //
+  // update_comment and delete_comment are intentionally NOT here: they exist
+  // ONLY on the standalone MCP server. The in-app agent deliberately exposes no
+  // hard comment edit/delete tool (comment edits are irreversible / not
+  // version-tracked; see the guardrail tests in ai-chat-tools.service.spec.ts),
+  // so there is nothing to unify — they stay inline in index.ts.
+
+  createComment: {
+    mcpName: 'create_comment',
+    inAppKey: 'createComment',
+    // CANONICAL: the in-app copy (the more-maintained one). It keeps the same
+    // rules as the MCP copy — inline-only, top-level requires a `selection`, no
+    // page-level comments, replies inherit the anchor, suggestedText must be
+    // unique — and adds the "retry with a corrected EXACT selection" and reply-
+    // to-reply-rejected guidance the MCP copy lacked. Execute-side validation
+    // (reject suggestedText on a reply, require a selection) stays per-layer.
+    description:
+      'Add an INLINE comment to a page, or reply to an existing top-level ' +
+      'comment (one level only — the backend rejects replies to replies). ' +
+      'The comment is anchored inline to the given exact `selection` text ' +
+      '(which gets highlighted); page-level comments are NOT supported. A ' +
+      'new top-level comment REQUIRES a `selection`. Replies inherit the ' +
+      "parent's anchor and take no selection. If the call fails with a " +
+      '"selection not found" error, retry with a corrected EXACT selection ' +
+      'copied verbatim from a single paragraph/block. You may also attach a ' +
+      '`suggestedText` proposing a replacement for the `selection` (a human ' +
+      'applies it from the UI); when set, the `selection` must occur exactly ' +
+      'once in the page. Reversible via the comment UI.',
+    tier: 'core',
+    catalogLine:
+      'createComment — add an inline comment (optionally with a suggested edit).',
+    // Reconciled schema: the field set is identical across both layers; the
+    // only constraint drift is `content`, which the MCP copy pinned to
+    // .min(1) while the in-app copy left unbounded — the stricter MCP form is
+    // kept (an empty comment body is never valid).
+    buildShape: (z) => ({
+      pageId: z.string().describe('The id of the page to comment on.'),
+      content: z.string().min(1).describe('The comment body as Markdown.'),
+      selection: z
+        .string()
+        .min(1)
+        .max(250)
+        .optional()
+        .describe(
+          'EXACT contiguous text from a SINGLE paragraph/block to anchor ' +
+            '(highlight) the comment on (<=250 chars, avoid spanning across ' +
+            'formatting boundaries). Required for a new top-level comment; ' +
+            'omit only when replying via parentCommentId.',
+        ),
+      parentCommentId: z
+        .string()
+        .optional()
+        .describe(
+          'Optional id of a TOP-LEVEL comment to reply to (one level ' +
+            'of replies only).',
+        ),
+      suggestedText: z
+        .string()
+        .min(1)
+        .max(2000)
+        .optional()
+        .describe(
+          'Optional proposed replacement (PLAIN TEXT) for the `selection`, ' +
+            'applied by a human via the UI (never auto-applied). REQUIRES a ' +
+            '`selection`; NOT allowed on a reply. When set, the `selection` ' +
+            'must be UNIQUE in the page — expand it with surrounding context ' +
+            '(still <=250 chars) if it occurs more than once, or the call is ' +
+            'refused.',
+        ),
+    }),
+  },
+
+  listComments: {
+    mcpName: 'list_comments',
+    inAppKey: 'listComments',
+    // CANONICAL: the two copies are near-identical; the MCP copy is the
+    // superset (it keeps the "(pagination is handled internally)" note the
+    // in-app copy dropped), so it is used verbatim.
+    description:
+      'List comments on a page in one call (pagination is handled ' +
+      'internally). By DEFAULT only ACTIVE threads are returned; resolved ' +
+      'threads (a resolved top-level comment and all its replies) are hidden ' +
+      'and their count reported as `resolvedThreadsHidden` so you can re-query ' +
+      'with `includeResolved: true` to see everything. Returns ' +
+      '`{ items, resolvedThreadsHidden }`. Content is returned as Markdown.',
+    tier: 'core',
+    catalogLine:
+      'listComments — list all comments on a page (including resolved).',
+    buildShape: (z) => ({
+      pageId: z.string().describe('ID of the page'),
+      includeResolved: z
+        .boolean()
+        .optional()
+        .describe('default only active threads; true — include resolved'),
+    }),
+  },
+
+  resolveComment: {
+    mcpName: 'resolve_comment',
+    inAppKey: 'resolveComment',
+    // CANONICAL: the MCP copy's richer wording, minus its snake_case reference
+    // to `delete_comment` (a sibling tool that does NOT exist in the in-app
+    // layer) — rephrased transport-neutrally per the registry convention.
+    description:
+      'Resolve (close) or reopen a top-level comment thread (reversible — ' +
+      'pass resolved=false to reopen). Only top-level comments can be ' +
+      'resolved; the server rejects resolving a reply. Resolving keeps the ' +
+      'thread and its replies intact (it is not a deletion).',
+    tier: 'core',
+    catalogLine: 'resolveComment — resolve or reopen a comment thread.',
+    // Reconciled schema: `resolved` drifted — the MCP copy made it optional
+    // with .default(true) (resolve is the common case, documented), the in-app
+    // copy made it required. The MCP form is kept (a strict superset: it never
+    // rejects a previously-valid input and adds a sensible default), and
+    // commentId keeps the MCP copy's stricter .min(1).
+    buildShape: (z) => ({
+      commentId: z
+        .string()
+        .min(1)
+        .describe('ID of the top-level comment thread to resolve or reopen'),
