Commit Graph

1040 Commits

Author SHA1 Message Date
claude code agent 227
c0d312d8f5 harden(transclusion): depth-cap the ProseMirror collectors (#55)
collectPageEmbedsFromPmJson (and the sibling collectors/remap) recursed with no
guard, so a pathological/cyclic non-JSON input could stack-overflow (RangeError).
Add a depth cap (1000, far above any real doc nesting) so such input is handled
gracefully. Normal documents are unaffected. Updates a stale test that asserted
the old throwing behavior.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-21 03:17:37 +03:00
claude code agent 227
5215913533 fix(security): env-configurable trustProxy with a safe default (#61)
trustProxy was unconditionally true, so req.ip came from a client-forgeable
X-Forwarded-For and the per-IP throttles (share-AI, /mcp brute-force) were
spoofable. Make it env-configurable (TRUST_PROXY) with a safe default that
trusts XFF only from loopback/private proxies, documented in .env.example.
NOTE: this changes the default from trust-all; deployments whose proxy is on a
public IP must set TRUST_PROXY (caveat documented).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-21 03:17:37 +03:00
claude code agent 227
e52f069fc6 fix(ws): resync the sidebar tree on socket reconnect (#66)
WS events missed during a disconnect (wifi blip, sleep) were lost, so the
sidebar tree silently diverged until a manual reload. On RECONNECT (not the
first connect) invalidate the root-sidebar-pages + sidebar-pages queries so the
tree refetches through the authorized API and re-converges.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-21 03:17:37 +03:00
claude code agent 227
ff342ca705 cleanup(page-embed): remove dead isPageEmbedCycle/isPageEmbedTooDeep (#71)
After a merge decideEmbedState became the canonical guard and inlines the
cycle/too-deep logic, leaving these predicates called only by their own tests.
Remove them (and their test blocks); keep PAGE_EMBED_MAX_DEPTH (used by
decideEmbedState). Production behavior stays covered by decide-embed-state.test.ts.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-21 03:17:37 +03:00
claude code agent 227
099d31f594 fix(ai): sandwich SAFETY_FRAMEWORK around the role persona (#68)
A custom AI-role's text preceded the only SAFETY_FRAMEWORK block and replaced
the persona, so a jailbreak in the role text sat before the safety rules.
buildSystemPrompt now emits SAFETY both before AND after the persona, with the
role/persona delimited as lower-trust (<role_persona note=...>); the default
persona is sandwiched too. Context (currently-viewing-page) preserved.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-21 03:17:37 +03:00
claude code agent 227
212bcea4d7 fix(page): movePage cycle guard + no phantom PAGE_MOVED (#67, #64)
#67: movePage didn't check the destination wasn't the page itself or inside its
own subtree, so MCP/REST/agent/fast-drag could persist+broadcast a cycle. Reject
before the update (self-parent, or moved page among the destination parent's
ancestors via getPageBreadCrumbs).
#64: movePage emitted PAGE_MOVED from a stale pre-read even when the row didn't
change / was concurrently deleted (phantom move). Gate the emit on
updateResult.numUpdatedRows !== 0n. Both are movePage hardening in one method.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-21 03:08:34 +03:00
claude code agent 227
