The #285 gate dropped every remapped (wrong-layout) candidate shorter than 3
chars, which broke the legitimate short prefix '/сщ' -> 'co' -> Code while '/co'
still worked. Replace the blanket length filter with a match-TYPE gate: the
original query and remaps >= 3 chars match fully (title/description/searchTerms);
a short (1-2 char) remap is restricted to a TITLE fuzzy-match. So '/сщ' -> 'co'
matches the 'Code' title again, while '/cy' -> 'сн' and '/b' -> 'и' still do not
surface Footnote (they only ever leaked in via the 'сноска'/'примечание'
searchTerm substrings, not the title).
Adds positive tests for /сщ and /co; keeps the /cy and /b negatives.
closes#283
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
F4: menu-items.layout.test.ts imports from './menu-items' (relative, no extension),
matching the sibling test files (was still the aliased '@/.../menu-items.ts').
F5: remove the dead 'candidate !== originalCandidate' clause from the remapped-candidate
filter — buildLayoutCandidates dedupes remaps against the original via Set, so the tail
after destructuring can never equal the original; the length gate is the only real
condition. Comment updated to state the dedup invariant instead.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This actually lands F1+F2 (round 1 pushed only the test rename by mistake).
F1: only the ORIGINAL query matches without length limits; remapped (wrong-layout)
candidates must be >= 3 chars before they can match, via a shared candidateMatchesItem
helper applied to both the item filter and the tie-break sort. Stops a 1-2 char ASCII
query from spuriously substring-matching Cyrillic searchTerms (/cy->сн no longer hits
'сноска', /b->и no longer hits 'примечание'), while keeping real wrong-layout commands
(/сщву->Code, /cyjcrf->Footnote), genuine short queries (/p, /h1) and Cyrillic terms
(/сноска->Footnote) working.
F2: reword the buildLayoutCandidates JSDoc (an ASCII query yields multiple candidates;
dedup only collapses when nothing is remappable).
Adds negative tests for /cy and /b.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
F1: only the ORIGINAL query does full matching; remapped (wrong-layout) candidates
must be >= 3 chars and differ from the original before they can match (via a shared
candidateMatchesItem helper, applied to both the filter and the tie-break sort). This
stops a short remapped candidate from substring-matching the only cyrillic searchTerms
(/cy->сн, /b->и no longer surface Footnote) while keeping real wrong-layout commands
(/сщву->Code, /cyjcrf->Footnote) and genuine cyrillic terms (/сноска->Footnote) working.
F2: fix the buildLayoutCandidates JSDoc (an ascii query yields multiple candidates,
not a single-element set).
F3: rename the test to menu-items.layout.test.ts + relative import, per sibling convention.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Typing a command with the wrong layout (e.g. Russian ЙЦУКЕН -> /сщву for 'code')
matched nothing and collapsed the popup. Add ЙЦУКЕН<->QWERTY layout maps and a
buildLayoutCandidates(query) = [original, RU->EN, EN->RU]; getSuggestionItems now
matches an item if ANY candidate hits (fuzzy title / description / searchTerms),
and the tie-break sort is candidate-aware. Keeping the original among candidates
preserves genuine Cyrillic search terms (сноска -> Footnote). One-function change;
slash-command.ts allow() reuses it, so the popup-collapse is fixed transitively.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
F1: escape the collaborative page title before interpolating into
<page_changed page="..."> (and the pre-existing openedPage attr) — strip
<>" and collapse whitespace, so a crafted title can't break out of the
attribute into the system prompt (cross-user injection).
F2: neutralize <page_changed>/</page_changed> occurrences inside the diff body
so a crafted line can't close the block early.
F3: remove the dead content_hash column (written every turn, never read) —
migration, repo, service hashing + crypto import, db.d.ts, spec asserts.
F4: test the best-effort catch branches (detectPageChange / snapshotOpenPage
swallow errors and don't break the turn).