+      resolved: z
+        .boolean()
+        .optional()
+        .default(true)
+        .describe(
+          'true (default) marks the thread resolved/closed; false reopens it',
+        ),
+    }),
+  },
+
+  checkNewComments: {
+    mcpName: 'check_new_comments',
+    inAppKey: 'checkNewComments',
+    // CANONICAL: the MCP copy (the more detailed of the two). The MCP layer's
+    // execute-side guard that rejects an unparseable `since` timestamp stays in
+    // its execute body (per-layer logic), not in the shared schema.
+    description:
+      'Check for new comments across pages in a space since a given ' +
+      'timestamp. Optionally scope to a page subtree (folder). Returns only ' +
+      'comments created after the specified time.',
+    tier: 'deferred',
+    catalogLine:
+      'checkNewComments — find comments in a space created after a timestamp.',
+    // Reconciled schema: `since` keeps the MCP copy's stricter .min(1) (the
+    // in-app copy left it unbounded); field descriptions use the MCP copy's
+    // more detailed wording (it carries an example timestamp).
+    buildShape: (z) => ({
+      spaceId: z.string().describe('Space ID to check for new comments'),
+      since: z
+        .string()
+        .min(1)
+        .describe(
+          "ISO 8601 timestamp — only return comments created after this time " +
+            "(e.g. '2026-03-10T00:00:00Z')",
+        ),
+      parentPageId: z
+        .string()
+        .optional()
+        .describe(
+          'Optional root page ID to scope the check to a subtree (folder). ' +
+            'Only pages under this parent will be checked.',
+        ),
+    }),
+  },
+
+  // --- table tools (unified from the per-layer inline definitions, #294) ---
+  //
+  // These tools carried a "NOT shared" note in BOTH layers because of a single
+  // parameter-NAME drift: the MCP layer named the table reference `table` while
+  // the in-app layer named it `tableRef`. #294 reconciles that drift by unifying
+  // on the MCP name `table` — renaming the MCP public parameter would break
+  // external MCP clients, whereas the in-app parameter is model-facing
+  // (prompt-only) and safe to rename. The in-app execute bodies now destructure
+  // `table` instead of `tableRef` (nothing else changes). Descriptions take the
+  // MCP copy's richer wording (it documented `#<index>`, padding, header-row
+  // behavior) plus the in-app copy's "Reversible via page history" note; sibling
+  // tool references are phrased transport-neutrally.
+  //
+  // NOT here (kept inline in index.ts): table_get / getTable. Its MCP tool name
+  // is noun-first (`table_get`) while the in-app key is verb-first (`getTable`),
+  // so it 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 (its in-app param was still
+  // aligned to `table` for consistency with the migrated trio below).
+
+  tableInsertRow: {
+    mcpName: 'table_insert_row',
+    inAppKey: 'tableInsertRow',
+    description:
+      'Insert a row of plain-text cells into a table. `table` is `#<index>` ' +
+      'from the page outline, or a block id inside it. `cells` is the text per ' +
+      "column (padded to the table's column count; an error if more cells than " +
+      'columns). `index` is the 0-based insert position (0 inserts before the ' +
+      'header); omit to append at the end. Reversible via page history.',
+    tier: 'deferred',
+    catalogLine: 'tableInsertRow — insert a row of plain-text cells into a table.',
+    buildShape: (z) => ({
+      pageId: z.string().min(1).describe('The id of the page.'),
+      table: z
+        .string()
+        .min(1)
+        .describe('"#<index>" from the page outline, or a block id in the table.'),
+      cells: z.array(z.string()).describe('The cell texts for the row (one per column).'),
+      index: z
+        .number()
+        .int()
+        .optional()
+        .describe('0-based insert position (0 inserts before the header); omit to append.'),
+    }),
+  },
+
+  tableDeleteRow: {
+    mcpName: 'table_delete_row',
+    inAppKey: 'tableDeleteRow',
+    description:
+      'Delete the row at 0-based `index` from a table (`table` is `#<index>` ' +
+      'from the page outline, or a block id inside it). Refuses to delete the ' +
+      "table's only row; an out-of-range `index` throws. Deleting `index` 0 " +
+      'removes the header row, and the next row becomes the new header. ' +
+      'Reversible via page history.',
+    tier: 'deferred',
+    catalogLine: 'tableDeleteRow — delete a table row at a 0-based index.',
+    buildShape: (z) => ({
+      pageId: z.string().min(1).describe('The id of the page.'),
+      table: z
+        .string()
+        .min(1)
+        .describe('"#<index>" from the page outline, or a block id in the table.'),
+      index: z.number().int().describe('0-based row index to delete.'),
+    }),
+  },
+
+  tableUpdateCell: {
+    mcpName: 'table_update_cell',
+    inAppKey: 'tableUpdateCell',
+    description:
+      'Set the plain-text content of cell [row, col] (0-based) in a table ' +
+      '(`table` is `#<index>` from the page outline, or a block id inside it). ' +
+      "Replaces the cell's content with a single text paragraph; for rich " +
+      "formatting, patch the cell's paragraph id (obtained from reading the " +
+      'table) instead. Reversible via page history.',
+    tier: 'deferred',
+    catalogLine: 'tableUpdateCell — set the text of a table cell at [row, col].',
+    buildShape: (z) => ({
+      pageId: z.string().min(1).describe('The id of the page.'),
+      table: z
+        .string()
+        .min(1)
+        .describe('"#<index>" from the page outline, or a block id in the table.'),
+      row: z.number().int().describe('0-based row index.'),
+      col: z.number().int().describe('0-based column index.'),
+      text: z.string().describe('The new cell text.'),
+    }),
+  },
 } satisfies Record<string, SharedToolSpec>;