05a7a4001f fix(share-ai): cap per-request output + unify provider errors (#60, #95)
#60: streamText had no maxOutputTokens, so one anonymous request could run up
the provider bill. Add maxOutputTokens (env SHARE_AI_MAX_OUTPUT_TOKENS, default
512) via resolveShareAiMaxOutputTokens().
#95: the anonymous path hand-built error strings, diverging from the unified
describeProviderError format used on the authenticated path; both onError blocks
now call describeProviderError so a share reader sees 402/429/503 causes in the
same form (and the stack is still logged). Both changes are in this one file and
share hunks, hence one commit.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-21 03:08:34 +03:00
claude code agent 227
5344a9bdde fix(auth): handle null-password (SSO/LDAP-only) accounts without bcrypt throw (#70)
A user with password=NULL passed the missing/disabled guard and reached
comparePasswordHash(pw, null), which native bcrypt rejects -> 500 on
/api/auth/login and, on /mcp, a leaky 401 that the brute-force limiter ignored
(enumeration oracle + limiter evasion). Treat a null/empty password like a
missing user in verifyUserCredentials (dummy compare for timing parity + unified
CREDENTIALS_MISMATCH_MESSAGE) and reject early in changePassword before bcrypt.
Contract spec asserts the null-password guard.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-21 03:07:53 +03:00
claude code agent 227
d79f709742 fix(share): surface the real error on the share AI widget (#87)
The widget hardcoded a generic 'Something went wrong' body and ignored
error.message, violating AGENTS.md. Render describeChatError(error.message, t) —
the same helper the internal chat uses — so a reader sees the real 402/429/503
cause instead of a bare 'try again'.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-21 03:07:53 +03:00
claude code agent 227
2b4ec0bfcc fix(share-ai): reject non-text message parts to close size-cap bypass (#63)
MAX_SHARE_MESSAGE_CHARS only counted text parts, so a forged non-text part
(tool-result/file/data) bypassed the cap and bloated the model input
(token-DoS); convertToModelMessages would also expand a forged tool-result. The
anonymous path runs no tools, so a client non-text part is never legitimate —
reject any message with a non-text part (isTextUIPart) before the size check.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-21 03:07:53 +03:00
claude code agent 227
e19849d980 fix(share-ai): fail-closed workspace limiter on Redis failure (#62)
The per-workspace anonymous share-AI cost cap failed OPEN on a Redis error
(return true => admit), so a Redis outage removed the cap entirely (unmetered
billable anonymous calls). The feature is optional, so unavailability is
harmless: fail CLOSED (return false => controller 429s) instead.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-21 03:07:53 +03:00
claude_code
4bf6d9f36b Merge develop into fix/html-embed-hardening (#46)
Some checks failed
Test / test (pull_request) Has been cancelled
Resolve the html-embed.spec.ts conflict as a union: both #46 and #49 (already in
develop) added different test cases to the same file. Keep all of them —
stripHtmlEmbedNodes gets #46's root-node case plus develop's deeply-nested,
non-object and empty-content cases; #46's collectHtmlEmbedSources and
stripDisallowedHtmlEmbedNodes suites and develop's hasHtmlEmbedNode suite all
kept; imports unioned. No production code conflicted.