F5: soften the overstated 'diff cannot smuggle instructions' comment to
defense-in-depth framing referencing the F1/F2 mitigations + safety sandwich.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add a new value "inline" to the image align attribute (alongside
left/center/right/floatLeft/floatRight). Inline images render as
inline-block containers, so consecutive ones form a row that wraps
naturally on narrow viewports; unlike the float modes, text does not
wrap around them.
- applyAlignment: reset-then-apply extended to display/vertical-align;
the reset restores the constructor's inline display:flex so non-inline
modes keep byte-identical styles and editor-ext stays independent of
the client CSS class
- image bubble menu: new "Inline (side by side)" button (IconLayoutColumns)
with active state, mirroring the float buttons
- i18n: key registered in en-US and ru-RU ("В ряд"), like the float labels
- tests: 3 new applyAlignment specs (apply, reset on switch-away, float->inline)
- no schema/MCP/markdown changes needed: align round-trips as data-align
F1: extract the navbar-visibility crux (width/height 0 or right<=0 -> hidden)
from getNavbarRect into a pure isNavbarRectVisible in dock-helpers.ts + 3 tests;
getNavbarRect calls it (identical null cases).
F2: base the dock/undock button's label/icon/title on the effective useDock state
(docked && dockRect present) rather than the raw docked flag, so a docked window
that fell back to floating (collapsed sidebar) doesn't show 'Undock'. Toggle
action unchanged; no remount.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Icons are rendered only as <item.icon style={...} stroke={2} />, so type the
field as ComponentType<{ style?; stroke? }> instead of typeof IconBold. stroke is
string|number to match Tabler's own prop type, so Tabler icons and the local
IconStress both satisfy it without the 'as unknown as' cast.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Unit-test the focus-restore guard: an external <input> active -> editor.view.focus
NOT called (deliberate move respected); a non-focusable element active -> focus
called once. Fake editor + fake timers (rAF via setTimeout stub); view.focus is a
spy. Regression lock for the guard that keeps focus out of the page-title input.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
F1: add code-block-view.test.tsx (mirrors the footnote structure harness) asserting
the language selector renders only when editor.isEditable, and the copy button is
present in both modes.
F2: remove the now-dead justify=flex-end on the absolutely-positioned menu Group.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Drag the floating AI-chat window onto the sidebar and release over it to DOCK it
— the window pins to the live navbar rect, overlaying the page tree; a drop-zone
highlight shows while dragging over it. Closing the chat re-shows the tree.
Undock via a header button or by dragging the docked window back onto content
(pops out floating at the drop point). The docked/floating mode persists in
localStorage and the docked window follows the navbar width (manual resize,
space<->shared route change) via a ResizeObserver + sidebar-toggle/transitionend
re-sync; when the navbar is collapsed/absent the window falls back to floating
instead of vanishing. Dock/undock only flips a mode atom + geometry — ChatThread
is never remounted, so an in-flight response stream is not interrupted.
Frontend only.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The agent rebuilds context from DB each turn and didn't know the user manually
edited the open page since its last response, so it could overwrite those edits.
Add a per-turn ephemeral <page_changed> note in the system prompt (twin of
INTERRUPT_NOTE, self-clearing) carrying a unified Markdown diff of what changed
since the END of the agent's previous turn.
- New ai_chat_page_snapshots table (migration + hand-declared db.d.ts/entity
types) storing the page Markdown per (chat,page) at each turn's end.
- Pure computePageChange util (whitespace-normalized unified diff via the
existing jsdiff dep, 6KB cap + getPage hint).
- Turn start: if the open page's updatedAt moved past the snapshot, diff current
vs snapshot; non-empty -> PAGE_CHANGED_NOTE in the safety sandwich.
- Turn end: upsert the snapshot on EVERY terminal path (onFinish/onError/onAbort,
once) so the agent's own edits are excluded by construction even on aborted
turns.
All best-effort (never breaks/latency-regresses a turn); fast path when updatedAt
is unchanged. Server-only.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Select a vowel and one click places a combining acute accent over it; clicking
again removes it (toggle). Inserts the literal Unicode char U+0301 right after
the letter — plain text, not a custom TipTap mark — so it survives HTML/Markdown
export, full-text search and public share with zero server/converter changes.
