Финальный линк Фазы 1б. Инвентарь тулов жил в 4 рукописных прозаических копиях
(SERVER_INSTRUCTIONS под regex-тестом; <tool_catalog>; имена в ai-chat.prompt.ts
без гарда; README) — роадмап #416 планировал 4 последовательных ручных правки
этого текста (#411/#412/#413/#415).
- SERVER_INSTRUCTIONS разбит (новый модуль server-instructions.ts): ROUTING_
PROSE (рукописные intent-подсказки «когда что» — осмысленно ручные, перенесены
ДОСЛОВНО со всеми предостережениями: <=250 у create_comment, soft-delete у
delete_page, baseHash у drawio_update, PUBLIC у share_page) + buildToolInventory()
— генерирует <tool_inventory> из реестра (mcpName + purpose из catalogLine,
группировка по TOOL_FAMILY, бакет OTHER ловит незамаппленное → тул нельзя
тихо потерять) + 5 inline MCP-only (INLINE_MCP_INVENTORY). Детерминирован
(семейства FAMILY_ORDER, имена localeCompare). regex-тест server-instructions
удалён; структурные гарантии — в новом tool-inventory.test.mjs (точное
членство множества сильнее старого \b-скрейпа).
- Имена тулов в ai-chat.prompt.ts → через экспорт PROMPT_TOOL_NAMES; новый гард
ai-chat.prompt.tool-names.spec.ts: каждое имя — реальный тул реестра, скан
guidance-нот на camelCase-токены падает на несуществующем (escape-
нейтрализация против ложных nThe-токенов).
- INLINE_TOOL_TIERS уже содержал ровно 8 genuinely-inline тулов (после #445) —
сжатие не потребовалось.
Критерий: добавление/переименование спека меняет инвентарь БЕЗ правки прозы.
Внутреннее ревью: APPROVE — фактическим прогоном подтверждено, что НИ ОДИН тул
из старого SERVER_INSTRUCTIONS не выпал (диф старый-vs-новый пуст; добавился
get_workspace, раньше прятавшийся в EXCEPTIONS); проза дословна; инвентарь
полон/детерминирован/без фантомов; гард краснеет на обеих ветках провала.
613 node + 289 jest зелёные. Стоит на #445 — мержить последним в стопке 1б.
README-каталоги вне обязательного скоупа (docs-скрипт) — в чек-лист #412.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Ядро Фазы 1б. Реестр (#294) шарил только метаданные (имя/схема/описание/tier),
но НЕ execute-логику — у каждого shared-тула было ДВА рукописных execute-тела
с копией маппинга аргументов (MCP registerShared в index.ts; in-app sharedTool
в ai-chat-tools.service, зеркалящий MCP-транспорт вручную). Корень
повторяющихся parity-багов (f46d89ea drawio, f8d26420 stashPage, fc9088b7
node-args): добавление одного тула = 7-9 согласованных ручных правок в двух
пакетах.
- SharedToolSpec расширен: канонический execute(client, args) (чистый JS —
свободно пересекает zod-мажорную границу v3/v4) + оверрайды
mcpExecute/inAppExecute/mcpOnly/inAppOnly для ОСОЗНАННЫХ per-layer различий.
client: DocmostClientLike (Pick из #446). Канон возвращает СЫРЬЁ, каждый хост
накладывает свой конверт (MCP jsonContent, in-app как есть); override владеет
результатом хоста целиком.
- Оба хоста → циклы по Object.values(SHARED_TOOL_SPECS): index.ts registerShared
39→0 (цикл), ai-chat-tools.service sharedTool ~40→1 (цикл). Добавление спека
автоматически регистрирует тул в ОБОИХ хостах — сценарий PR #434 невозможен
по построению.
- Осознанные различия через overrides (ни одно не сплющено к одному хосту):
оба mcpExecute+inAppExecute — createPage/movePage/deletePage/
exportPageMarkdown/createComment (guardrails, конверты, проекции, тексты
ошибок); execute+inAppExecute — getPage/renamePage/resolveComment;
execute+mcpExecute — stashPage (resource_link+structuredContent),
checkNewComments (since-guard только на MCP).
- Оставлены inline (по делу): update_comment/delete_comment (MCP-only, in-app
не даёт хард-правку/удаление комментов), search/transformPage (per-transport
дивергенция — hybrid RRF / без deleteComments), table_get (noun-vs-verb
naming clash — уедет после camelCase #412), getCurrentPage/updatePageContent/
listSidebarPages/getComment/getPageHistory (in-app-only, per-request state).
- Guard-тесты (contract-parity, phantom-catalog) сохранены — теперь инварианты,
не «последняя линия».
Внутреннее ревью: APPROVE, построчная BEFORE/AFTER-сверка по каждому shared-тулу
на обоих хостах — ни одного изменённого per-host поведения (порядок/дефолты
аргументов, guard'ы, конверты, проекции сохранены), множества тулов побайтово
совпадают (48 in-app, 45 MCP), кросс-zod-граница чистая (нет z. в execute),
611 mcp + 260 server тестов зелёные. Ядро Фазы 1б, стоит на #446 — мержить после.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Восстановленный отложенный долг #294: @docmost/mcp не отдавал .d.ts, поэтому в
сервере жили ТРИ дрейфующие ручные копии одних и тех же имён/сигнатур
(DocmostClientLike ~230 строк, копия SharedToolSpec, name-only HOST_CONTRACT_
METHODS-тест). In-app execute-тела зовут клиент ПОЗИЦИОННО, так что перестановка
параметра в client.ts доезжала до прода рантайм-ошибкой без сигнала на компиляции.
- declaration:true (+declarationMap) в packages/mcp/tsconfig.json; types-экспорт
в package.json (exports → conditional {types, default} для . и ./http;
require.resolve/dynamic-import резолвят default → build/index.js, рантайм не
тронут). build/index.d.ts эмитится, реэкспортит DocmostClient + SharedToolSpec.
Правок исходников пакета для эмита НЕ потребовалось.
- DocmostClientLike → Pick<DocmostClient, 48 методов> из type-only import
(стёрт на компиляции, ESM/CJS-границу не задевает); ручное зеркало удалено.
- SharedToolSpec → type-only реэкспорт из пакета; ручная копия удалена.
- client-host-contract.test.mjs удалён целиком — имена И сигнатуры теперь
проверяет tsc.