Full suite green: server 651, client (16 files), editor-ext 56, mcp 247.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-21 01:59:22 +03:00
claude_code
d7681b4fb6 Merge develop into fix/page-template-demo-issues (#45)
Some checks failed
Test / test (pull_request) Has been cancelled
Resolve conflicts from the parallel page-embed refactor that landed in develop
via #49:
- page-embed-view.tsx: keep develop's canonical decideEmbedState for the
  cycle/depth/availability guard; keep #45's #39 chrome cleanup (single source
  link, IconFileText fallback) and #40 refresh remount key. Drop #45's now-unused
  isPageEmbedCycle/isPageEmbedTooDeep wiring.
- page-embed-picker.tsx: use develop's excludeHost util; drop #45's duplicate
  filterPageEmbedOptions and its test.
- page-embed-ancestry-context.test.tsx: keep #45's superset suite.
- page-template-access.spec.ts: keep develop's constructor args; update the two
  deleteByReferenceAndSources assertions to the new 4-arg workspace-scoped
  signature introduced by #45 (#36 defense-in-depth).

Full suite green: server 624, client 219, editor-ext 56, mcp 247.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-21 01:51:09 +03:00
claude_code
8b8b05e005 Merge remote-tracking branch 'gitea/develop' into feat/ai-chat-collapse-on-focus 2026-06-21 01:33:47 +03:00
claude_code
3695dbdf7f Merge remote-tracking branch 'gitea/develop' into fix/ai-chat-current-page 2026-06-21 01:29:37 +03:00
claude_code
4fa8882c58 Merge remote-tracking branch 'gitea/develop' into feat/share-chat-reuse-internal 2026-06-21 01:28:14 +03:00
claude_code
730486ad12 test(mcp): keep real mcp-auth.helpers in gate spec mock (forward-compat with #49)
Some checks failed
Test / test (pull_request) Has been cancelled
After develop merged, mcp.service.ts calls decideBasicGate from mcp-auth.helpers.
The gate spec mocked the whole module returning only FailedLoginLimiter, so the
merged code crashed with 'decideBasicGate is not a function' (7/7 failing).
Spread jest.requireActual('./mcp-auth.helpers') so the real helpers are kept and
the gate exercises real logic; keep only FailedLoginLimiter stubbed so its
constructor runs without a real sweep timer.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-21 01:25:36 +03:00
claude_code
5f3a3d3ec0 Merge remote-tracking branch 'gitea/develop' into fix/mcp-security-followups 2026-06-21 01:21:57 +03:00
claude_code
f63719a21c fix(share): neutralize own-origin absolute links in public-share AI chat
isExternalHttpUrl treated any http(s):// URL as external, so an absolute link
back to the app's own host (e.g. https://self/p/{uuid}, /settings/members)
emitted by the assistant stayed clickable on the anonymous share, leaking
internal UUIDs/structure and pointing at auth-gated routes. Classify a link as
external only when its host differs from window.location.host; unparseable URLs
are treated as internal (fail-closed). Tests cover own-origin absolute (flag
on -> inert), external host (kept with safe rel/target), dangerous schemes, and
no behavior change for the internal chat (flag off).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-21 01:20:11 +03:00
claude_code
d96f94a80a ci: run the test suites on push/PR + quarantine broken stock scaffolds
Some checks failed
Test / test (pull_request) Has been cancelled
Add .github/workflows/test.yml (pnpm + Node 22): on pull_request and push
to develop it installs, builds @docmost/editor-ext and runs `pnpm -r test`
across all packages (server Jest, client Vitest, editor-ext Vitest,
packages/mcp node:test). So tests now run automatically in CI, not just
on demand.

To make the run green, quarantine the 16 pre-existing stock NestJS
`should be defined` scaffold specs via jest `testPathIgnorePatterns` —
they never compiled (missing DI providers / lib0 ESM) and assert nothing
useful. Tracked for a proper fix/removal in issue #56. Verified each
pattern drops only its scaffold (46 of 62 suites still collected) and the
full `pnpm -r test` is green: server 587, client 185, editor-ext 56,
mcp 247.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-21 00:44:21 +03:00
claude code agent 227
e0aac5aa04 feat(share): public-share AI chat reuses the internal chat's presentation (#41)
The public-share widget was a separate minimal impl: plain-text answer, static
'Thinking…', no markdown, no tool-cards. Now it renders through the internal
chat's debugged presentational layer (MessageList/MessageItem/TypingIndicator/
ToolCallCard), so a share gets the same incremental streaming, animated typing
indicator, markdown, and tool-call cards. The share keeps its anonymous
transport (useChat + DefaultChatTransport '/api/shares/ai/stream',
credentials:'omit').

The shared components were already prop-driven (UIMessage[] + isStreaming) with
no transport/auth coupling; made the new props additive optionals (emptyState,
showCitations, neutralizeInternalLinks) all defaulting to current behavior, so
the internal chat is unchanged.

Security (review-caught): rendering assistant markdown on the ANONYMOUS share
made internal links (/p/{id}, /settings/...) clickable, which the old plain-text
render didn't. renderChatMarkdown gains neutralizeInternalLinks (true only on
the share): a one-shot DOMPurify afterSanitizeAttributes hook (added/removed by
reference around a single sanitize) strips href from internal/relative/non-http(s)
links (rendered inert) and keeps external http(s) links with
rel=noopener noreferrer nofollow target=_blank. Tests cover both the link
neutralization and the absence of any global-hook leak into internal renders.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-21 00:04:18 +03:00
claude code agent 227
f6e216cb87 feat(ai-chat): auto-collapse the chat window on page focus, expand on header (#42)
The floating chat window covered page content; you could only collapse it
manually. Now it auto-collapses to its header (visual collapse only — ChatThread
stays mounted so an in-flight stream isn't interrupted) when you interact with
the page, and expands again from the header.