packages/prosemirror-markdown/test/gitmost-transcript-neutralization.test.ts

@@ -0,0 +1,124 @@
+import { describe, expect, it } from "vitest";
+// Import DIRECTLY from src (NOT the docmost-client barrel, which pulls in
+// collaboration.ts and mutates global DOM at import time).
+import { convertProseMirrorToMarkdown } from "../src/lib/markdown-converter.js";
+import { markdownToProseMirror } from "../src/lib/markdown-to-prosemirror.js";
+
+/**
+ * gitmost #377 (round-1 review, finding #1) — proof, against the REAL
+ * converter, that the transcript-insert boundary defense survives git-sync.
+ *
+ * The web bridge (apps/client .../gitmost/gitmost-recording.ts,
+ * `gitmostInsertTranscriptIntoEditor`) appends each transcript line as a
+ * PARAGRAPH text node. The paragraph serializer here (`case "paragraph"`) emits
+ * that text VERBATIM with no block-escape, so a line whose text begins with a
+ * col-0 markdown block trigger would, on the doc -> markdown -> doc git-sync
+ * cycle, silently re-parse into a heading / list / quote / callout / code block.
+ * That missing block-escape is the pre-existing root cause; the bridge's
+ * boundary defense prepends an invisible zero-width space (U+200B) to a line
+ * that begins with such a trigger, shifting it off column 0.
+ *
+ * This test keeps a COPY of the bridge's trigger regex (the bridge is in a
+ * different package and can't be imported here) and asserts:
+ *   1. bare trigger lines DO corrupt (documents the root cause), and
+ *   2. the ZWSP-neutralized form round-trips as a single PARAGRAPH with the
+ *      text byte-preserved.
+ */
+
+const ZWSP = "​"; // U+200B
+
+// MUST stay in sync with GITMOST_MD_BLOCK_TRIGGER_RE in the client bridge.
+const MD_BLOCK_TRIGGER_RE =
+  /^(?:#{1,6}(?:\s|$)|[-*+](?:\s|$)|>|\d+[.)](?:\s|$)|```|~~~|\||([-*_])(?:\s*\1){2,}\s*$)/;
+
+const doc = (...nodes: any[]) => ({ type: "doc", content: nodes });
+const para = (t: string) => ({
+  type: "paragraph",
+  content: [{ type: "text", text: t }],
+});
+
+const roundtrip = async (text: string) => {
+  const md = convertProseMirrorToMarkdown(doc(para(text)));
+  const back = await markdownToProseMirror(md);
+  return back.content as any[];
+};
+
+describe("gitmost transcript neutralization (git-sync round-trip)", () => {
+  // Lines that, at column 0, the serializer's missing block-escape would let
+  // git-sync re-parse into a non-paragraph block.
+  const triggerLines = [
+    "- dash",
+    "* star",
+    "+ plus",
+    "> quote",
+    "# hash",
+    "1. one",
+    "1) one",
+    "> [!info] note",
+    "```js",
+    "~~~",
+    // Solid + spaced thematic breaks — these re-parse into a `horizontalRule`,
+    // which carries NO text, so a bare separator line LOSES its text entirely
+    // (round-2 finding). `_` also only forms a block via this construct.
+    "---",
+    "***",
+    "___",
+    "- - -", // spaced dash break (solid form is caught by [-*+]\s too, but this is the break)
+    "_ _ _",
+  ];
+
+  it("BARE trigger lines corrupt into non-paragraph blocks (root cause)", async () => {
+    for (const line of triggerLines) {
+      const blocks = await roundtrip(line);
+      // At least one produced block is NOT a paragraph — i.e. corruption.
+      const allParagraphs = blocks.every((b) => b.type === "paragraph");
+      expect(
+        allParagraphs,
+        `expected "${line}" to corrupt when inserted bare`,
+      ).toBe(false);
+    }
+  });
+
+  it("BARE solid thematic breaks corrupt into a text-LOSING horizontalRule", async () => {
+    // The severe case: no text node survives. Documents why neutralization
+    // matters more here than for list/quote (where the text survived).
+    for (const line of ["---", "***", "___"]) {
+      const blocks = await roundtrip(line);
+      expect(blocks.map((b) => b.type)).toContain("horizontalRule");
+      // No block carries the original text anywhere.
+      const flat = JSON.stringify(blocks);
+      expect(flat).not.toContain(line);
+    }
+  });
+
+  it("ZWSP-neutralized trigger lines round-trip as a single paragraph, text preserved", async () => {
+    for (const line of triggerLines) {
+      // The regex must actually classify each as a trigger.
+      expect(MD_BLOCK_TRIGGER_RE.test(line), `regex missed "${line}"`).toBe(
+        true,
+      );
+      const neutralized = ZWSP + line;
+      const blocks = await roundtrip(neutralized);
+
+      expect(blocks).toHaveLength(1);
+      expect(blocks[0].type).toBe("paragraph");
+      // Text is byte-preserved (ZWSP + original line), so the display is the
+      // original line with only an invisible leading character.
+      expect(blocks[0].content[0].text).toBe(neutralized);
+    }
+  });
+
+  it("normal host-prefixed lines never match the trigger regex and round-trip byte-exact", async () => {
+    for (const line of [
+      "You: hello there",
+      "Speaker 1: - and then a dash mid-line",
+      "Speaker 2: 1. not a list",
+    ]) {
+      expect(MD_BLOCK_TRIGGER_RE.test(line)).toBe(false);
+      const blocks = await roundtrip(line);
+      expect(blocks).toHaveLength(1);
+      expect(blocks[0].type).toBe("paragraph");
+      expect(blocks[0].content[0].text).toBe(line);
+    }
+  });
+});