- Позиционная безопасность: never-called __assertClientCallContract(client:
DocmostClientLike) воспроизводит каждый позиционный вызов с типизированными
плейсхолдерами (AI-SDK стирает вход execute-замыканий в any, иначе позиционные
вызовы не проверялись). Перестановка параметров client.ts → ошибка компиляции
сервера ровно тут. Loose as-касты в ai-chat-tools.service не потребовали
правок; as any не добавлялся.
Внутреннее ревью: APPROVE. Runtime resolution через conditional exports не сломан
(разобрано для прод-инсталляции, не только symlink); покрытие
__assertClientCallContract полное (48 call-sites == union == assert, сверено
программно); Pick полон; демонстрация reorder → TS2345 в assert. Единственная
находка (Promise<any> в части возвратов) предсуществующая в client.ts, вне
цели PR. Стоит на #447 (закрытие skew build/vs/src) — мержить после него.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Два структурных слепых пятна: (1) сервер грузит СКОМПИЛИРОВАННЫЙ
packages/mcp/build/index.js, а серверные guard-тесты читают src/tool-specs.ts —
правка src без пересборки оставляет тесты зелёными, но рантайм расходится со
спеками; (2) спеки добавляют в packages/mcp, а parity-тесты живут в jest-сьюте
apps/server — PR, трогающий только пакет, проходит зелёным, сломанная in-app-
проводка всплывает уже на develop (кейс f46d89ea).
- REGISTRY_STAMP: codegen (scripts/gen-registry-stamp.mjs) считает sha256 от
нормализованного (CRLF→LF, один хвостовой \n снят) сырого текста
src/tool-specs.ts, пишет src/registry-stamp.generated.ts (gitignored),
index.ts реэкспортит → попадает в build/. Вшит в build/pretest/watch ДО tsc.
- Loader (dev/test): computeSrcRegistryStamp пересчитывает стамп из src рядом с
build/index.js (dev-vs-prod по existsSync, любая ошибка → null), сверяет с
build-стампом → при рассинхроне бросает «build is stale — rebuild». В prod
(src нет) и на pre-#447 билдах (нет REGISTRY_STAMP) — чистый no-op.
- CI: job mcp-server-parity собирает shared-deps+mcp (регенерит стамп) и гоняет
ОБА сьюта вместе (mcp node:test + server guard-спеки) — именованный гейт, его
нельзя случайно расщепить.
- AGENTS.md: правка спеков требует ребилда @docmost/mcp.
Тесты (20): mcp-сайд (детерминизм, нормализация, desync-гард стамп-vs-билд) +
server-сайд (null при отсутствии src = prod no-op; mismatch → throw точного
сообщения; pre-#447 no-op). Кросс-импл equality-гард: один фиксированный вход →
один хэш на ОБЕИХ сторонах, ловит рассинхрон двух нормализаций. Внутреннее
ревью: APPROVE WITH SUGGESTIONS (обе — покрытие guard'а — закрыты этим тестом).
Мутационно: любой из двух normalize-имплов расходится → equality-тест краснеет.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
resolve/unresolve enqueue — fire-and-forget (void ...catch(warn)): смысл #399
в том, что недоступность очереди НЕ должна ронять HTTP-запрос. Delete-путь уже
покрыт (enqueue awaited перед hard-delete), а reject resolve-пути — нет. Тест:
generalQueue.add реджектит -> resolveComment всё равно resolves (не throws) +
warn залогирован (ошибка проглочена на микротаске после возврата, поэтому
flushMicrotasks перед ассертом). Мутационно: сделать enqueue awaited без catch
-> тест краснеет.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
resolve/unresolve/delete comment-mark синхронно дёргали collab-gateway
(handleYjsEvent) прямо в HTTP-запросе — сетевой раунд-трип к collab на
горячем пути. Теперь:
- DB-строка пишется синхронно (источник истины) с общим таймстампом;
- сама mark-операция уходит идемпотентным COMMENT_MARK_UPDATE-джобом на
GENERAL_QUEUE, воркер проигрывает тот же handleYjsEvent;
- resolve/unresolve — fire-and-forget (best-effort), delete — await энкью
ДО необратимого hard-delete (durability split);
- race-guard: устаревшее ПРОТИВОПОЛОЖНОЕ событие (ts <= updatedAt строки и
состояние расходится) пропускается, а не флипает mark в устаревшее;
- DI-цикл обойдён ленивым moduleRef.get(CollaborationGateway, strict:false).
Внутренний цикл: 2 прохода. Правки по внутреннему ревью: `<` → `<=` в
race-guard (безопасная обработка суб-миллисекундной ничьи двух
противоположных тогглов); задокументирован сознательный компромисс —
транзиентный page.updated-broadcast из воркера несёт только {id}, теряя
name/avatarUrl «кто редактировал» (lastUpdatedById выставляется верно,
косметика, самочинится на следующем реальном редактировании).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
PR #434 (drawio stage 1) added drawioGet/drawioCreate/drawioUpdate to the
shared tool-spec registry with in-app metadata (inAppKey, deferred tier,
catalogLine) but wired them only in the standalone MCP server, breaking the
contract-parity and phantom-catalog unit tests on develop CI.
- expose drawioGet/drawioCreate/drawioUpdate in forUser() via sharedTool(),
mirroring the MCP transport's argument mapping (format ?? 'xml' default,
flat schema regrouped into the client's `where` object, positional
baseHash pass-through)
- extend the DocmostClientLike hand-mirror with the three client methods
- append the three names to the HOST_CONTRACT_METHODS drift-guard whitelist
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- F1 [medium — the substantive one]: hasRestrictedPagesInWorkspace is now UNCACHED
(a plain EXISTS per call, like its sibling hasRestrictedPagesInSpace). Caching it
(even 5s) reintroduced an access-control leak the space path never had: a
concurrent whole-workspace read in the insert->commit window of the FIRST
restricted page could re-populate `false` under withCache (read-then-set, no
del-during-read guard) and override the insert-time bust, leaking that page to
unauthorized users for up to the TTL. Uncaching removes both the DB/cache
asymmetry and the TOCTOU race; the space path already accepts this per-call cost.
Reverted the now-unnecessary insertPageAccess cache-bust and removed the dead
HAS_RESTRICTED_PAGES_IN_WORKSPACE cache key.
- F2 [test]: page-permission-workspace-filter.int-spec.ts (real PG) — the
short-circuit returns the full input set with zero restrictions AND filters out
the page the user can't reach when a restriction is present (proving the authz
behavior is unchanged), the 0->1 transition flips immediately, and the flag is
per-workspace scoped.