- document mousedown listener in the CAPTURE phase, armed only when
  windowOpen && !minimized; collapses on a pointer-down outside the window.
  Guards: ignore clicks inside the window and inside any Mantine [data-portal]
  (the chat-list kebab menu + delete-confirm modal render in portals).
- Header click expands: startDrag distinguishes click vs drag by a 4px
  threshold (minimizedRef avoids a stale closure); an expand-click doesn't
  persist geometry.
- Reset minimized=false when the window opens (no sticky collapsed state).
- a11y: when minimized, the title is the keyboard expand affordance
  (role=button, tabIndex, aria-label Expand, Enter/Space) — kept off the
  dragBar container so no role=button wraps the Minimize/Close buttons.
- Pure helpers shouldCollapseOnOutsidePointer + isHeaderClick with vitest tests.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-20 23:45:43 +03:00
claude_code
90d3fab483 test: cover features since 053a9c0d + repair test tooling
Add ~330 tests across server (Jest), client (Vitest), editor-ext (Vitest)
and packages/mcp (node:test) for the gitmost features added since
053a9c0d: AI chat, AI agent roles, public-share assistant, MCP per-user
auth, HTML embed, page templates/embed, realtime tree, tree
expand/collapse, and the AI-settings UI.

Test-tooling fixes (prerequisite, were silently hiding coverage):
- Repair 3 page-template specs broken by the 11-arg TransclusionService
  constructor; they never compiled, so template access-control / content
  -leak / unsync-strip coverage was fictitious.
- Build @docmost/editor-ext before server tests via a `pretest` hook;
  the stale dist omitted the new HtmlEmbed/PageEmbed exports (TS2305).
- Let jest resolve the .tsx email templates: add `tsx` to
  moduleFileExtensions and widen the ts-jest transform to (t|j)sx?.

Behaviour-preserving "extract pure core" refactors that the tests drive:
- server: resolveShareAssistantRequest + uiMessageTextLength
  (public-share controller), decideBasicGate + mapAuthResultToResponse
  (mcp), buildErrorAssistantRecord (ai-chat), jsonbObject export (roles).
- client: render-raw-html + shouldExecute/canEdit, decide-embed-state,
  page-embed picker utils, tree-socket reducers, open/close branch maps,
  isEndpointConfigured/resolveKeyField; buildTreeWithChildren now treats
  a permission-trimmed orphan as a root instead of crashing.

Deferred (need a test DB or HTTP harness, documented in the specs):
repo-level Postgres integration tests and the public-share XFF E2E.
Pre-existing DI/lib0-ESM suite failures are untouched and out of scope.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-20 23:40:40 +03:00
claude code agent 227
1f457b060c fix(mcp): security review follow-ups (#24)
Post-merge hardening from the #13 security review:
- isInitializeRequestBody now delegates to the SDK isInitializeRequest (same
  predicate as packages/mcp/http.ts), so a bare {method:'initialize'} with no
  id/params no longer triggers the side-effecting login() (audit-spam /
  user_sessions growth) before http.ts 400s it.
- Bind the Bearer path to the instance workspace: verifyBearerAccess rejects a
  token whose payload.workspaceId != the instance workspace (resolved via
  workspaceRepo.findFirst, consistent with the Basic path); optional param so
  it's a no-op when unset.
- Close the user-enumeration timing oracle in verifyUserCredentials: the
  missing/disabled branch now runs a bcrypt compare against a module-level dummy
  hash whose cost (12) matches production saltRounds, so both paths take one
  equal-cost bcrypt compare; the exact CREDENTIALS_MISMATCH_MESSAGE is preserved.
- Document the trusted-proxy requirement for the spoofable per-IP brute-force
  limiter in .env.example (trustProxy is on; deploy behind a trusted proxy).
- Add real-execution coverage for enforceBasicLoginGate (SSO enforced / EE-MFA
  bundled vs not / user-MFA / workspace-enforced-MFA) instead of stubbing the gate.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-20 23:36:53 +03:00
claude code agent 227
424761753e fix(html-embed): shrink the collab broadcast window with an early onChange guard (#26)
A non-admin's transient htmlEmbed could execute in other open editors until the
debounced (10s) onStoreDocument strip. Add a ~300ms onChange-debounced early
strip (guardHtmlEmbed) that converges the shared ydoc for everyone far sooner.