patches/ai@6.0.134.patch

@@ -0,0 +1,68 @@
+diff --git a/dist/index.js b/dist/index.js
+index ae447a12f7823ec0a00837ee9f0eb809a610d5f8..a3402b2c2d021ef432cfa76e35d370073d525135 100644
+--- a/dist/index.js
++++ b/dist/index.js
+@@ -6578,9 +6578,19 @@ function createOutputTransformStream(output) {
+         controller.enqueue({ part: chunk, partialOutput: void 0 });
+         return;
+       }
+-      text2 += chunk.text;
+       textChunk += chunk.text;
+       textProviderMetadata = (_a21 = chunk.providerMetadata) != null ? _a21 : textProviderMetadata;
++      if (output == null) {
++        // PATCH(docmost #OOM): no output strategy requested -> publish each
++        // text-delta immediately and do NOT build cumulative partialOutput
++        // snapshots. Unpatched, the default text() output snapshots the ENTIRE
++        // accumulated turn text on every delta (O(n^2) memory) and those
++        // snapshots pile up in the never-consumed leftover tee branch of
++        // DefaultStreamTextResult.baseStream -> heap OOM on long agent turns.
++        publishTextChunk({ controller });
++        return;
++      }
++      text2 += chunk.text;
+       const result = await output.parsePartialOutput({ text: text2 });
+       if (result !== void 0) {
+         const currentJson = JSON.stringify(result.partial);
+@@ -6959,7 +6969,7 @@ var DefaultStreamTextResult = class {
+         })
+       );
+     }
+-    this.baseStream = stream.pipeThrough(createOutputTransformStream(output != null ? output : text())).pipeThrough(eventProcessor);
++    this.baseStream = stream.pipeThrough(createOutputTransformStream(output)).pipeThrough(eventProcessor);
+     const { maxRetries, retry } = prepareRetries({
+       maxRetries: maxRetriesArg,
+       abortSignal
+diff --git a/dist/index.mjs b/dist/index.mjs
+index 663875332e3f9a9bd167c25583c515876f42951b..b840b0502c9894df983e0154805abb80e70e6331 100644
+--- a/dist/index.mjs
++++ b/dist/index.mjs
+@@ -6501,9 +6501,19 @@ function createOutputTransformStream(output) {
+         controller.enqueue({ part: chunk, partialOutput: void 0 });
+         return;
+       }
+-      text2 += chunk.text;
+       textChunk += chunk.text;
+       textProviderMetadata = (_a21 = chunk.providerMetadata) != null ? _a21 : textProviderMetadata;
++      if (output == null) {
++        // PATCH(docmost #OOM): no output strategy requested -> publish each
++        // text-delta immediately and do NOT build cumulative partialOutput
++        // snapshots. Unpatched, the default text() output snapshots the ENTIRE
++        // accumulated turn text on every delta (O(n^2) memory) and those
++        // snapshots pile up in the never-consumed leftover tee branch of
++        // DefaultStreamTextResult.baseStream -> heap OOM on long agent turns.
++        publishTextChunk({ controller });
++        return;
++      }
++      text2 += chunk.text;
+       const result = await output.parsePartialOutput({ text: text2 });
+       if (result !== void 0) {
+         const currentJson = JSON.stringify(result.partial);
+@@ -6882,7 +6892,7 @@ var DefaultStreamTextResult = class {
+         })
+       );
+     }
+-    this.baseStream = stream.pipeThrough(createOutputTransformStream(output != null ? output : text())).pipeThrough(eventProcessor);
++    this.baseStream = stream.pipeThrough(createOutputTransformStream(output)).pipeThrough(eventProcessor);
+     const { maxRetries, retry } = prepareRetries({
+       maxRetries: maxRetriesArg,
+       abortSignal