- F3 [doc]: documented the deploy-time write-lock in the migration header — the
non-CONCURRENT GIN trigram builds take a SHARE lock that blocks writes on
pages/users/… for minutes on a large tenant; run in a maintenance window or
build CONCURRENTLY out-of-band for big installs.
- F4 [doc]: corrected the jwt.strategy comment — the reused req.raw.workspace is
the middleware's selectAll superset (not "the exact row this query returns"),
harmless because AuthWorkspace already preferred that object.
Gate: server tsc 0; the new int-spec 3/3 on real Postgres.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
One migration + targeted hot-path fixes. API behavior 1:1 (schema change = added
indexes + a byte-identical f_unaccent function-body swap, see below).
- Trigram + composite indexes (20260705T120000-perf-indexes.ts): GIN trigram on
LOWER(f_unaccent(title/name)) for pages/users/groups (the /search/suggest
leading-wildcard LIKE did a seq scan per keystroke — EXPLAIN now confirms
Bitmap Index Scan on idx_pages_title_trgm), + page_history(page_id,id DESC),
comments(page_id,id). DEVIATION (verified byte-identical): PG18 cannot inline
the two-arg f_unaccent body during index creation, so up() swaps it to the
schema-qualified single-arg `SELECT public.unaccent($1)` — same dictionary,
identical output for all inputs, so the tsvector trigger + main @@ search stay
consistent with NO reindex; down() restores the exact two-arg body.
- Auth path: jwt.strategy reuses req.raw.workspace when workspaceId matches (the
middleware already validated it) instead of re-querying; domain.middleware
caches the workspace lookup (withCache 15s, invalidated in all 8 WorkspaceRepo
mutators, with a Date reviver for the JSON-serialized cache). USER + SESSION
caching DEFERRED — the invalidation surface (role change doesn't revoke
sessions; revocation includes background jobs) can't be safely covered, and a
missed hook on a security path is worse than the win.
- AI re-embed coalescing: aiQueue.add gets {jobId: embed-<id>, delay: 30s} so
active editing collapses to one job (worker reads current page state).
- filterAccessiblePageIds: hasRestrictedPagesInWorkspace short-circuit skips the
recursive-ancestor CTE when a workspace has zero restricted pages (wired from
search/favorites/notifications/recent/created-by). EXISTS on the same pageAccess
table the CTE anti-joins → no false-positive / no access leak. Busts the cache
on insertPageAccess so a 0->1 restricted transition takes effect immediately
(review F1).
- Small: syncTransclusion guarded by a family-node probe (both old+new content, so
the removal path is preserved); mention notifications enqueue only when the set
gained a member; redis maintainLock clears a prior interval (leak fix).
Skipped as risky (flagged): global ValidationPipe transform change; a pool-wide
statement_timeout (would kill long CREATE INDEX migrations on the same pool).
NOTE: kept the trash query's `content` select — the trash UI reads page.content
for its preview modal (review F3, would have regressed).
Gate: server tsc 0; jest page-permission/auth/search/persistence 15 suites pass;
migration up+down+idempotency verified on real PG18 with EXPLAIN confirming index
use. No new deps.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Аудит при подготовке #413 нашёл дрейфующие дубли между packages/mcp и
packages/prosemirror-markdown. Четыре дедупа (поведение тулов не меняется):
1. node-ops: форк ~960 строк сведён в ОДНУ копию в prosemirror-markdown (живая
mcp-версия — строгое надмножество замороженного #293-seed'а пакета; сверено по
git-истории, новая пакет-копия байт-в-байт == прежней mcp-копии). Barrel-экспорт
полной поверхности; mcp/client.ts/page-search.ts/transforms.ts/collaboration.ts
импортируют из пакета; тесты переехали. node-ops тянет stripInlineMarkdown ->
пакет-локальная text-normalize.ts несёт только этот примитив (mcp-версия —
домен #408; заголовок документирует дубликацию + источник истины).
2. footnote-lex/footnote-analyze (vestigial legacy [^id]: диагностика): сведены к
одному fence-aware предупреждению 'reference-style footnotes -> use ^[...]'
(полезно для класса #410); footnote-lex удалён.
3. footnote-authoring -> примитивы (footnoteContentKey/makeFootnoteDefinition/
generateFootnoteId) перенесены в пакетный footnote.ts, одна реализация конвенции.
4. parse-node-arg -> перенесён в prosemirror-markdown (не mcp: сервер CommonJS не
импортирует ESM-only @docmost/mcp, но нативно импортирует пакет), обе копии
удалены, консьюмеры перенаправлены.
canonicalizeFootnotes/ENFORCEMENT RULE #228 и comment-anchor/json-edit/text-normalize
(mcp) не тронуты. API-поверхность node-ops оставлена чистой для #409/#413.
closes#414
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
wrapToolsWithCommentSignal всегда ставил свой toModelOutput, молча выбрасывая
собственный toModelOutput инструмента (латентная ловушка — будущий тул со своим
toModelOutput тихо сломался бы). Теперь база = origToModelOutput(info) при наличии,
иначе воспроизведённый дефолт SDK; no-signal путь возвращает базу дословно, signal-
путь = части базы (modelOutputToParts: text/json/content) + элемент сигнала последним.
execute по-прежнему возвращает СЫРОЙ результат -> part.output/цитаты байт-идентичны.
Дефолтный путь (единственный исполняемый сегодня) байт-идентичен и SDK-дефолту, и
до-фиксовому signal-пути (проверено повторным ревью). json-ветку загардил ?? null
для симметрии с fallback. +2 теста: тул со своим text/content toModelOutput —
база честно сохраняется и в no-signal, и в signal (сигнал добавлен последним).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Комментарии человека находят агента сами: короткая эфемерная строка
'new comments: N on page … — call listComments(pageId)' в результате ЛЮБОГО
tool-вызова, mid-turn. Общий хелпер packages/mcp/src/comment-signal.ts
(createCommentSignalTracker: watermark + per-page debounce + working-set;
buildCommentSignalLine; defangCommentSignalTitle).
- Standalone MCP (index.ts): второй wrapper в choke point registerTool (паттерн
метрик #402) — отдельный {type:'text'} content-элемент, форма результата не
меняется. Источник: rate-limited listComments по working-set, title через
getPageRaw только на hit. State per-session.