Safety-critical details:
- Scheduled from onChange ONLY for non-admins AND only when the workspace toggle
  is ON (cached per-document in onLoadDocument), so the common toggle-OFF case
  does zero extra work.
- guardHtmlEmbed does ALL async work (toggle + persisted allow-list read) FIRST,
  then performs fromYdoc -> strip -> fragment.delete -> applyUpdate in a single
  SYNCHRONOUS, await-free block, so no inbound Yjs update can interleave and a
  concurrent edit can never be clobbered. Bails if document.isDestroyed.
- Reuses the #29 preserve logic (admin-vetted embeds survive; only the non-admin's
  new ones are stripped). Loop-safe (corrective update has null origin -> no
  reschedule; post-strip no embed -> cheap no-op). Per-document timer cleared on
  unload. onStoreDocument stays the authoritative backstop.

The irreducible residual is only the very first inbound broadcast before the
debounce fires — Hocuspocus exposes no synchronous beforeBroadcast filter.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-20 23:20:02 +03:00
claude code agent 227
b7ea8c850e fix(html-embed): preserve admin's existing embed on a non-admin co-editor's store (#29)
The collab persist strip keyed to the storing connection's user, so when a
non-admin co-editor stored, it removed an admin's legitimately-authored embed
too (data loss). Now: toggle OFF still strips all (feature disabled); toggle ON
+ non-admin storer strips only NEWLY-introduced embeds and preserves those
already present in the persisted content (admin-vetted), via new helpers
collectHtmlEmbedSources + stripDisallowedHtmlEmbedNodes (identity = attrs.source,
already-vetted HTML). The ydoc reflect is now guarded by a deep-equal check so
an unrelated non-admin edit that touches no new embed doesn't churn the doc.
A non-admin still cannot add a new embed. Documents the allow-list TOCTOU
(best-effort snapshot read outside the lock; converges on next store).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-20 23:02:01 +03:00
claude code agent 227
8191c37daa test(html-embed): real-execution gate tests for create/duplicate/import (#27)
The create/duplicate/import gate tests asserted gate presence via brittle
expect(SRC).toMatch(/regex/) over the source text plus a reimplemented
applyGate() stand-in, so a refactor could break the real gate while they still
passed. Rewrite both specs to execute the REAL methods (PageService.create /
duplicatePage; ImportService.importPage; FileImportTaskService.processGenericImport)
with each caller role and assert on the PERSISTED content via hasHtmlEmbedNode:
member -> stripped, admin/owner+toggle ON -> preserved, toggle OFF -> stripped
for everyone, unknown/missing role -> fail-closed. No source-regex assertions
remain.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-20 22:49:18 +03:00
claude code agent 227
39f3eacf89 test(page-templates): cover lookupTemplate anti-leak + edge cases (#33)
- The security-relevant catch->not_found branch in lookupTemplate (returns
  not_found instead of raw content when comment-mark stripping throws) is now
  tested by forcing the strip to throw with a malformed text node, asserting no
  content/marks leak.
- not_found for a soft-deleted source resolved through the REAL
  filterViewerAccessiblePageIds (deletedAt-excluded), not the stubbed filter.
- Rename the misleading 'honours <=50 cap' test to reflect it only exercises
  dedup (the cap lives in the DTO, never engaged in the service unit).
- Cover the onlyTemplates search filter (restricts to is_template=true).
Also fix two pre-existing failing 'should be defined' specs (search service +
controller) that couldn't resolve the @InjectKysely token via createTestingModule.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-20 22:37:35 +03:00
claude code agent 227
bc1ea792f5 test(page-templates): cover duplicatePage pageEmbed remap + reference sync (#32)
Extract the per-node pageEmbed remap decision into a shared pure helper
(remapPageEmbedSourceId) and use it BOTH in PageService.duplicatePage and the
JSON walker, so the test guards the real production path (not a mirror that
could drift). Behavior is identical: source in the copied set -> new copy id;
otherwise keep the original. Add jest coverage (16 tests): the remap helper
(in-set/out-of-set/null/nested), syncPageTemplateReferences toDelete (stale refs
removed with the right workspaceId), and insertTemplateReferencesForPages
multi-workspace grouping/filtering.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-20 22:37:35 +03:00
claude code agent 227
98769155d3 test(page-templates): cover client pageEmbed cycle/self-embed/depth guard (#31)
The cycle/self-embed/depth guard (PAGE_EMBED_MAX_DEPTH=5) lives only on the
client and is the sole protection against runaway nested rendering — and was
untested. Extract the inline predicates into pure, behavior-identical exported
helpers (isPageEmbedCycle, isPageEmbedTooDeep in the ancestry context;
filterPageEmbedOptions in the picker) so they're unit-testable without mounting
the heavy Tiptap NodeView, and add vitest coverage (20 tests): ancestry chain/
host accumulation, cycle (ancestor-in-chain + top-level self-embed), too-deep at
the cap, and picker host-exclusion.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-20 22:37:35 +03:00
claude code agent 227
4f46f91db4 test(page-templates): fix TransclusionService spec constructor arity
The transclusion specs predated two added constructor params, so they failed to
compile (TS2554: expected 11 args, got 10) and the suites couldn't run. Add the
missing mock args: workspaceRepo (param 11) in the lookup/access specs, and
pageTemplateReferencesRepo (param 4, which had shifted pageRepo into the wrong
slot) in the unsync-html-embed spec. All three suites now compile and pass.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-20 22:22:56 +03:00
claude_code
c5f44a6eee Merge branch 'develop' into feat/footnotes
Resolve conflicts at shared registration points by unioning both features
(footnotes + the already-merged html-embed / page-embed work):
- slash-menu/menu-items.ts, editor extensions.ts: keep both imports + configures
- collaboration.util.ts: register footnote nodes and pageEmbed
- editor-ext marked.utils.ts: register footnote + html-embed markdown extensions
- editor-ext package.json/tsconfig.json/vitest.config.ts: union of test config
  (jsdom env for footnote DOM tests + combined test/spec include glob)