pnpm-lock.yaml

@@ -44,6 +44,9 @@ overrides:
   ip-address: 10.1.1
 
 patchedDependencies:
+  ai@6.0.134:
+    hash: f60bfc3357e01e1f3978c6c40fdd65aeb33fefaad7179cde8676465b6c5ff4d9
+    path: patches/ai@6.0.134.patch
   scimmy@1.3.5:
     hash: 775d80f86830b2c5dd1a250c9802c10f8fc3da3c7898373de5aa0c23993d1673
     path: patches/scimmy@1.3.5.patch
@@ -504,6 +507,9 @@ importers:
       vite:
         specifier: 8.0.5
         version: 8.0.5(@types/node@22.19.1)(esbuild@0.28.0)(jiti@2.4.2)(less@4.2.0)(sugarss@5.0.1(postcss@8.5.14))(terser@5.39.0)(tsx@4.21.0)(yaml@2.8.3)
+      vite-plugin-compression2:
+        specifier: 2.5.3
+        version: 2.5.3
       vitest:
         specifier: 4.1.6
         version: 4.1.6(@opentelemetry/api@1.9.0)(@types/node@22.19.1)(@vitest/coverage-v8@4.1.6)(happy-dom@20.8.9)(jsdom@25.0.0)(vite@8.0.5(@types/node@22.19.1)(esbuild@0.28.0)(jiti@2.4.2)(less@4.2.0)(sugarss@5.0.1(postcss@8.5.14))(terser@5.39.0)(tsx@4.21.0)(yaml@2.8.3))
@@ -543,6 +549,12 @@ importers:
       '@docmost/pdf-inspector':
         specifier: 1.9.6
         version: 1.9.6
+      '@docmost/prosemirror-markdown':
+        specifier: workspace:*
+        version: link:../../packages/prosemirror-markdown
+      '@fastify/compress':
+        specifier: ^9.0.0
+        version: 9.0.0
       '@fastify/cookie':
         specifier: ^11.0.2
         version: 11.0.2
@@ -623,10 +635,10 @@ importers:
         version: 8.3.0(socket.io-adapter@2.5.4)
       ai:
         specifier: ^6.0.134
-        version: 6.0.134(zod@4.3.6)
+        version: 6.0.134(patch_hash=f60bfc3357e01e1f3978c6c40fdd65aeb33fefaad7179cde8676465b6c5ff4d9)(zod@4.3.6)
       ai-sdk-ollama:
         specifier: ^3.8.1
-        version: 3.8.1(ai@6.0.134(zod@4.3.6))(zod@4.3.6)
+        version: 3.8.1(ai@6.0.134(patch_hash=f60bfc3357e01e1f3978c6c40fdd65aeb33fefaad7179cde8676465b6c5ff4d9)(zod@4.3.6))(zod@4.3.6)
       bcrypt:
         specifier: ^6.0.0
         version: 6.0.0
@@ -2700,6 +2712,9 @@ packages:
   '@fastify/busboy@3.1.1':
     resolution: {integrity: sha512-5DGmA8FTdB2XbDeEwc/5ZXBl6UbBAyBOOLlPuBnZ/N1SwdH9Ii+cOX3tBROlDgcTXxjOYnLMVoKk9+FXAw0CJw==}
 
+  '@fastify/compress@9.0.0':
+    resolution: {integrity: sha512-PZRg+ut5xd/ubsGPWfoPNryoCOtEdHboIWpDieTUHov1gKdLitF8mRmT3JbqNnRbelQXSNXUsIpakAEKR6AcTQ==}
+
   '@fastify/cookie@11.0.2':
     resolution: {integrity: sha512-GWdwdGlgJxyvNv+QcKiGNevSspMQXncjMZ1J8IvuDQk0jvkzgWWZFNC2En3s+nHndZBGV8IbLwOI/sxCZw/mzA==}
 
@@ -4493,6 +4508,15 @@ packages:
   '@rolldown/pluginutils@1.0.0-rc.7':
     resolution: {integrity: sha512-qujRfC8sFVInYSPPMLQByRh7zhwkGFS4+tyMQ83srV1qrxL4g8E2tyxVVyxd0+8QeBM1mIk9KbWxkegRr76XzA==}
 
+  '@rollup/pluginutils@5.4.0':
+    resolution: {integrity: sha512-MfPp06CjRLfXQ3wY0R8vJDYBy/MvVcc9OulEfR0B8Iv9ko+GCNaRZ+EpJYFl27LhKsZK0o420sYCRHCjfCgeUg==}
+    engines: {node: '>=14.0.0'}
+    peerDependencies:
+      rollup: ^1.20.0||^2.0.0||^3.0.0||^4.0.0
+    peerDependenciesMeta:
+      rollup:
+        optional: true
+
   '@selderee/plugin-htmlparser2@0.11.0':
     resolution: {integrity: sha512-P33hHGdldxGabLFjPPpaTxVolMrzrcegejx+0GxjrIb9Zv48D8yAIA/QTDR2dFl7Uz7urX8aX6+5bCZslr+gWQ==}
 
@@ -5682,6 +5706,10 @@ packages:
     resolution: {integrity: sha512-/XrFJgzQQQHpti1raDJC6m4ws6aNktmjBlhk8Fdlk7LwCEuDoieEJJY9OFHjfiFJFFRM2tK+Ky/IsfbbmlMu1w==}
     engines: {node: ^22.22.2 || ^24.15.0 || >=26.0.0}
 
+  abort-controller@3.0.0:
+    resolution: {integrity: sha512-h8lQ8tacZYnR3vNQTgibj+tODHI5/+l06Au2Pcriv/Gmet0eaj4TwWH41sO9wnHDiQsEj19q0drzdWdeAHtweg==}
+    engines: {node: '>=6.5'}
+
   abstract-logging@2.0.1:
     resolution: {integrity: sha512-2BjRTZxTPvheOvGbBslFSYOUkr+SjPtOnrLP33f+VIWLzezQpZcqVg7ja3L4dBXmzzgwT+a029jRx5PCi3JuiA==}
 
@@ -6064,6 +6092,9 @@ packages:
   buffer@5.7.1:
     resolution: {integrity: sha512-EHcyIPBQ4BSGlvjB16k5KgAJ27CIsHY/2JBmCRReo48y9rQ3MaUzWX3KVlBa4U7MyX02HdVj0K7C3WaB3ju7FQ==}
 