- In-app (ai-chat-tools.service.ts): execute ВСЕГДА возвращает сырой результат
(part.output/цитаты не трогаются), сигнал доставляется модели через отдельный
toModelOutput ({type:'content', value:[raw, signal]}) — зеркало MCP; no-signal
ветка точно воспроизводит дефолт SDK. Источник: REST-probe (осознанный форк от
DB-count из ТЗ — чтобы не менять конструктор сервиса и не ломать спеки).
- Инъекционная защита: в сигнал идут только count+pageId+defanged-title, НИКОГДА
текст комментария (untrusted). Per-page watermark (не глобальный) — комментарии
на второй странице не теряются.
closes#417
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
В ai@6 упавший тул — это tool-error часть в step.content ({type,toolCallId,
toolName,input,error}), а не элемент toolResults. Раньше serializeSteps писал
только toolCalls+toolResults (ошибка терялась, orphan tool-call без результата),
а assistantParts эмитил заглушку 'Tool call did not complete.' (реальный текст
ошибки терялся для мультиходового реплея — модель не знала, почему упало, и
повторяла ошибку).
- StepLike расширен полем content; новый хелпер normalizeToolError (Error/string/
object -> строка, обрезка через существующий compactValue/лимиты).
- serializeSteps: на каждый tool-error пушит парный {toolName, error} тем же
паттерном, что успешный {toolName, output} -> колонка tool_calls фиксирует сбой.
- assistantParts: при наличии tool-error эмитит output-error с РЕАЛЬНЫМ текстом;
заглушка остаётся только для по-настоящему непарных вызовов (прерванных).
- docs/reading-ai-logs.md обновлён под новую форму + cutover-оговорка.
Обратно совместимо: старые строки читаются как раньше, error-элемент аддитивен.
closes#407
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Read-tool outputs were compacted at a 4000-byte gate before being stored
in metadata.parts, which is re-sent to the model every later turn. Whole-page
reads (tens of KB) got shrunk to a 500-char preview plus a "[truncated N chars]"
marker; on the next turn the model read that marker as a source truncation and
re-read the page, wasting tokens and producing wrong behavior.
- Raise MAX_TOOL_OUTPUT_BYTES 4000 -> 200_000 so normal reads are stored and
replayed verbatim; only a single >200 KB output is compacted as a backstop.
- Reword the inline string marker so it reads as a replay-history elision, not
a source truncation, and tells the model it can re-call the tool.
- Update doc comments and enlarge the compactToolOutput unit-test inputs above
the new 200 KB gate.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The Researcher role wrote 40 literal `^[...]` and zero real footnotes: its
incremental write path (insertNode/editPageText) doesn't parse markdown, and
the footnote-capable tool was MCP-only. Promote three tools from inline
MCP-only to the shared registry so the in-app agent gets them too.
- tool-specs.ts: insertFootnote/insertImage/replaceImage added to
SHARED_TOOL_SPECS (mcpName/schema/description moved VERBATIM from the inline
registrations — MCP names + behaviour unchanged for external clients).
- index.ts: the 3 inline registerTool calls become registerShared; drop the
"MCP-only by design" comments.
- ai-chat-tools.service.ts: register the 3 in-app via sharedTool ->
client.insertFootnote/insertImage/replaceImage (imageUrl->url,
attachmentId->oldAttachmentId mapping).
- tool-tiers.ts: insertFootnote -> core (else the original asymmetry recurs —
footnote tool hidden while editPageText is core); images -> deferred.
- research/{en,ru}.yaml FOOTNOTES: `^[...]` parses ONLY on a whole-markdown
write (create/update/import); for a pinpoint citation to existing text use
insertFootnote; via editPageText/insertNode it stays literal.
- json-edit.ts guardrail: an edit_page_text `replace` containing a `^[...]`
token is refused into failed[] with an insert_footnote hint, mirroring the
existing formatting-marker refusal. (Slightly broader net than that mirror —
a literal `^[a-z]` regex class in a replace is also refused; accepted
defense-in-depth, has a no-false-positive test.)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Follow-up to the merged MCP-hang fix (16b476a2); post-hoc review DO. The
fix itself is unchanged and prod-working — this adds the coverage + one
coherence fix the review asked for.
Tests (ai-chat.service.setup-abort.spec.ts):
- onLateResolve late-close: a toolset that resolves AFTER the setup race
was lost has its leased clients released (close spy asserted).
- pure 60s deadline (signal NOT aborted): the turn proceeds Docmost-only
(reaches streamText, run NOT finalized 'aborted') — the defense-in-depth
backstop, previously untested.
- legacy no-runId: a setup abort does NOT re-throw (the `runId &&` guard);
together with the deadline test this locks both halves of the catch guard.
Coherence (external-mcp/mcp-clients.service.ts buildEntry):
- The per-server connects now run via Promise.all instead of a sequential
for-await, so total build time is bounded by the slowest single server
(~2×CONNECT_TIMEOUT_MS) instead of the sum. At >6 all-timing-out servers
the sequential build exceeded the outer 60s deadline, inverting the
"per-server bound primary, outer deadline is backstop" invariant. Merge
is done in a sequential post-Promise.all loop in server order, so tool-key
precedence/disambiguation, outcomes and client ordering are byte-identical
to the sequential build; each server keeps its own timeout + failure-close.
Comments: the onLateResolve note now says it releases the lease (refcount),
not force-closes transports (the cache owns them); the invariant comment
reflects the parallel build.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A single NUL (U+0000) in model/tool output (e.g. a truncated multibyte
read of a web page) is rejected by Postgres in BOTH the content (text) and
toolCalls/metadata (jsonb) columns, so it failed EVERY write of the
streaming assistant row ("invalid input syntax for type json") and silently
dropped the turn's content from the DB while the live stream still showed it.
- add stripNulChars: deep-strips NUL from all strings, returns the same
reference when there is nothing to strip (no needless clone)
- apply it at the flushAssistant choke point (covers content + toolCalls +
metadata for the seed, per-step and terminal writes)
- tests: deep-strip, same-reference, end-to-end via flushAssistant
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Reduce the default external-MCP silence timeout from 5 min to 1 min and the
overall call timeout from 15 min to 2 min. Update related tests and comments.
Increase Fastify's JSON body limit to 25 MiB (configurable via
HTTP_JSON_BODY_LIMIT) to accommodate large AI‑chat payloads.
BREAKING CHANGE: shorter MCP timeouts may abort long‑running tool calls that
previously succeeded with the older defaults.