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-20 22:21:07 +03:00
claude code agent 227
a6ba19f0dc feat(ai-chat): add get_current_page tool for proxy-robust page context (#43, hardness #2)
The current page id was only injected as text in the system prompt, which a
proxy (CLIProxyAPI) can rewrite/truncate, so the agent could lose track of 'this
page'. Add a getCurrentPage tool the model can call to read the open page (id +
title) from the server-side request context (forUser now takes openedPage,
threaded from body.openPage — the same value used for the system prompt). The
inline system-prompt line is kept as belt-and-suspenders. Reads/writes still go
through the CASL-enforced page tools by id, so this is strictly not worse than
the existing prompt hint — just delivered over a channel the proxy can't mangle.

User-approved on the issue. Completes #43 together with the hardness-1 fix.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-20 22:19:40 +03:00
claude code agent 227
ada1dce739 fix(ai-chat): resolve the current page for agent context (#43, hardness #1)
AiChatWindow derived the open page via useParams(), but it's mounted in a
pathless parent layout route where :pageSlug isn't matched, so useParams()
returned {} and openPage was ALWAYS null — the agent never received current-page
context (couldn't resolve 'this page'/'the current page'). Derive pageSlug from
useMatch('/s/:spaceSlug/p/:pageSlug') against the full pathname instead, so it
resolves regardless of where the component sits in the route tree. No-match
behavior is unchanged (undefined -> query disabled -> openPage null).

Addresses Hardness #1 of #43. Hardness #2 (proxy resilience: a get_current_page
tool / hidden user-message context so identity doesn't depend on the system
prompt surviving CLIProxyAPI) remains open.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-20 21:57:03 +03:00
claude code agent 227
8ee4279d30 harden(html-embed): make stripHtmlEmbedNodes total with a root-type check (#30)
stripHtmlEmbedNodes only filtered children, so a (never-in-practice) bare
htmlEmbed root node would be returned as-is. Add a defensive root check that
returns an embed-free doc, making the helper total — it can never return a node
for which hasHtmlEmbedNode is true. Adds a unit test for the root case.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-20 21:52:32 +03:00
claude code agent 227
6a052b88b4 fix(html-embed): strip embeds at serve time on authenticated read paths (#28)
Completes the workspace htmlEmbed kill-switch. The public-share path already
strips at serve time when the toggle is OFF, but the authenticated read paths
(/info and /history/info) returned page/history content with embeds intact, so
a disabled feature kept executing for in-workspace view-only viewers until the
page was next saved. Now both paths resolve the workspace toggle and run
stripHtmlEmbedNodes when it's OFF (fail-closed on a missing workspace), before
any markdown/html format conversion. Admin-authored content only — completeness,
not privilege escalation. Injects WorkspaceRepo into PageController.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-20 21:52:32 +03:00
claude code agent 227
79d096ed7a fix(page-templates): defense-in-depth workspace checks (#36)
Consistency hardening from #17 review (not currently exploitable):
- toggleTemplate now explicitly rejects a page outside the caller's workspace
  (page.workspaceId !== user.workspaceId -> NotFound, avoiding existence leak)
  instead of relying solely on the space-membership model.
- PageTemplateReferencesRepo.deleteByReferenceAndSources is now workspace-scoped
  (adds a workspaceId filter + param), matching the 'scope by workspaceId
  everywhere' invariant; the sole caller threads its workspaceId.
The PAGE_TEMPLATE_THROTTLER limit is intentionally left as-is (the issue's
throttle item was 'consider only'; no change without usage data).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-20 21:42:49 +03:00
claude code agent 227
a15cccf557 chore(page-templates): remove dead findReferencePageIdsBySource (#34)
The 'used in N pages' reverse-navigation method had zero callers in the merged
PR #17 — unreachable, untested code. Remove it. The reverse-navigation feature
can be (re)added with the method if/when it's actually built.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-20 21:42:49 +03:00
claude code agent 227
22887c474a chore(page-templates): tidy ts suppression in duplicatePage pageEmbed remap (#37)
Replace bare //@ts-ignore (no space, no reason) with // @ts-expect-error plus a
reason on the pageEmbed sourcePageId reassignment, matching the codebase style.
ProseMirror Attrs is read-only typed, so the reassignment genuinely errors —
@ts-expect-error is valid here.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-20 21:42:30 +03:00
claude code agent 227
4536d27ad2 fix(page-templates): never strand a page-embed id in-flight (#35)
In the page-embed lookup flush(), the success branch cleared inFlightRef and
resolved waiters only for ids present in the response items. A short/partial
server response would leave a requested id stuck in inFlightRef forever (the
subscribe/refresh path is guarded by !inFlightRef.has(id)) and its refresh()
promise would never resolve. After processing returned items, also clear +
resolve any requested id that wasn't returned, mirroring the catch branch.
Cannot trigger under today's exact-mapping server contract; this is hardening.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-20 21:42:30 +03:00
claude code agent 227
a85dd607bd fix(footnotes): tighten the gap between a definition's number and text (#44)
The footnote definition number ('1.') sat ~19px from its text because two
spacings stacked: the 1.5em (24px) marker min-width box (wider than the ~15px
glyph) plus a 10px flex gap. Reduce the flex gap to 0.4em (about one space) and
right-align the number within the 1.5em column so the period sits next to the
text and multi-digit numbers (10, 11, ...) stay aligned. Reads like '1. text'.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-20 21:29:02 +03:00
claude code agent 227
b8655ae52c fix(page-templates): make page-embed Refresh actually re-render (#40)
The read-only embed renderer mounts a Tiptap EditorProvider with the looked-up
content, but Tiptap consumes the `content` option only at initial mount. After
Refresh busted the lookup cache and re-fetched fresh content, the new content
prop never reached the sub-editor, so the embed appeared not to update at all.

Key PageEmbedContent on result.sourceUpdatedAt (the source page's updatedAt,
already returned by the lookup and bumped on every persisted content change) so
the component and its EditorProvider remount and apply the refreshed content
when the source changes.

Note: server-side freshness vs. live collab edits is bounded by the 10s persist
debounce (collaboration.gateway.ts) — that separate limitation stays documented
in #40 and is out of scope here; this commit fixes the client never re-rendering.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-20 21:26:42 +03:00
claude code agent 227
c9eb495688 fix(page-templates): clean up page-embed node chrome (#39)
Two design problems on the whole-page embed (pageEmbed) node:

- Double selection frame: the generic square cyan .ProseMirror-selectednode
  outline stacked on top of the rounded .includeWrap border. Add node-pageEmbed
  to the existing outline:none rule (already covering the transclusion nodes) so
  only the single rounded border remains.
- Redundant 'open source' controls: the floating toolbar's external-link button
  duplicated the header badge title link. Remove the toolbar button; the badge
  title is now the single way to open the source (kept Refresh + ... menu).
  Also swap the badge fallback icon IconArrowsMaximize (read as 'expand') for a
  neutral IconFileText.

Follow-ups from review: render the badge whenever the source resolves (so the
only open-source link can't vanish when title+icon are empty), and label the
link (title/aria-label) + add the 'Open source page' i18n key (en-US, ru-RU).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-20 21:21:32 +03:00
claude code agent 227
859223db1a fix(page-templates): show a template marker icon in the page tree (#38)
Template pages were toggleable but indistinguishable in the sidebar tree.
Render an IconTemplate next to the title when node.isTemplate is true, wrapped
in a Tooltip(label='Template') with an aria-label + role='img' for AT. The
icon is a child of the row Link so clicks navigate as normal; pointer events
stay enabled so the tooltip's hover handlers fire. Adds the 'Template' i18n
key to en-US and ru-RU (other locales fall back to en-US).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-20 21:15:43 +03:00
claude_code
19ae6a0efa Merge pull request 'feat(editor): page templates — live whole-page embed (MVP)' (#17) from feat/page-templates into develop
Some checks failed
Develop / build (push) Has been cancelled
2026-06-20 20:34:44 +03:00
claude_code
2b3fc926cc Merge remote-tracking branch 'gitea/develop' into feat/html-embed-admin
# Conflicts:
#	apps/server/src/core/workspace/services/workspace.service.ts
2026-06-20 20:18:44 +03:00
claude_code
e9e9f74ec6 Merge remote-tracking branch 'gitea/develop' into feat/page-templates
# Conflicts:
#	apps/server/src/integrations/throttle/throttle.module.ts
#	apps/server/src/integrations/throttle/throttler-names.ts
2026-06-20 20:18:42 +03:00
claude code agent 227
52efd37fd9 fix(page-templates): import ThrottleModule into collab app so it boots
PageTemplateController (added on this branch) guards its lookup/toggle routes
with UserThrottlerGuard, which depends on the throttler options provided by
ThrottleModule. CollaborationModule -> TransclusionModule registers that
controller, and the collab server bootstraps CollabAppModule, which did not
import ThrottleModule. The API server's AppModule does, so :3000 booted, but
the collab server (:3001) crashed at startup with
'Nest can't resolve dependencies of the UserThrottlerGuard ... THROTTLER:MODULE_OPTIONS'.
Without collab the editor can't sync, so live editing was broken on this branch.

Import ThrottleModule into CollabAppModule, mirroring AppModule, so the guard
resolves in the collab process too.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-20 20:10:09 +03:00
claude_code
6128920264 Merge pull request 'feat(public-share): selectable agent-role identity + fix floating-icon overlap' (#25) from feat/share-assistant-identity-and-branding into develop 2026-06-20 19:59:48 +03:00
claude_code
cf29a0fc11 0.93.0 2026-06-20 19:57:37 +03:00