Insert/remove is a single transaction (one Ctrl+Z), inherits the letter's marks
(bold/italic/color), and restores the original selection so the active state
toggles correctly. Editable bubble menu only. New pure helper stress-accent.ts
(+ 5 unit tests). i18n: en 'Stress' / ru 'Ударение'.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The row/column grip and cell-chevron menus are Mantine <Menu>s with
returnFocus:true whose targets live outside the editor's contenteditable. After
a menu action focus returns to that outside target, so ProseMirror's undo keymap
never sees Ctrl+Z until the user clicks back into a cell. Add
refocusEditorAfterMenuClose(editor): on the next frame (after Mantine's
returnFocus) restore editor focus via view.focus(), unless the user intentionally
moved to another input/editable. Wired into both onClose paths (the shared
row/column lifecycle hook + cell-chevron).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The code-block control panel (language selector + copy) took a full row above
the code. Move both to an absolute overlay in the top-right corner and hide the
language selector until the block is hovered/focused; the copy button stays
always visible. In read-only the language selector isn't rendered at all. The
<pre> (editable contentDOM) stays FIRST in the DOM so click hit-testing (#146)
is not regressed; the panel leaves the flow via position:absolute.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The banner only offered 'Make permanent'. Add a secondary destructive
'Move to trash' button that soft-deletes the note now instead of waiting for
TTL expiry, reusing the tree/header soft-delete path (useTreeMutation.handleDelete):
optimistic tree removal, the undo-toast, the deletedAt cache stamp, and the
redirect to space home. No confirm modal (project convention = undo-toast).
Gated on the existing Edit permission. Client-only, no server/i18n changes
(both labels already exist).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- F1: gate the card on rows-WITH-text (`thread.some(row => row.text.length > 0)`)
instead of thread length. A text-less root whose only reply is also text-less
would otherwise open an empty <Paper> (the render already filters empty rows).
New test locks it (parent + reply both empty → no card).
- F2: ESTIMATED_CARD_HEIGHT 200 -> 300 (= CARD_MAX_HEIGHT) so the flip-above
decision reserves the real worst-case height and a tall thread near the
viewport bottom flips up instead of overflowing off-screen.
vitest 19/19, tsc 0, eslint 0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Captures the non-obvious gotchas that make bringing up a local instance
painful: the collaboration server is a THIRD process (pnpm dev starts only
API + client) that must be built before running (tsx/ts-node fail on NestJS
DI); APP_SECRET must be identical between the API and collab servers or every
realtime connection is rejected with "Invalid collab token"; Vite binds
localhost so LAN access needs --host; a stale @docmost/editor-ext white-
screens the client; pgvector is mandatory; migrations don't auto-run in dev.
Also documents that demo/test passwords should be a simple one-word
alphanumeric (no special chars, which get mangled through shells/JSON/URLs).
Referenced from AGENTS.md (Commands + Two-server-processes sections).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Per maintainer feedback: show the comment author and the whole thread (parent
+ replies), but as simple "Author: text" lines — no avatars, timestamps, or
thread chrome ("it's already clear they're comments on one entry, one after
another"). Also lengthen the open delay so the card doesn't pop up on a
passing glance.
- Render each comment in the thread as a plain line: bold "Name:" + text,
parent first then replies (createdAt asc). Empty-text comments are skipped.
- OPEN_DELAY_MS 120 -> 350.
- Drop the avatar/relative-time/divider UI (and the CustomAvatar/timeAgo
imports). buildThread (root + direct replies) is unchanged — the comment
model is flat, so direct children of the root are the full thread.
Tests updated to the "Author: text" shape (textContent-based, incl. ordering).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Lock the feature's central invariant — the tooltip must never intercept the
comment-mark's click (which opens the side panel). pointer-events:none is the
single property guaranteeing that, and it was unasserted: a regression dropping
it from the style object would let a lingering card swallow the click with no
test failing. Assert it in the "shows after delay" test.