A transient network blip during an external-MCP handshake left createMCPClient
pending forever (@ai-sdk/mcp does not settle on abort). getOrBuildEntry caches the
per-workspace build PROMISE, so the never-settling connect poisoned the cache and
every later turn hung at step_count=0 before streamText — the run never finalized,
the row stayed 'running', and the chat was permanently blocked with
A_RUN_ALREADY_ACTIVE (an explicit Stop could not interrupt the un-signalled setup).
- mcp-clients: wrap connect() in a settling timeout (connectWithTimeout) that
closes a late-arriving client, so a hung handshake rejects instead of poisoning
the build cache; the bad server is skipped and the build completes.
- mcp-clients: close a connected-but-unregistered client when tools() fails,
fixing a pre-existing transport leak in buildEntry.
- ai-chat.service: bound the toolsFor build with the run's abort signal AND a
deadline (raceAgainstAbortAndTimeout); re-throw an explicit Stop only when a run
exists (runId) so the run finalizes as 'aborted', and keep the legacy
socket-bound path unchanged; settle the run 'aborted' vs 'error' accordingly.
- tests: cache-not-poisoned + orphan-client-close-once + Stop-during-setup
finalizes the run once.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Snapshot the editor selection at send time (same live-ref pattern as
openPageRef in prepareSendMessagesRequest), carry it nested inside
openPage so it dies with the page on a fail-closed resolve, and surface
it to the model only through the existing core getCurrentPage tool.
The selection TEXT is returned exclusively in the tool result (untrusted
collaborative-page content, treated as data by SAFETY_FRAMEWORK); the
system prompt gets only a fixed one-line flag, never the text/before/
after. sanitizeSelection caps text (4000), before/after (200), blockIds
(<=64 chars, <=20). Selection is a hint, not ground truth — the tool
description tells the agent to localize the fragment before editing.
closes#388
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Under a fully parallel `pnpm -r test` run the suite computed five separate
bcrypt cost-12 hashes (one per test, ~300ms idle, multi-second with all cores
saturated) and tripped jest's default 5s per-test timeout ("DISABLED user"
test, suite 31s under load vs 6s isolated).
- compute the hash ONCE in a top-level beforeAll (covers both describe
blocks) and share the read-only string across the five former call sites
- jest.setTimeout(30_000) at module scope so the per-test bcrypt compares
inside verifyUserCredentials get headroom under load too
No change to test names, assertions, order, or the CREDENTIALS_MISMATCH
contract semantics. Suite: 8/8 green, 4.8s isolated (was ~6s).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Do 1 [test-coverage]: контроллерный тест на флаг-гейт begin-hook open() — при
OFF beginRun зовётся (durable-ран независим), а streamRegistry.open НЕ зовётся
(закрывает регресс: пустая entry → non-null paused attach → зависший SSE вместо
204); при ON — open зовётся с (chatId, runId).
Do 2 [stability]: байт-кап очереди pending paused-подписчика. pendingBytes +
overflowed на Subscriber; в paused-ветке ingestFrame при превышении
SUBSCRIBER_MAX_BUFFERED_BYTES (8MB) подписчик помечается overflowed, pending
чистится, он выбрасывается из entry.subscribers (как overflowed-entry). start()
на overflowed → onEnd (чистый 204-эквивалент, без частичного реплея). Контракт
«start() в том же тике, что attach()» задокументирован в коде — кап это
структурный бэкстоп для phase-2 Redis-await шва. Юнит-тест: paused A + live B,
9×1MB > cap → A выброшен (0 доставок), B получает все 9 живьём, поздний start(A)
→ один onEnd без реплея.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
PR 1 of 2 (#381, фаза 1.5 #184): серверный реестр SSE-стримов агентских ранов,
чтобы любая вкладка могла подключиться к живому рану с реплеем кадров + живым
хвостом. «Спящий» — весь провод за флагом AI_CHAT_RESUMABLE_STREAM (off по
умолчанию); клиент (PR 2) ещё не написан.
- ai-chat-stream-registry.service.ts: in-memory реестр (open/bind/abortEntry/
attach). attach — снапшот+подписка в ОДНОМ синхронном блоке (инвариант 4: нет
await между `subscribers.add` и `frames.slice()`), paused-подписчик, overflow,
retention с identity-guard (инвариант 2), open поверх live entry даёт ровно
один onEnd (инвариант 3), anchor против кросс-ранового реплея (инвариант 6).
- ai-chat.controller.ts: begin-хук open(chatId, runId) + GET-attach эндпоинт
(403 чужой чат; 204 нет-entry/finished/anchor-мисматч; cleanup до первой
записи + recheck req.raw.destroyed; cap→destroy).
- ai-chat.service.ts: tee SSE-кадров в реестр (consumeSseStream + generateMessageId,
гейт на runId && flag) + abortEntry из внешнего catch.
- environment.service.ts: флаг isAiChatResumableStreamEnabled().
Флаг OFF ⇒ байт-в-байт legacy И #184-фаза-1 (нет start.messageId, нет tee).
Инжектируемые провайдеры НЕ @Optional() → поломка вайринга роняет старт, а не
тихо выключает фичу.
Тесты: registry unit (16), controller.attach (9), service pipe-options (4, вкл.
flag-off-with-runId негатив), int-spec ai-chat-attach (6, реальный MockLanguageModelV3).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- 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>
Integrates the designer-handoff Roles Catalog modal, wired to the real API; the
parent ai-agent-roles.tsx and the { opened, onClose, roles } contract are
unchanged.
- Server importFromCatalog now returns per-role lists (createdRoles /
skippedRoles with a reason) alongside the existing counters (compat-preserving),
so the UI can name the conflicting/installed roles.
- New pure view-model (catalog-bundle-model.ts): bundlePhase (empty | allNew |
allInstalled | updates | mixed, ignoring the transient 'skipped'),
installedLangForRole (same-slug-different-language hint), mapCatalogRoleToView —
all unit-tested without mounting.
- Bundle cards with a summary status in the collapsed header (eager useQueries
fan-out over all bundles, sharing the existing per-bundle cache keys), a single
primary action per bundle, checkboxes + select/deselect-all, an inline result
plaque that keeps the modal open, per-bundle and global 'Update all' request
series with progress, and the other-language hint.
- The partial-result plaque distinguishes the skip reason: only a name-conflict
offers 'Rename & install'; an already-installed race is informational (a rename
re-import would just skip again and self-heal into a false success).