+  buffer@6.0.3:
+    resolution: {integrity: sha512-FTiCpNxtwiZZHEZbcbTIcZjERVICn9yq/pDFkTl95/AxzD1naBctN7YO68riM/gLSDY7sdrMby8hofADYuuqOA==}
+
   bullmq@5.76.10:
     resolution: {integrity: sha512-LWve7SpQjYSpCP2GEsWmoyzTz2H37L8HRmSTu3YihYsTOr5kJxrfEX6aEV7m6eskEMWXSHZYTMZepX6qNaH6CQ==}
     engines: {node: '>=12.22.0'}
@@ -6819,6 +6850,9 @@ packages:
     resolution: {integrity: sha512-KIN/nDJBQRcXw0MLVhZE9iQHmG68qAVIBg9CqmUYjmQIhgij9U5MFvrqkUL5FbtyyzZuOeOt0zdeRe4UY7ct+A==}
     engines: {node: '>= 0.4'}
 
+  duplexify@3.7.1:
+    resolution: {integrity: sha512-07z8uv2wMyS51kKhD1KsdXJg5WQ6t93RneqRxUHnskXVtlYYkLqM0gqStQZ3pj073g687jPCHrqNfCzawLYh5g==}
+
   ecdsa-sig-formatter@1.0.11:
     resolution: {integrity: sha512-nagl3RYrbNv6kQkeJIpt6NJZy8twLB/2vtz6yN9Z4vRKHN4/QZJIEbqohALSgwKdnksuY3k5Addp5lg8sVoVcQ==}
 
@@ -7062,6 +7096,9 @@ packages:
     resolution: {integrity: sha512-MMdARuVEQziNTeJD8DgMqmhwR11BRQ/cBP+pLtYdSTnf3MIO8fFeiINEbX36ZdNlfU/7A9f3gUw49B3oQsvwBA==}
     engines: {node: '>=4.0'}
 
+  estree-walker@2.0.2:
+    resolution: {integrity: sha512-Rfkk/Mp/DL7JVje3u18FxFujQlTNR2q6QfMSMB7AvCBx91NGj/ba3kCfza0f6dVDbw7YlRf/nDrn7pQrCCyQ/w==}
+
   estree-walker@3.0.3:
     resolution: {integrity: sha512-7RUKfXgSMMkzt6ZuXmqapOurLGPPfgj6l9uRZ7lRGolvk0y2yocc35LdcxKC5PQZdn2DMqioAQ2NoWcrTKmm6g==}
 
@@ -7073,6 +7110,10 @@ packages:
     resolution: {integrity: sha512-aIL5Fx7mawVa300al2BnEE4iNvo1qETxLrPI/o05L7z6go7fCw1J6EQmbK4FmJ2AS7kgVF/KEZWufBfdClMcPg==}
     engines: {node: '>= 0.6'}
 
+  event-target-shim@5.0.1:
+    resolution: {integrity: sha512-i/2XbnSz/uxRCU6+NdVJgKWDTM427+MqYbkQzD321DuCQJUqOuJKIA0IM2+W2xtYHdKOmZ4dR6fExsd4SXL+WQ==}
+    engines: {node: '>=6'}
+
   eventemitter2@6.4.9:
     resolution: {integrity: sha512-JEPTiaOt9f04oa6NOkc4aH+nVp5I3wEjpHbIPqfgCdD5v5bUzy7xQqwcVO2aDQgOWhI28da57HksMrzK9HlRxg==}
 
@@ -9082,6 +9123,9 @@ packages:
   peberminta@0.9.0:
     resolution: {integrity: sha512-XIxfHpEuSJbITd1H3EeQwpcZbTLHc+VVr8ANI9t5sit565tsI4/xK3KWTUFE2e6QiangUkh3B0jihzmGnNrRsQ==}
 
+  peek-stream@1.1.3:
+    resolution: {integrity: sha512-FhJ+YbOSBb9/rIl2ZeE/QHEsWn7PqNYt8ARAY3kIgNGOk13g9FGyIY6JIl/xB/3TFRVoTv5as0l11weORrTekA==}
+
   pend@1.2.0:
     resolution: {integrity: sha512-F3asv42UuXchdzt+xXqfW1OGlVBe+mxa2mqI0pg5yAHZPvFmY3Y6drSf/GQ1A86WgWEN9Kzh/WrgKa6iGcHXLg==}
 
@@ -9327,6 +9371,10 @@ packages:
   process-warning@5.0.0:
     resolution: {integrity: sha512-a39t9ApHNx2L4+HBnQKqxxHNs1r7KF+Intd8Q/g1bUh6q0WIp9voPXJ/x0j+ZL45KF1pJd9+q2jLIRMfvEshkA==}
 
+  process@0.11.10:
+    resolution: {integrity: sha512-cdGef/drWFoydD1JsMzuFf8100nZl+GT+yacc2bEced5f9Rjk4z+WtFUTBu9PhOi9j/jfmBPu0mMEY4wIdAF8A==}
+    engines: {node: '>= 0.6.0'}
+
   prom-client@15.1.3:
     resolution: {integrity: sha512-6ZiOBfCywsD4k1BN9IX0uZhF+tJkV8q8llP64G5Hajs4JOeVLPCwpPVcpXy3BwYiUGgyJzsJJQeOIv7+hDSq8g==}
     engines: {node: ^16 || ^18 || >=20}
@@ -9622,6 +9670,10 @@ packages:
     resolution: {integrity: sha512-9u/sniCrY3D5WdsERHzHE4G2YCXqoG5FTHUiCC4SIbr6XcLZBY05ya9EKjYek9O5xOAwjGq+1JdGBAS7Q9ScoA==}
     engines: {node: '>= 6'}
 