Adds CommentHoverPreview, mounted in page-editor next to <EditorContent>:
hovering a `.comment-mark[data-comment-id]` span shows a small floating card
(createPortal, position:fixed, pointer-events:none so it never intercepts the
mark's click) with the parent comment's plain text. Uses useCommentsQuery
(shares the ["comments", pageId] cache with the side panel — no extra
request). Skips unknown/not-yet-loaded, resolved (data-resolved attr or
resolvedAt/resolvedById), and empty-text comments. A ~120ms open delay avoids
flicker; hides on mouseout / mousedown / scroll(capture) / resize / page
change. commentContentToText flattens the comment's ProseMirror doc
(stringified or parsed) to plain text, preserving hardBreaks as newlines and
never throwing. Main editor only (read-only / shares / history out of scope).
closes#268
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Review round 1 on the scroll-position feature:
- F1: add two tests for the hook's subtlest invariants — (a2) the restore
target is captured synchronously at mount and survives a fresh scroll@0
overwriting storage on load (a regression moving the capture into an effect
would now fail); (a3) restore runs at most once per mount even when called
again (the wiring effect can re-run).
- F2: log instead of silently swallowing sessionStorage errors in
readStorage/writeStorage (AGENTS.md "errors must never be swallowed" rule);
no user notification since a missed scroll restore is not actionable.
- F3: document the hard dependency on PageEditor remounting per page
(key={page.id}) at the refs declaration — the per-mount refs are not reset
on an in-place pageId change, so removing that key would break restore on
the 2nd page.
vitest 9/9, tsc 0, eslint 0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds useScrollPosition(pageId): saves window.scrollY to sessionStorage
(key gitmost:scroll-position:<pageId>) on throttled scroll / pagehide /
visibilitychange / cleanup, capturing the previously-saved value
synchronously at mount before any handler can overwrite it with the fresh 0.
restoreScrollPosition() (wired in page-editor.tsx to fire once the live
content is laid out, !showStatic && editor) yields to a #hash anchor, then
polls the document height and scrolls to the saved Y once the content is
tall enough, with a 5s timeout clamped to the max reachable position. All
storage access is try/caught so a disabled/quota'd Storage never breaks the
page. The in-flight restore poll is held in a ref and cancelled on unmount,
so a fast SPA navigation can't scroll the next page. closes#266
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Reviewer round 1 on the #260 collab-doc-name fix:
- F1: replaceImage is the one path where the resolved UUID gates BOTH the
collab-doc open AND the per-page mutex key (withPageLock(pageUuid)). Add a
deterministic test to resolve-page-id-collab-doc-name.test.mjs: it gates
/files/upload so replaceImage parks mid-upload holding its lock, asserts the
doc opened as page.<uuid> (never page.<slug>), and probes the SHARED
page-lock chain — a withPageLock(UUID) probe must stay blocked while
replaceImage holds it (with a free-key probe as a non-vacuity guard). The
test fails if the lock key is reverted to the slugId (verified).
- F2: drop the dead `pageIdCache.set(uuid, uuid)` — resolvePageId returns on
the isUuid() short-circuit before the cache is ever read with a uuid key, so
only slugId->uuid entries are stored/read. Comment corrected to match.
MCP suite 430/430, tsc 0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Real root cause of the silent MCP edit loss: the web editor always opens the
collaboration document by the page UUID (`page.${page.id}`), but the MCP
opened it by the agent-supplied id — usually a slugId — so `page.${pageId}`
became `page.<slugId>`. For one DB page that is TWO independent Yjs documents;
both persist to the same `pages` row (findById/updatePage resolve id or
slugId), so the human tab's debounced store overwrites the agent edit
(last-store-wins) — gone after reload, never shown live. The slugId doc also
made the server's transclusion sync + embedding reindex throw Postgres 22P02.
Fix:
- MCP (primary): resolvePageId(pageId) returns the canonical UUID — a UUID
short-circuits with no network call, a slugId resolves once via getPageRaw
and is cached both ways. Every collab-write path (mutatePageContent /
updatePageContentRealtime / replacePageContent and the mutate/replace/
unlocked seams) now opens by the resolved UUID, so the MCP and the editor
share ONE Yjs doc. replaceImage's whole-operation page lock also keys on the
UUID so it serializes against the other (now-UUID-keyed) writes.