- All strings i18n'd (en/ru); mock handoff code (SEED/mockImport/delay) removed.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Move every SERVER Markdown->ProseMirror path off the editor-ext markdown layer
(`markdownToHtml`, a second marked-based parser) onto the canonical
`@docmost/prosemirror-markdown` package, and add a foreign-markdown normalizer at
the import boundary.
Code:
- `ImportService.processMarkdown` (single `.md` upload) now parses
`markdownToProseMirror(normalizeForeignMarkdown(md))` directly — no HTML hop.
- `PageService.parseProsemirrorContent` markdown case (page create/update with
`format: 'markdown'`) same.
- `FileImportTaskService` (zip import) parses markdown with the package, then
serializes to HTML (`jsonToHtml`) so the SHARED HTML attachment / internal-link
pipeline (processAttachments + formatImportHtml + processHTML) keeps handling
`.md` and `.html` imports uniformly. The markdown PARSE — the drift source — no
longer goes through editor-ext; the PM->HTML->PM hop that follows is lossless
plumbing for attachment resolution, not a second parse.
- `canonicalizeFootnotes` stays as an idempotent #228 safety net for the HTML
path (a no-op on the already-canonical markdown output).
Normalizer (`integrations/import/utils/foreign-markdown.ts`): a TEXT pre-pass,
NOT a parser fork. The strict canonical parser does not accept GFM `[^id]`
reference footnotes (and would misread `[^id]: def` as a CommonMark link-ref
definition, silently corrupting the ref into a bogus link), so the normalizer
rewrites reference footnotes into canonical inline `^[def]` before parsing.
Callout surfaces (`:::type` and `> [!type]`) are intentionally NOT touched — the
canonical parser already accepts BOTH natively, so normalizing them would be
redundant and risk degrading its nesting/code-fence-aware handling.
Fixtures-first: foreign-markdown.spec pins the normalizer and the end-to-end
acceptance (no literal `[^id]`/`:::` leaks; re-export is canonical). The two
footnote-canonicalize specs are updated to the canonical output — the parser
assigns fresh `fn-*` ids, so they now assert by definition BODY order (still
reference-ordered, deduped, orphan-free).
FINAL CHECK: `grep -rn "htmlToMarkdown\|markdownToHtml" apps/server/src` (non
-test) is now empty — both editor-ext markdown-layer functions are gone from the
server.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- F1: added in-app execute tests for the two wirings that ACTUALLY changed in the
migration (the contract-parity test only checks advertised schema keys, not
execute bodies): movePage forwards the newly-added optional `position` to
client.movePage (and passes undefined position + null parent when omitted); the
table trio (insert/delete/updateCell) forwards the unified `table` param
positionally. A field destructured under the wrong name would have silently
passed undefined to the client (execute is any-cast, tsc won't catch it).
- F2: rewrote the three migrated descriptions that hardcoded snake_case sibling
tool names (which the in-app camelCase layer exposes under different ids,
violating the registry's own transport-neutral-prose convention) into neutral
prose: getPage "use get_page_json" -> "use the lossless page-JSON read tool";
updatePageJson "get_page_json -> ... -> update_page_json" -> "read the page-JSON
view -> modify -> write it back", "prefer rename_page" -> "prefer the rename-page
tool"; exportPageMarkdown "import_page_markdown round-trip" -> "page-Markdown
import round-trip" (the last was a direct regress — the in-app base said the
camelCase importPageMarkdown). (stashPage's pre-existing get_page_json mention is
out of scope, per the reviewer.)
Gate: mcp build 0; ai-chat-tools.service + tool-tiers (catalog-partition) pass,
incl. the 5 new execute-wiring tests.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Migrates share_page / sharePage into the transport-agnostic spec registry
(schema + description declared once; each transport keeps only its execute/auth):
- sharePage (deferred) -> SHARED_TOOL_SPECS; index.ts uses registerShared(),
ai-chat uses sharedTool(); removed from INLINE_TOOL_TIERS.
Drift reconciled (documented inline): both inline copies already carried the
"only share when the user explicitly asked" security framing, so the old
"per-transport divergence" note in BOTH layers was STALE — there was no real
behavioral divergence, only wording drift. The canonical description merges the
MCP copy's URL-format + idempotency detail with the in-app copy's reversibility
note and keeps the shared security framing. pageId keeps the MCP copy's stricter
.min(1). The MCP execute keeps its own `searchIndexing ?? true` default
(per-layer, not part of the shared schema).
Intentionally NOT migrated (kept inline — genuinely divergent, as their existing
notes state):
- search / searchPages: the in-app tool is a semantic+keyword hybrid (RRF) with
in-process access control and a tuned schema (limit 1-20); the MCP `search` is
a plain REST full-text search (limit up to 100). Different behavior AND schema.
- docmost_transform / transformPage: the in-app tool deliberately omits the
`deleteComments` schema field (a comment-deletion guardrail) and carries a
shorter description. Different schema.
Gate: mcp build 0 + node --test 458/458 (page-search excluded — hangs only under
the local re2->RegExp type-shim, its source untouched), server jest 775 incl.
tool-tiers catalog-partition + shared-spec contract parity, server tsc 0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Migrates the three-layer page tools into the transport-agnostic spec registry
(schema + description declared once; each transport keeps only its execute/auth):
- getPage, listPages (core), createPage, movePage, renamePage, deletePage,
updatePageJson, exportPageMarkdown (deferred) -> SHARED_TOOL_SPECS; index.ts
uses registerShared(), ai-chat uses sharedTool(); removed from
INLINE_TOOL_TIERS. Tiers preserved from CORE_TOOL_KEYS (getPage/listPages =
core, the rest deferred).
delete_page is genuinely three-layer (in-app deletePage exists), so it IS
migrated — not MCP-only. Its H4 guardrail is preserved: the shared schema
exposes ONLY pageId, so no permanentlyDelete/forceDelete flag can reach the
client (still asserted by ai-chat-tools.service.spec.ts).
Descriptions merged (documented inline): each canonical text takes the MCP
copy's richer structural notes plus the in-app copy's reversibility framing.
Schema DRIFT reconciled (documented inline):
- createPage.content: MCP pinned .min(1) but the in-app copy left it unbounded
and DOCUMENTS an empty body as valid ("may be empty" — creating an empty page
to fill later is a real use). Kept the looser no-min form: create_page now also
accepts an empty body (harmless) and no previously-valid in-app input is
rejected. title/spaceId keep the MCP .min(1) (empty is never valid).
- movePage: MCP exposed an optional `position` (fractional-index) field the
in-app copy lacked. Unified by KEEPING position — the in-app client already
accepts an optional position arg, so the in-app execute now forwards it;
optional, so no previously-valid call is rejected. `parentPageId` is nullable
on both (real JSON null -> root); the MCP execute keeps its 'null'/'' string
coercion as a per-layer robustness fallback.
- getPage/renamePage/updatePageJson/exportPageMarkdown/listPages: kept the MCP
copy's stricter .min(1) on ids where the in-app copy was unbounded.
Per-transport execute logic preserved: getPage's {title,markdown} projection,
updatePageJson's JSON-string normalization, list_pages' default limit/tree, and
move_page's cycle guard + positive-confirmation check all stay in their execute
bodies.
Intentionally NOT touched: updatePageContent (Markdown-based body update; no MCP
equivalent) and getTable (name-convention divergence, see tables family) stay
inline.
Gate: mcp build 0 + node --test 458/458 (page-search excluded — hangs only under
the local re2->RegExp type-shim, its source untouched), server jest 770 incl.
tool-tiers catalog-partition + shared-spec contract parity + deletePage H4
guardrail, server tsc 0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Migrates the three-layer table WRITE tools into the transport-agnostic spec
registry (schema + description declared once; each transport keeps only its
execute/auth):
- tableInsertRow, tableDeleteRow, tableUpdateCell -> SHARED_TOOL_SPECS;
index.ts uses registerShared(), ai-chat uses sharedTool(); removed from
INLINE_TOOL_TIERS (all three are deferred; not in CORE_TOOL_KEYS).
Drift reconciled (documented inline): the four table tools previously carried a
"NOT shared" note in both layers over a single parameter-NAME drift — the MCP
layer named the table reference `table`, the in-app layer `tableRef`. Unified on
the MCP name `table` (renaming the public MCP parameter would break external MCP
clients; the in-app parameter is model-facing/prompt-only and safe to rename).
The in-app execute bodies now destructure `table`. Descriptions took the MCP
copy's richer wording (documents `#<index>`, padding, header-row behavior) plus
the in-app copy's "Reversible via page history" note; both fields keep the MCP
copy's stricter .min(1) (in-app left them unbounded); sibling tool references
phrased transport-neutrally.
Intentionally NOT migrated (kept inline): table_get / getTable. Its MCP tool
name is noun-first (`table_get`) while the in-app key is verb-first (`getTable`),
which breaks the snake_case(inAppKey) naming convention the registry enforces
(shared-tool-specs.contract.spec.ts). Renaming the public MCP tool would break
external clients, so it stays per-transport — but its in-app reference param was
still aligned to `table` (was `tableRef`) for consistency with the migrated trio.
Gate: mcp tsc 0 + node --test 458/458 (page-search excluded — hangs only under
the local re2->RegExp type-shim, its source is untouched), server jest 730 incl.
tool-tiers catalog-partition + shared-spec contract parity, server tsc 0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Migrates the three-layer comment tools into the single transport-agnostic spec
registry (schema + model-facing description declared once; each transport keeps
only its execute/auth):
- createComment, listComments, resolveComment, checkNewComments — moved to
SHARED_TOOL_SPECS; index.ts uses registerShared(), ai-chat uses sharedTool();
removed from INLINE_TOOL_TIERS (tier/catalogLine now on the spec). Tiers
preserved from CORE_TOOL_KEYS (create/list/resolve = core, check = deferred).
Intentionally NOT migrated (kept MCP-inline): update_comment / delete_comment —
they are MCP-only by design; the in-app AI-chat layer deliberately has no
updateComment/deleteComment (comment edits are irreversible / not
version-tracked), asserted by ai-chat-tools.service.spec.ts. A registry spec's
tier/catalogLine are in-app metadata and the catalog-partition test forbids a
deferred spec without a live in-app tool, so these stay per-transport.
Drift reconciled (documented inline): createComment/listComments/checkNewComments
took the more-maintained/superset description + stricter .min(1) guards.
resolveComment: `resolved` drifted (MCP optional+default(true) vs in-app
required) — kept the MCP superset, so in-app resolveComment now accepts an
omitted `resolved` (defaults to resolve) — a deliberate, backward-compatible
unification (never rejects a previously-valid input).
Gate: mcp build 0 + node --test 480/480, ai-chat 654, tool-tiers (incl. F3
catalog-partition) 16/16, server tsc 0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- F10 [stability]: closeMetricsServer() now calls server.closeIdleConnections()
+ server.unref() after server.close(). server.close()'s callback doesn't fire
until keep-alive sockets drain, and the scraper (VictoriaMetrics/vmagent) holds
an idle keep-alive socket — so onModuleDestroy's awaited close would hang until
the scraper disconnects or the orchestrator SIGKILLs on the kill-grace window.
closeIdleConnections() drops idle keep-alive sockets so shutdown completes
immediately (Node 22, per the Dockerfile base).
- F9 [test]: client-telemetry.module.spec.ts pins the E1=B register() gate — the
core of the "public endpoint OFF by default" decision: flag unset / any non-
"true" value ("false"/""/"0"/…) → empty controllers+providers (route absent);
"true"/"TRUE" → registers VitalsController + VitalsService. A flag-inversion or
truthiness regression that reopened the anonymous disk-fill surface now fails.
- F11 [regression/perf]: the db_query_duration_seconds token work (firstSqlToken
regex + Set lookup) is now gated on isMetricsEnabled() in database.module.ts, so
a non-metrics deployment pays NOTHING per query (previously observeDbQuery
no-op'd but the token was still computed on every query). Also hoisted the
13-element known-token Set to a module const (KNOWN_SQL_TOKENS) so it's built
once, not per query.
Gate: server tsc 0; metrics + vitals + client-telemetry suites pass (incl. the
new register-gate test).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Maintainer resolved E1 as variant B: the public vitals sink + client collection
must be OFF by default (else client_metrics grows unbounded on a self-host deploy
with no external pruner, via an unauthenticated public endpoint).
- F1: new operator flag CLIENT_TELEMETRY_ENABLED (default OFF), SEPARATE from
METRICS_PORT (Grafana reads the table directly, independent of the scrape port).
ClientTelemetryModule.register() provides VitalsController ONLY when the flag is
true (route absent otherwise); the flag reaches the client via window.CONFIG
(config.ts isClientTelemetryEnabled), and initVitals() early-returns when off.
- F2/F3 [throttler]: this repo's ThrottlerGuard applies EVERY named throttler to
every guarded route unless skipped. The new VITALS bucket therefore (a) newly
bound collab-token → 429 behind shared/NAT IPs, and (b) the vitals route didn't
skip the stricter public-share-ai (5/min) bucket → effective 5/min not 120.
Fix (additive, global config unchanged): vitals.controller @SkipThrottle the
other buckets + @Throttle VITALS 120/min; collab-token adds VITALS_THROTTLER to
its existing @SkipThrottle (restoring its prior effectively-unthrottled state).
- F4: metrics node:http server is closed on shutdown (MetricsServerLifecycle
OnModuleDestroy → closeMetricsServer(), fired by enableShutdownHooks).
- F5: docSize outside [0, int4-max] drops to null (keeping the event) instead of
overflowing int4 and failing the WHOLE batch insert (+ 2 tests).
- F6: .env.example documents METRICS_PORT (no default — unset = subsystem OFF) +
CLIENT_TELEMETRY_ENABLED; fixed the inaccurate "default 9464" wording.
- F7: disabled/non-sampled sessions install ZERO observers — isVitalsActive()
(enabled && sampled) gates reportClientMetric AND the page-editor
measurePageOpen + dispatchTransaction wrapping.
- F8: kept db.d.ts hand-added (wontfix) — this repo HAND-CURATES db.d.ts (verified
across recent fork migrations a32fba63/8c5b57eb/fdeede00); codegen would be the
deviation. The ClientMetrics interface maps the migration 1:1.
Gate: server tsc 0, client tsc 0, server metrics/vitals/telemetry/throttle 21
tests, client route-template 5. No new deps.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Squashed for a clean rebase onto develop (was 19 commits; the reviewer approved
the net diff at fb246080). Detaches an agent run from the HTTP request/browser
window: a run is a first-class lifecycle object (ai_chat_runs), a browser
disconnect no longer kills it, a concurrent-run insert-gate prevents double runs,
and a reopened chat live-follows a still-running run via a polled observer merge.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The metrics INFRA is already deployed (VictoriaMetrics scraping docmost:9464,
Grafana dashboards, alerts) with a target `gitmost-app` that is red because the
app half didn't exist. This is that half. The contract (metric names, port,
table, endpoint) is FIXED by the deployed infra and matched exactly.
Server (prom-client):
- A bare node:http `/metrics` server on METRICS_PORT (default 9464), SEPARATE
from the Fastify :3000 listener so /metrics never exists publicly; the whole
subsystem is OFF when METRICS_PORT is unset.
- collectDefaultMetrics() + http_request_duration_seconds{method,route,status}
via a Fastify onResponse hook using the ROUTE TEMPLATE (req.routeOptions.url,
never the raw URL — bounded cardinality; 404 -> "unknown"), EXCLUDING SSE/
streaming responses (would record the connection lifetime and poison p95).
- db_query_duration_seconds (Kysely log callback, labelled by the leading SQL
token), bullmq_queue_depth{queue} (getJobCounts every 15s) +
bullmq_job_duration_seconds{queue} (worker completed/failed),
collab_store_duration_seconds (around onStoreDocument).
- POST /api/telemetry/vitals — PUBLIC (sendBeacon) but IP-throttled; ~16KB body
cap, <=50 events/batch, metric-name + rating whitelist, attr truncated to 120
chars, batch insert; malformed/foreign/oversized silently dropped and 200'd (no
browser retry). New migration `client_metrics` (schema byte-identical to the
contract, both indexes, conditional grafana_ro GRANT; no app-side retention —
the maintenance container prunes >90d).
Client (web-vitals):
- initVitals() decides sampling ONCE per session (25%, sessionStorage) BEFORE
subscribing; onINP/onLCP/onCLS/onTTFB (attribution) buffered + flushed via
navigator.sendBeacon on visibilitychange:hidden and a timer (not fetch-per-
metric). Custom: editor_tx_ms (dispatchTransaction sync-part timer, >8ms, with
doc_size), page_open_ms, longtask_ms. Route labels are templates only; no
titles/slugs/text.
Gate: server + client tsc 0, frozen install 0 (added prom-client + web-vitals +
regenerated the lock), server metrics/vitals tests 11, client route-template 5,
and the migration verified valid against real Postgres.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- F1: document AI_CHAT_DEFERRED_TOOLS in .env.example (AI_* section) — default
ON = deferred loading (compact catalog + loadTools), =false restores the old
"all tools always active" behavior.
- F2: integration test of the ON path in ai-chat-stream.int-spec.ts — a deferred
tool activated via loadTools is active on the SAME turn's next step but a fresh
turn starts cold (CORE + loadTools only), proving the per-turn activatedTools
Set does not leak across turns/chats. Drives the real streamText loop with a
MockLanguageModelV3 and inspects recorded per-step activeTools-filtered tools.
- F3: replace the magic toHaveLength(28) in tool-tiers.spec.ts with a two-way
partition against the LIVE in-app toolset (AiChatToolsService.forUser keys):
every non-core tool must appear in buildInAppDeferredCatalog and every catalog
entry must map to a real non-core tool — so a future tool forgotten in
INLINE_TOOL_TIERS fails the suite instead of silently vanishing from the agent.
No production logic change (mechanism was already reviewed correct).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The in-app AI agent shipped all ~41 tool schemas on every model step. This
adds a two-tier catalog: core tools (frequent or one-line) stay always-active;
the rest are advertised as a compact catalog and their full schema is fetched
on demand via the loadTools meta-tool, wired through ai@6 prepareStep's
per-step activeTools.
- tools/tool-tiers.ts: CORE_TOOL_KEYS, INLINE_TOOL_TIERS, applyLoadTools,
catalog builders (+ tool-tiers.spec.ts, 13 cases).
- ai-chat.service.ts prepareAgentStep: returns activeTools =
[...CORE_TOOL_KEYS, loadTools, ...activatedTools]; per-turn activated Set.
- ai-chat.prompt.ts: buildToolCatalogBlock renders the deferred catalog.
- mcp/tool-specs.ts: tier + catalogLine metadata (external snake_case /mcp
transport unchanged).
- EnvironmentService.isAiChatDeferredToolsEnabled(): AI_CHAT_DEFERRED_TOOLS,
default ON per issue intent (kill-switch =false restores old behavior).
Gate: server ai-chat 631/631, tool-tiers 13/13, mcp 472/472, tsc clean.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
agent_coder
vvzvlad
agent_vscode
agent_coder
claude code agent 227