+  readable-stream@4.7.0:
+    resolution: {integrity: sha512-oIGGmcpTLwPga8Bn6/Z75SVaH1z5dUut2ibSyAMVhmUggWpmDn2dapB0n7f8nwaSiRtepAsfJyfXIO5DCVAODg==}
+    engines: {node: ^12.22.0 || ^14.17.0 || >=16.0.0}
+
   readdirp@3.6.0:
     resolution: {integrity: sha512-hOS089on8RduqdbhvQ5Z37A0ESjsqz6qnRcffsMU3495FuTdqSm+7bhJ29JvIOsBDEEnan5DPu9t3To9VRlMzA==}
     engines: {node: '>=8.10.0'}
@@ -10010,6 +10062,9 @@ packages:
   stream-browserify@3.0.0:
     resolution: {integrity: sha512-H73RAHsVBapbim0tU2JwwOiXUj+fikfiaoYAKHF3VJfA0pe2BCzkhAHBlLG6REzE+2WNZcxOXjK7lkso+9euLA==}
 
+  stream-shift@1.0.3:
+    resolution: {integrity: sha512-76ORR0DO1o1hlKwTbi/DM3EXWGf3ZJYO8cXX5RJwnul2DEg2oyoZyjLNoQM8WsvZiFKCRfC1O0J7iCvie3RZmQ==}
+
   strict-event-emitter-types@2.0.0:
     resolution: {integrity: sha512-Nk/brWYpD85WlOgzw5h173aci0Teyv8YdIAEtV+N88nDB0dLlazZyJMIsN6eo1/AR61l+p6CJTG1JIyFaoNEEA==}
 
@@ -10150,6 +10205,9 @@ packages:
     resolution: {integrity: sha512-g9ljZiwki/LfxmQADO3dEY1CbpmXT5Hm2fJ+QaGKwSXUylMybePR7/67YW7jOrrvjEgL1Fmz5kzyAjWVWLlucg==}
     engines: {node: '>=6'}
 
+  tar-mini@0.2.0:
+    resolution: {integrity: sha512-+qfUHz700DWnRutdUsxRRVZ38G1Qr27OetwaMYTdg8hcPxf46U0S1Zf76dQMWRBmusOt2ZCK5kbIaiLkoGO7WQ==}
+
   tar-stream@2.2.0:
     resolution: {integrity: sha512-ujeqbceABgwMZxEJnk2HDY2DlnUZ+9oEcb1KzTVfYHio0UE6dG71n60d8D2I4qNvleWrrXpmjpt7vZeF1LnMZQ==}
     engines: {node: '>=6'}
@@ -10193,6 +10251,9 @@ packages:
     resolution: {integrity: sha512-nt6AMGKW1p/70DF/hGBdJB57B8Tspmbp5gfJ8ilhLnt7kkr2ye7hzD6NVG8GGErk2HWF34igrL2CXmNIkzKqKw==}
     engines: {node: '>=18'}
 
+  through2@2.0.5:
+    resolution: {integrity: sha512-/mrRod8xqpA+IHSLyGCQ2s8SPHiCDEeQJSep1jqLYeEUClOFG2Qsh+4FU6G9VeqpZnGW/Su8LQGc4YKni5rYSQ==}
+
   tiny-invariant@1.3.3:
     resolution: {integrity: sha512-+FbBPE1o9QAYvviau/qC5SE3caw21q3xkvWKBtja5vgqOWIHHJ3ioaq1VPfn/Szqctz2bU/oYeKd9/z5BL+PVg==}
 
@@ -10598,6 +10659,9 @@ packages:
     resolution: {integrity: sha512-BNGbWLfd0eUPabhkXUVm0j8uuvREyTh5ovRa/dyow/BqAbZJyC+5fU+IzQOzmAKzYqYRAISoRhdQr3eIZ/PXqg==}
     engines: {node: '>= 0.8'}
 
+  vite-plugin-compression2@2.5.3:
+    resolution: {integrity: sha512-ItPgqQWkcnBbVw7is9OKwiZ8v6+ju9rYROl5Lp6QfQDEx/d55AwJQb/KLpsQqsU9HoigYBsZ8tK6I02UwJNvEw==}
+
   vite@8.0.5:
     resolution: {integrity: sha512-nmu43Qvq9UopTRfMx2jOYW5l16pb3iDC1JH6yMuPkpVbzK0k+L7dfsEDH4jRgYFmsg0sTAqkojoZgzLMlwHsCQ==}
     engines: {node: ^20.19.0 || >=22.12.0}
@@ -12937,6 +13001,15 @@ snapshots:
 
   '@fastify/busboy@3.1.1': {}
 
+  '@fastify/compress@9.0.0':
+    dependencies:
+      '@fastify/accept-negotiator': 2.0.1
+      fastify-plugin: 5.1.0
+      mime-db: 1.54.0
+      minipass: 7.1.3
+      peek-stream: 1.1.3
+      readable-stream: 4.7.0
+
   '@fastify/cookie@11.0.2':
     dependencies:
       cookie: 1.1.1
@@ -14952,6 +15025,12 @@ snapshots:
 