- Server (defense + kills the 22P02 noise): onStoreDocument passes the resolved
page.id — not the raw doc-name id — to syncTransclusion, the embedding queue,
the mention-notification job, addContributors, and the in-tx history read.
Content store and the empty-guard are untouched.
Tests: a new MCP test stands up a real Hocuspocus server and asserts a slugId
input opens `page.<uuid>` (never `page.<slugId>`), with UUID short-circuit and
single-resolve caching; the server spec asserts the side-effects receive the
UUID for a `page.<slugId>` doc. closes#260
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The reviewer noted the in-order emitter's else branch (a NOT-next-to-emit
segment failing → buffer an empty placeholder so the drain can skip it,
use-streaming-dictation.ts:215-218) was the one reachable ordering branch
left uncovered. Add a non-vacuous case: with 3 segments, reject seq 1
(out of order) → one notification, nothing emitted; resolve seq 0 → "alpha";
resolve seq 2 → "gamma". The seq-2 flush proves the empty placeholder let the
emitter advance PAST the failed seq 1 — without the else branch the drain
would stall at the missing seq 1 and "gamma" would never emit.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
After "Reindex now" the "Indexed X of Y" counter froze at 0 until a manual
reload. Root cause is purely client-side: right after the mutation the
client still holds the PRE-reindex settings snapshot, which for an already
fully-indexed workspace reads reindexing=false, indexed>=total. The
deadline-clearing effect evaluated isReindexComplete() against that stale
snapshot, read it as "done", and cleared the poll deadline before the first
post-reindex poll ever landed — so polling never ran and the counter stayed
at 0 (a reload just fetched one fresh snapshot).
Gate completion on having actually observed the active run: a
reindexSeenActiveRef, reset on each new reindex (mutation onSuccess, before
setting the deadline) and latched true once a poll reports reindexing=true.
isReindexComplete(status, seenActive) and nextReindexPollInterval now require
seenActive, so the stale fully-indexed snapshot no longer reads as finished.
The server pre-seeds reindexing=true from enqueue time, so seenActive latches
early and a genuine completion still stops polling promptly; the
REINDEX_POLL_CAP_MS cap is checked first and always wins, so polling can
never run away. closes#262
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Backfill the two genuinely-uncovered infra-free units from the #244 Part B
test backlog (the rest was already covered by #248/#257):
- use-streaming-dictation: the in-order transcription emitter. Drives the
real hook via renderHook with mocked VAD + deferred transcribeAudio so the
test controls response order. Asserts out-of-order HTTP responses still
emit text in segment order; whitespace trimmed and empty results dropped
while the sequence advances; a failed segment shows one notification and is
skipped so later segments still flush; a response resolving after cancel()
is dropped (stale-epoch guard).
- internal-link-paste (handleInternalLink / createMentionAction): validateFn
reject → no resolve/dispatch; resolve → mention node with the resolved page
+ anchor dispatched via replaceWith at pos; "Untitled" fallback; reject →
raw url inserted as text under a link mark; createMentionAction wiring to
getPageById on success + failure.
Test-only; no production code changed.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Update GitHub workflow services to pull PostgreSQL and Redis images from `mirror.gcr.io` instead of Docker Hub. This avoids anonymous pull rate‑limit failures on shared GitHub runner IPs by using the Docker Hub pull‑through cache.
The [Unreleased] compare link still pointed at v0.93.0 even though the
0.94.0 release section already exists, and there was no [0.94.0]
link-reference at all (the header was unresolvable). Point [Unreleased] at
v0.94.0...HEAD and add [0.94.0]: v0.93.0...v0.94.0 so every version header
resolves. closes#258
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
stashPage is declared in the server's DocmostClientLike interface and
shipped as the stash_page MCP tool (client.ts, tool-specs.ts, index.ts),
but the hand-maintained HOST_CONTRACT_METHODS mirror in the contract test
was never updated — so the drift-guard test failed and broke CI's
unit-test job. Add the missing name; both directions now agree.
agent_coder
vvzvlad
claude_code