   '@rolldown/pluginutils@1.0.0-rc.7': {}
 
+  '@rollup/pluginutils@5.4.0':
+    dependencies:
+      '@types/estree': 1.0.9
+      estree-walker: 2.0.2
+      picomatch: 4.0.4
+
   '@selderee/plugin-htmlparser2@0.11.0':
     dependencies:
       domhandler: 5.0.3
@@ -16325,6 +16404,10 @@ snapshots:
 
   abbrev@5.0.0: {}
 
+  abort-controller@3.0.0:
+    dependencies:
+      event-target-shim: 5.0.1
+
   abstract-logging@2.0.1: {}
 
   accepts@1.3.8:
@@ -16355,17 +16438,17 @@ snapshots:
 
   agent-base@7.1.4: {}
 
-  ai-sdk-ollama@3.8.1(ai@6.0.134(zod@4.3.6))(zod@4.3.6):
+  ai-sdk-ollama@3.8.1(ai@6.0.134(patch_hash=f60bfc3357e01e1f3978c6c40fdd65aeb33fefaad7179cde8676465b6c5ff4d9)(zod@4.3.6))(zod@4.3.6):
     dependencies:
       '@ai-sdk/provider': 3.0.8
       '@ai-sdk/provider-utils': 4.0.21(zod@4.3.6)
-      ai: 6.0.134(zod@4.3.6)
+      ai: 6.0.134(patch_hash=f60bfc3357e01e1f3978c6c40fdd65aeb33fefaad7179cde8676465b6c5ff4d9)(zod@4.3.6)
       jsonrepair: 3.13.3
       ollama: 0.6.3
     transitivePeerDependencies:
       - zod
 
-  ai@6.0.134(zod@4.3.6):
+  ai@6.0.134(patch_hash=f60bfc3357e01e1f3978c6c40fdd65aeb33fefaad7179cde8676465b6c5ff4d9)(zod@4.3.6):
     dependencies:
       '@ai-sdk/gateway': 3.0.77(zod@4.3.6)
       '@ai-sdk/provider': 3.0.8
@@ -16773,6 +16856,11 @@ snapshots:
       base64-js: 1.5.1
       ieee754: 1.2.1
 
+  buffer@6.0.3:
+    dependencies:
+      base64-js: 1.5.1
+      ieee754: 1.2.1
+
   bullmq@5.76.10:
     dependencies:
       cron-parser: 4.9.0
@@ -17526,6 +17614,13 @@ snapshots:
       es-errors: 1.3.0
       gopd: 1.2.0
 
+  duplexify@3.7.1:
+    dependencies:
+      end-of-stream: 1.4.4
+      inherits: 2.0.4
+      readable-stream: 2.3.8
+      stream-shift: 1.0.3
+
   ecdsa-sig-formatter@1.0.11:
     dependencies:
       safe-buffer: 5.2.1
@@ -17952,6 +18047,8 @@ snapshots:
 
   estraverse@5.3.0: {}
 
+  estree-walker@2.0.2: {}
+
   estree-walker@3.0.3:
     dependencies:
       '@types/estree': 1.0.9
@@ -17960,6 +18057,8 @@ snapshots:
 
   etag@1.8.1: {}
 
+  event-target-shim@5.0.1: {}
+
   eventemitter2@6.4.9: {}
 
   eventemitter3@4.0.7: {}
@@ -20223,6 +20322,12 @@ snapshots:
 
   peberminta@0.9.0: {}
 
+  peek-stream@1.1.3:
+    dependencies:
+      buffer-from: 1.1.2
+      duplexify: 3.7.1
+      through2: 2.0.5
+
   pend@1.2.0: {}
 
   perfect-freehand@1.2.0: {}
@@ -20494,6 +20599,8 @@ snapshots:
 
   process-warning@5.0.0: {}
 
+  process@0.11.10: {}
+
   prom-client@15.1.3:
     dependencies:
       '@opentelemetry/api': 1.9.0
@@ -20915,6 +21022,14 @@ snapshots:
       string_decoder: 1.3.0
       util-deprecate: 1.0.2
 
+  readable-stream@4.7.0:
+    dependencies:
+      abort-controller: 3.0.0
+      buffer: 6.0.3
+      events: 3.3.0
+      process: 0.11.10
+      string_decoder: 1.3.0
+
   readdirp@3.6.0:
     dependencies:
       picomatch: 2.3.2
@@ -21368,6 +21483,8 @@ snapshots:
       inherits: 2.0.4
       readable-stream: 3.6.2
 
+  stream-shift@1.0.3: {}
+
   strict-event-emitter-types@2.0.0: {}
 
   string-length@4.0.2:
@@ -21528,6 +21645,8 @@ snapshots:
 
   tapable@2.3.0: {}
 
+  tar-mini@0.2.0: {}
+
   tar-stream@2.2.0:
     dependencies:
       bl: 4.1.0
@@ -21577,6 +21696,11 @@ snapshots:
 
   throttleit@2.1.0: {}
 
+  through2@2.0.5:
+    dependencies:
+      readable-stream: 2.3.8
+      xtend: 4.0.2
+
   tiny-invariant@1.3.3: {}
 
   tinybench@2.9.0: {}
@@ -21981,6 +22105,13 @@ snapshots:
 
   vary@1.1.2: {}
 
+  vite-plugin-compression2@2.5.3:
+    dependencies:
+      '@rollup/pluginutils': 5.4.0
+      tar-mini: 0.2.0
+    transitivePeerDependencies:
+      - rollup
+
   vite@8.0.5(@types/node@20.19.43)(esbuild@0.28.0)(jiti@2.4.2)(less@4.2.0)(sugarss@5.0.1(postcss@8.5.14))(terser@5.39.0)(tsx@4.21.0)(yaml@2.8.3):
     dependencies:
       lightningcss: 1.32.0
@@ -22361,8 +22492,7 @@ snapshots:
 
   xpath@0.0.34: {}
 
-  xtend@4.0.2:
+  xtend@4.0.2: {}
-    optional: true
 
   y-indexeddb@9.0.12(yjs@13.6.30(patch_hash=1ceeb66dba1f86545c98a3ff7f5152aff9b35caf409091cef9caedb5e65c8810)):
     dependencies: