* MUL-4016: fix mention tokenizer stacktrace backtracking
Co-authored-by: multica-agent <github@multica.ai>
* fix(editor): de-ambiguate escaped-label regexes to kill ReDoS (MUL-4016)
The mention/slash/file-card label regexes used `(?:\\.|[^\]])` where both
alternatives can consume a backslash. On an unterminated match, each `\x`
run is enumerated 2^n ways — pasting a Java stacktrace (`\~\[...\]`) or a
crafted ~50-char string freezes the main thread for seconds (GitHub #4881).
Exclude backslash from the char class (`[^\]\\]`) so a backslash can only be
consumed by `\\.`. The alternatives become disjoint and matching is linear;
legal escaped-bracket labels like `David\[TF\]` still parse unchanged.
Fixed in all four sites that shared the pattern:
- mention-extension.ts tokenize()
- slash-command-extension.ts start() + tokenize()
- file-card.tsx FILE_CARD_MARKDOWN_RE
- packages/ui/markdown/file-cards.ts NEW_FILE_CARD_RE (runs on every
read-only comment/description render, not just the editor)
Adds adversarial regression tests (repeated `\a` + missing closing bracket)
that fail in ~10-40s against the old regexes and pass in <1ms after the fix.
Builds on #4889's marker-first mention start().
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
* fix(editor): escape backslash in mention/slash labels for round-trip (MUL-4016)
Follow-up to the de-ambiguation fix, addressing Howard's PR review.
The linear tokenizer now treats "\" as an escape lead (\\.), so a label whose
serialized form contains a bare "\" adjacent to the closing "]" no longer parses
back — the "\]" is consumed as an escaped bracket and swallows the boundary. The
old ambiguous regex tolerated this by chance; the de-ambiguation exposes it.
mention/slash renderMarkdown escaped only [ and ], not \. Switch both to the
shared escapeMarkdownLabel() (escapes [ ] \ ( )) and mirror it on parse with
replace(/\\([[\]\\()])/g, "$1"), matching what file-card already does. This also
converges the three tokenizers on one escape contract. file-card was already
correct and is unchanged.
Adds parameterized round-trip tests for labels containing "\" / "\]" / parens
(e.g. "A\\", "ends\\", "a\\]b"); these fail on the old serializer and pass now.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
The real ordered-list caret bug (MUL-3973). Typing 1. in the comment box
persists the draft "1. \n\n"; on remount @tiptap/markdown parses that empty
item into a schema-invalid, childless listItem, leaving the document with an
AllSelection instead of a text cursor — so the browser paints the caret on the
following block and it can't be moved back into the list.
Verified against the REAL editor (real createEditorExtensions + @tiptap/markdown,
no @tiptap/react mock). Non-empty items round-trip fine; only the empty-item
round-trip corrupts, which the reverted #4813 never exercised.
- repairEmptyListItems(): rebuild from JSON so every list/task item leads with a
paragraph (covers empty items and nested items whose first child is a
sub-list); reset the AllSelection first (else setContent collapses the list);
keep the whole repair off the undo stack; restore the prior caret (sync path)
or land in the list item (mount).
- Called in onCreate (mount) and after the sync-effect setContent.
- Real-editor tests incl. undo-does-not-revive and nested-list schema validity.
Co-authored-by: multica-agent <github@multica.ai>
* feat(composio): server-side connect flow + connections REST (Notion MVP) (MUL-3720) (#4608)
* feat(composio): server-side connect flow + connections REST (Notion MVP) (MUL-3720)
Compose the merged server/pkg/composio SDK into a user-facing connection
manager: signed-state connect handshake, local user_composio_connection
mirror, idempotent disconnect, and a per-user MCP session helper (not yet
wired into task dispatch).
- migration 127_user_composio_connection (no FK/cascade, per DB rules)
- sqlc queries: upsert (idempotent on user_id+connected_account_id), list
active, owner-scoped get, mark revoked
- internal/integrations/composio: signed HMAC-SHA256 state, BeginConnect,
CompleteCallback (idempotent upsert), ListConnections, Disconnect
(upstream 404 = idempotent success), CreateMCPSession (no-op when empty,
pins connected_accounts per toolkit), CallbackRedirect
- REST handlers under /api/integrations/composio (user-scoped, 503 when
COMPOSIO_API_KEY unset): connect/init, callback (302), connections list,
delete
- router wiring gated by COMPOSIO_API_KEY; COMPOSIO_AUTH_CONFIGS_JSON maps
toolkit->auth_config (MVP: notion); state secret from COMPOSIO_STATE_SECRET
or derived from JWT_SECRET; callback base from COMPOSIO_CALLBACK_BASE_URL
or MULTICA_PUBLIC_URL
- tests: state (expire/tamper/wrong-secret), service (mapping, callback
idempotency, non-success, disconnect owner/404 idempotency, MCP pin),
handlers (httptest), redact regression for Bearer mcp_ tokens
MVP scope: Notion only; no task-dispatch overlay, sharing, or webhook
event handling (later stages).
Co-authored-by: multica-agent <github@multica.ai>
* fix(composio): bind callback account to user + idempotent revoked disconnect (MUL-3720)
Address PR 4608 review (CHANGES_REQUESTED):
- callback: verify connected_account_id with Composio before mirroring it.
The signed state only proved user/toolkit/exp, so a valid state paired with
a tampered connected_account_id would be written verbatim. CompleteCallback
now calls ListConnectedAccounts and fails closed (ErrAccountVerification)
unless the account belongs to the state's user (composio_user_id == multica
user id) and was created under the toolkit's auth config. No row is written
on mismatch / unknown account / upstream error.
- disconnect: short-circuit to a no-op when the local row is already revoked,
before touching upstream. Previously a second DELETE re-hit Composio and a
non-404 upstream error surfaced as a 502, breaking the 204-idempotent
contract.
- CreateMCPSession: document the v1 single-active-connection-per-(user,toolkit)
constraint and make duplicate selection deterministic (newest-wins, rows are
connected_at DESC) instead of order-dependent map overwrite. Stage 3 owns the
real single-account-enforcement vs multi-account-shape decision.
Tests: tampered/wrong-auth-config/unknown-account callback rejection, revoked-row
disconnect no-op (asserts upstream not re-hit). composio pkg 85% coverage; all
green.
Co-authored-by: multica-agent <github@multica.ai>
* feat(composio): list all toolkits + dynamic auth-config resolution (MUL-3720)
Yushen's follow-up to the Notion MVP: surface the full Composio toolkit
catalog, render it in Settings, and drop the static env mapping in favor of
dynamic auth-config discovery.
Config correctness (per Composio docs):
- Remove COMPOSIO_AUTH_CONFIGS_JSON entirely. The toolkit→auth_config mapping
is now resolved at request time from the project's /auth_configs (cached,
5-min TTL), so enabling a toolkit is a dashboard action, not a redeploy.
- Do NOT add COMPOSIO_PROJECT_ID. The project API key (x-api-key) authenticates
to exactly one project; the project is resolved from the key. Only org-level
endpoints use x-org-api-key, which this integration never calls.
Backend:
- SDK: server/pkg/composio/auth_configs.go — ListAuthConfigs (toolkit_slug,
is_composio_managed, show_disabled, limit, cursor).
- service: dynamic resolver (authConfigMap cache; betterAuthConfig prefers a
custom/white-label config over Composio-managed, newest wins); BeginConnect
and CompleteCallback resolve via it; ListToolkits fetches the full catalog
(paginated, capped) annotated with connectable = has an enabled auth config,
connectable-first ordering.
- handler + route: GET /api/integrations/composio/toolkits (user-scoped, 503
when COMPOSIO_API_KEY unset) returning slug/name/logo/category/connectable.
Frontend:
- core: ComposioToolkit/ComposioConnection types, api client methods, and
composio query options (@multica/core/composio).
- views: Settings → Integrations now has a Composio section rendering every
toolkit as a card with search. Connect is gated on `connectable`;
non-connectable toolkits show a muted "not configured" hint instead of a
dead button. Connected toolkits show a badge + Disconnect (with confirm).
- i18n: composio block added to en/zh-Hans/ja/ko settings.
Tests: SDK + service (dynamic resolution, custom-over-managed preference,
connectable flag, resolver-error soft-degrade) and handler toolkits endpoint;
composio pkg 85.7% coverage. go build/vet/gofmt clean; core+views typecheck,
core+views lint, and core tests (691) all green.
Co-authored-by: multica-agent <github@multica.ai>
* fix(composio): close cross-toolkit callback fail-open by signing auth_config_id into state (MUL-3720)
Re-review blocker: CompleteCallback resolved the toolkit's auth config at
callback time and ignored a resolve error/empty result, while
verifyAccountOwnership skipped the auth-config comparison when the expected
value was empty. A user could then pass another toolkit's connected_account_id
into this toolkit's callback — the owner check passed and it was written under
the wrong toolkit_slug/account binding.
Fix: the auth_config_id is already resolved in BeginConnect (before the state
is signed), so sign it into the state and compare it exactly at callback. No
re-resolve, no fail-open. verifyAccountOwnership now fails closed when the
expected auth config is empty (rejects instead of skipping) and requires an
exact match — closing the cross-toolkit binding gap.
Tests: state round-trips auth_config_id; BeginConnect signs it; callback
rejects wrong/cross-toolkit auth config and an empty (no-mapping) auth config
fails closed. composio pkg 85.2% coverage, all green.
Frontend (non-blocking): the Composio settings tab now surfaces an error when
the connections query fails instead of silently rendering everything as
unconnected.
Co-authored-by: multica-agent <github@multica.ai>
* fix(composio): hide Settings section entirely when integration unconfigured (MUL-3720)
Decision (option 2, hide-then-merge): don't show a card that leaks the internal
COMPOSIO_API_KEY env-var name to every end user. IntegrationsTab now gates the
whole Composio section (heading + body) on the toolkits query — a 503 means the
key is unset, so the section is withheld instead of rendering the not-configured
card. Admin-only setup guidance is a later, role-gated affordance.
Removed the notConfigured card (and now-unused ApiError import) from
ComposioTab; it only mounts when configured. views typecheck + lint clean.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
* feat(composio): Stage 2 frontend polish — callback toast, last_used & expired UI, e2e (MUL-3718) (#4688)
* feat(composio): callback toast + refresh, last_used & expired UI, e2e (MUL-3718)
Co-authored-by: multica-agent <github@multica.ai>
* fix(composio): real callback redirect route + StrictMode-safe toast dedup (MUL-3718 review)
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
* fix(composio): callback endpoint should not require Multica auth (MUL-3843) (#4709)
* fix(composio): move OAuth callback out of the Auth group (MUL-3843)
Composio 302-redirects the browser to /api/integrations/composio/callback
at the end of the OAuth flow, but PR #4608 mounted it inside the cookie-auth
middleware group. When the session cookie is absent (expired session,
SameSite=Strict / Safari ITP, private window, self-hosted callback subdomain)
the Auth middleware returned a hard 401 and a JSON blob instead of the
settings redirect, breaking the flow.
Identity never came from the cookie anyway: it is carried by the HMAC-signed
state param that CompleteCallback verifies (signature, expiry, replay) and
cross-checked by verifyAccountOwnership; h.Composio == nil still 503s. So the
callback is registered alongside the other public OAuth/webhook routes; the
other four composio endpoints stay session-gated.
Refs MUL-3843, MUL-3715.
Co-authored-by: multica-agent <github@multica.ai>
* fix(composio): correct stale callback routing comments (MUL-3843)
The package header and ComposioCallback doc comments still described the
callback as sitting under the Auth middleware group. After the route was
moved out (this PR), update both to state it is a public route whose identity
comes from the signed state — addressing review nit from 张大彪.
Refs MUL-3843.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
* feat(composio): inject MCP overlay into agent runtime at task dispatch (MUL-3721) (#4704)
Stage 3 of the Composio epic. Wires the per-user Composio MCP session into
every agent task so the agent process sees the initiator's connected tools
without any prompt-time plumbing.
Server side
- Migration 128 adds agent_task_queue.runtime_mcp_overlay JSONB plus a
BEFORE-UPDATE trigger that wipes the column on any transition into a
terminal status (completed / failed / cancelled). A trigger is the single
source of truth — future queries that flip status cannot bypass it.
- composio.Service.BuildTaskOverlay(userID) reuses CreateMCPSession and
emits the Claude-style { mcpServers: { composio: { type: http, url,
headers } } } shape the daemon's existing sidecar generators consume.
Returns (nil, nil) on zero active connections so we never burn a
Composio session for a user with nothing to call.
- TaskService grows a Composio ComposioOverlayBuilder seam, wired in
router.go after composiointeg.NewService succeeds. Five enqueue paths
(issue / mention / quick-create / chat / auto-retry) attach the overlay
after CreateAgentTask returns and before the daemon is notified — so
every claim reads a settled row, with no second daemon hop. Best-effort:
a builder failure logs and proceeds with no overlay.
- resolveInitiatorFromTriggerComment derives the initiator user from the
trigger comment when it was authored by a member. Agent-authored
triggers are not treated as initiators (their connected-apps view is
empty by construction).
Daemon side
- handler/daemon.go claim path merges task.runtime_mcp_overlay onto
agent.mcp_config via mergeMCPOverlay before populating
TaskAgentData.McpConfig. Overlay wins on server-name collisions
because it carries the live user-scoped session URL. Errors fall back
to the agent config unchanged — a bad overlay must not surprise-disable
saved MCP tools. The existing execenv sidecar generators (cursor /
codex / openclaw / opencode / hermes-kiro) need no changes: they keep
consuming the merged result through TaskAgentData.McpConfig.
Tests
- 9 merge cases (mcp_overlay_test): both-nil short-circuit, agent-only
pass-through, overlay-only canonicalization, two-side merge, name
collision (overlay wins), top-level key preservation, malformed agent
fallback, malformed overlay fallback, non-object server rejection.
- 4 dispatch cases (composio): zero-connections returns nil without
CreateSession, happy-path emits the right shape with the right user
id, empty-URL defensive branch, SDK error surfacing.
- 4 TaskService helper cases: nil Composio is a no-op (Queries-safe),
invalid initiator does not call the builder, nil overlay skips the
UPDATE, builder error swallowed without panic.
- Migration 128 verified to roll up + down + up cleanly against the test
database.
Out of scope (deferred): assignment-triggered enqueue paths with no
trigger comment get no overlay attached today (no initiator UUID flows
through enqueueIssueTask in that case). Retry paths recompute the overlay
fresh from the parent's initiator_user_id instead of inheriting the bearer
from the parent row, so a stale token can never resurface on a retry.
Co-authored-by: Eve <eve@multica.ai>
Co-authored-by: multica-agent <github@multica.ai>
* feat(composio): per-agent allowlist + originator-scoped MCP overlay (MUL-3869) (#4736)
* feat(composio): per-agent allowlist + originator-scoped MCP overlay (MUL-3869)
Stage 3.1 of the Composio epic (MUL-3721 parent). PR #4704 wired in the
runtime_mcp_overlay column and a per-task dispatch hook; this change
inverts the default from "all-on" to opt-in and locks the overlay to the
agent owner's own connected apps:
- Agents carry composio_toolkit_allowlist TEXT[]. NULL or [] => no MCP.
Owner-only read/write; non-owner GET/PUT silently redacts/drops the
field (same shape as mcp_config).
- agent_task_queue carries originator_user_id UUID. Set from the
top-of-chain HUMAN at every enqueue path:
* issue/mention comment by member -> author_id
* issue/mention comment by agent -> inherit via comment.source_task_id
-> parent task originator_user_id
* quick-create -> requester_id
* chat -> initiator_user_id
* retry -> SQL-inherited from parent row
* autopilot -> NULL (system-driven)
- BuildTaskOverlay (composio dispatch) now takes (ctx, originatorUserID,
agent) and short-circuits on five gates: invalid originator,
originator != agent.owner_id, empty allowlist, empty intersection of
allowlist ∩ active connections, defensive empty session URL. Composio
CreateSession is called with BOTH `toolkits.slugs` (the intersection)
AND `connected_accounts` (the pinned account ids), narrowing the
tool-router twice.
- The originator-vs-owner gate closes the agent-fanout privacy hole: any
workspace member who can @-mention a public agent used to project the
owner's connected apps into their run. Now the overlay only mounts
when the human at the top of the chain IS the agent owner.
Tests:
- dispatch_test.go covers all 5 gates plus uppercase/whitespace slug
normalisation.
- task_runtime_mcp_overlay_test.go covers the no-op gates of the new
applyRuntimeMCPOverlay signature.
- agent_composio_allowlist_test.go (handler): owner roundtrip
(list/empty/null), workspace-admin silent-drop, owner-only GET
visibility, pure normaliseComposioToolkitAllowlist.
- resolve_originator_test.go (service, DB-backed): member-authored,
agent-authored inherits via comment.source_task_id, invalid id.
Migration 129 up/down/up verified against docker postgres.
Co-authored-by: multica-agent <github@multica.ai>
* chore(composio): gofmt + regenerate sqlc with v1.31.1 (MUL-3869 review nits)
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: Eve <eve@multica-ai.local>
Co-authored-by: multica-agent <github@multica.ai>
* fix(composio): accept nested connected account auth config
* feat(views): creator-only MCP tab for per-agent Composio allowlist (MUL-3870) (#4743)
Stage 3.2 frontend on top of the Stage 3.1 backend (MUL-3869, 4708dba97).
Adds an agent-detail tab that lets the agent owner pick which of their own
active Composio connections this agent may mount as MCP servers, writing the
selection to agent.composio_toolkit_allowlist via the existing PUT /api/agents.
- core/types: composio_toolkit_allowlist (+ _redacted) on Agent; tri-state
composio_toolkit_allowlist on UpdateAgentRequest (omit/no-change, null/clear,
array/replace), matching the backend contract.
- core/agents: useUpdateAgentAllowlist - optimistic mutation hook (patches the
cached workspace agent list, rolls back on error, invalidates on settle).
- views: AgentMcpTab renders the owner's active connections as checkboxes;
empty state links to Settings -> Integrations; defensive redacted state.
- views: wired into AgentOverviewPane as tab "composio_mcp", labeled "MCP Apps"
to disambiguate from the existing raw-JSON "MCP" (mcp_config) tab. The entry
is gated to the creator (currentUserId === agent.owner_id), matching the
backend's owner-only read/write of the allowlist.
- i18n: tabs.composio_mcp + tab_body.composio_mcp.* in en/ja/ko/zh-Hans.
- tests: agent-mcp-tab.test.tsx (gating, toggle->allowlist body, active-only,
empty, redacted); e2e/agent-mcp.spec.ts (creator sees tab + PUT body,
non-creator hidden) with Composio + agent endpoints mocked at the boundary.
Note: the product spec says "creator"; the schema has no creator_id - the
backend gate and redaction are keyed on owner_id, so the tab uses owner_id.
Co-authored-by: multica-agent <github@multica.ai>
* fix(composio): mount remote MCP for codex
* feat(agents): agent invocation permission system (MUL-3963) (#4844)
* feat(agents): agent invocation permission system (permission_mode + invocation targets)
MUL-3963: split who may INVOKE an agent out of the overloaded visibility
column into an explicit, extensible model on feature/composio-integration.
- DB: agent.permission_mode (private|public_to) + agent_invocation_target
table (workspace/member/team targets) + lossless backfill from visibility
(migration 130).
- canInvokeAgent: owner-only for private (NO admin bypass, NO A2A bypass);
public_to honours the allow-list; A2A judged by the top-of-chain originator.
- All trigger paths rewired: issue assign, comment @agent/@squad, chat,
quick-create, autopilot, squad leader, child-done.
- Agent API: permission_mode + invocation_targets on responses and
create/update (owner-only writes); legacy visibility kept as a derived field
so old clients never see a permission widening.
- Composio: BuildTaskOverlay now FOLLOWS invocation permission and uses the
agent OWNER connection (removed the originator==owner gate); front-end warns
when a shared agent enables Composio apps.
- CLI: --permission-mode / --public-to-workspace / --public-to-member (legacy
--visibility still mapped).
- Frontend: AccessPicker (Private / workspace / specific people / team soon),
permission rules mirror canInvokeAgent, Composio warning banner.
- Tests: migration backfill, admin cannot invoke others private, public_to
workspace/member whitelist, A2A by originator, Composio overlay uses owner
connection.
Co-authored-by: multica-agent <github@multica.ai>
* feat(agents): stackable, mixed public_to invocation targets (MUL-3963)
Follow-up on PR #4844: public_to now supports selecting MULTIPLE, MIXED
targets on one agent (e.g. Public to workspace + specific people + team),
with canInvokeAgent admitting on ANY matching target (OR).
- Frontend AccessPicker: reworked from a single exclusive kind into a
stackable multi-select — an "Everyone in workspace" toggle, a member
multi-select checklist, and a (disabled, v1) team placeholder can be
combined freely. Emits the full union of selected targets; empty union
collapses to Private. Existing team targets are preserved across saves.
Added the access.public_group locale string (en/zh-Hans/ja/ko).
- Backend already supported this (agent_invocation_target is multi-row per
agent; create/update take a target ARRAY and batch-replace the whole
allow-list; canInvokeAgent OR-matches). Added tests to lock it in:
mixed member+team targets, overlapping-member batch replace, and
workspace+member stacking then narrowing.
Refs MUL-3963.
Co-authored-by: multica-agent <github@multica.ai>
* fix(agents): address review on invocation permission (MUL-3963)
张大彪 review on PR #4844 — three blockers + product ruling + nits:
1. Migration 130: drop the FK/cascade on agent_invocation_target
(agent_id, created_by) per the Multica no-FK rule; relationships are now
maintained in the app layer (matching MUL-3515 §4). Added
DeleteAgentInvocationTargetsByArchivedRuntimeAgents and call it before
DeleteArchivedAgentsByRuntime in all three runtime-delete paths
(runtime.go x2, runtime_profile.go) so hard-deleting agents can't orphan
target rows.
2. revokeAndRemoveMember: prune the leaving member's member-target grants
(DeleteAgentInvocationTargetsByMember) in the same tx as the member-row
delete, so a re-invited user can't reclaim a stale invocation grant.
3. Empty public_to is a phantom — parsePermissionInput now normalises a
public_to with no resolvable targets to a single workspace target, so
`--permission-mode public_to` alone (and any empty target array) means
"public to workspace" instead of "shared but nobody can run it".
Product ruling: the system/no-human-originator → workspace-target path in
canInvokeAgent is a deliberate, documented exception (webhook/system/
workspace-wide automation); member/team targets still fail closed without a
resolved originator. Documented in code + locked with a test.
Nits: refreshed the stale "originator must be owner" comments — models.go
(via migration 130 COMMENT ON COLUMN + sqlc regen for composio_toolkit_allowlist
and originator_user_id) and agent-mcp-tab.tsx — to the owner-connection +
invocation-permission rules.
Tests: member remove/re-add regression, system workspace exception + member
fail-closed, empty public_to → workspace (plus the earlier mixed/overlap/
batch-replace suite). Migration 130 applied to the test DB; Go handler/service/
composio suites green; views typecheck clean.
Refs MUL-3963.
Co-authored-by: multica-agent <github@multica.ai>
* fix(agents): scope member invocation-target cleanup to one workspace (MUL-3963)
张大彪 3rd review — cross-workspace permission bug + comment nits:
- DeleteAgentInvocationTargetsByMember was a GLOBAL delete by user id, so
removing a user from workspace A also wiped their member-target grants on
agents in workspace B. Scoped it to a single workspace by joining through
agent.workspace_id; revokeAndRemoveMember now passes (workspaceID, userID).
- Regression test TestRevokeMember_InvocationTargetCleanupIsWorkspaceScoped:
same user allow-listed by agents in two workspaces; removal from one leaves
the other workspace's target intact.
- Nits: refreshed the remaining stale "originator == agent.owner_id" /
"owner-vs-originator" comments — CreateRetryTask (agent.sql, regenerated),
and the AgentResponse allowlist doc + ListAgents/UpdateAgent redaction
rationale in agent.go — to the owner-connection + invocation-permission rule.
Migration 130 applied to the test DB; Go handler/service/composio suites green;
go vet clean.
Refs MUL-3963.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
* fix(agents): agent access owner-only editable, read-only for others (MUL-3963) (#4853)
* fix(agents): make agent access owner-only editable, read-only for others (MUL-3963)
Interaction bug: a non-owner (incl. workspace admin) could open the AccessPicker
and set an agent public — the backend silently ignored it and the UI bounced
back to private. Access is owner-only, so non-owners must see a read-only state
and the backend must reject real changes explicitly.
Frontend:
- AccessPicker renders a static, non-interactive read-only state when the
viewer is not the owner: the current access value + a lock affordance + a
tooltip "Only the agent owner can change who can run this agent." No clickable
trigger is rendered, so a non-owner can never open a control the backend would
reject (the GitHub/Notion pattern for permission settings you can see but not
edit). The editable multi-select picker is unchanged for the owner.
- agent-detail-inspector gates the picker on ownership specifically
(currentUserId === agent.owner_id), NOT the general canEdit (which also admits
admins, who may edit other fields but not access).
- New locale key access.owner_only_readonly (en/zh-Hans/ja/ko).
Backend:
- UpdateAgent now returns an explicit 403 when a non-owner submits a REAL
permission change (permissionInputChangesAgent compares requested mode +
target set against the persisted state); a no-op resubmit (admin PATCH-as-PUT
echoing unchanged permission) is still tolerated so admin edits of other
fields keep working. Replaces the previous silent-drop that caused the bounce.
Tests:
- access-picker.test.tsx: non-owner gets a non-interactive read-only display
with the owner-only tooltip; owner gets an interactive picker; owner can pick
a member and stack workspace + member.
- TestUpdateAgent_AccessChangeIsOwnerOnly: admin real change → 403; admin no-op
resubmit → 200; admin editing other fields → 200; owner change → 200.
Incidental: fixed a pre-existing base typecheck break in
slash-command-suggestion.test.tsx (stray `signal` arg not in the suggestion
items type) that otherwise fails the whole @multica/views typecheck.
Refs MUL-3963.
Co-authored-by: multica-agent <github@multica.ai>
* fix(agents): compare legacy visibility, not expanded permission, for no-op detection (MUL-3963)
PR #4853 review: permissionInputChangesAgent expanded a legacy-only
visibility:"private" into a real private permission and compared it against the
agent's actual permission. A member-only public_to agent derives legacy
visibility "private", so an admin PATCH-as-PUT echoing visibility:"private"
while editing another field was misread as a public_to→private downgrade and
rejected with 403 — contradicting the "unchanged permission no-op is allowed"
contract.
Fix (per review): when a request carries ONLY legacy `visibility` (no
permission_mode / invocation_targets), derive the agent's CURRENT legacy
visibility from its real targets and compare the legacy string values. Equal =
no-op (allowed); a real legacy change (e.g. "workspace") still returns 403.
Requests that carry permission_mode / invocation_targets keep the precise
mode+target comparison.
Regression test TestUpdateAgent_LegacyVisibilityNoOpForMemberOnlyPublicTo:
member-only public_to agent — admin submitting visibility:"private" + a
non-permission field → 200 with targets unchanged; admin submitting
visibility:"workspace" → 403.
Go handler/composio suites green; migration 130 applied; go vet clean.
Refs MUL-3963.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
* feat(composio): brief agents on connected apps
* feat(composio): gate MCP apps behind feature flag
* fix(mobile): parse agent invocation permissions
* fix(tests): update agent fixtures for access fields
---------
Co-authored-by: multica-agent <github@multica.ai>
Co-authored-by: Multica Eve <eve@devv.ai>
Co-authored-by: Eve <eve@multica.ai>
Co-authored-by: Eve <eve@multica-ai.local>
* 📝 [需求] 有序列表光标切换缺陷提案 / docs: requirements for 26-editor-ordered-list-cursor
Co-authored-by: multica-agent <github@multica.ai>
* fix(editor): keep the caret in the list item when the comment box remounts
Typing `1.` makes an ordered list, then switching issues and back yanked the
caret to the next line and re-yanked it there whenever the user moved it. The
comment box remounts on issue switch (key={id}) and feeds the persisted draft
back as defaultValue; the content-sync effect treated the redrawn draft as an
external change because @tiptap/markdown reserializes an ordered list slightly
differently from its source, re-parsed it, and restored the caret with a bare
Math.min offset clamp that resolved onto the list's structural (non-text) gap.
- Short-circuit the sync effect when the incoming parser INPUT matches what was
last applied (seeded at mount), so an identical draft is never re-parsed.
- Replace the Math.min clamp with clampSelectionToText, which snaps the prior
offsets to the nearest valid text position via TextSelection.between.
Tests: restore-selection unit tests (real editor) + a content-editor remount
regression test.
Co-authored-by: multica-agent <github@multica.ai>
* fix(editor): seed the caret guard synchronously and keep it in sync with the doc
Addresses cross-review findings on the content-sync guard:
- Seed appliedIncomingRef at render instead of in onCreate. Tiptap v3's
onCreate is deferred past the first sync-effect run, so the guard was null on
that run and a remount could still re-parse the ordered-list draft (the mock's
synchronous onCreate had hidden this).
- Advance appliedIncomingRef on the Guard 3 short-circuit too, so a later
external change back to an older value (e.g. a collaborator reverts the
description) is no longer mistaken for an already-applied input and swallowed.
Tests: add a suppressed-onCreate remount case and an A->B->A external-revert
regression; 23 passed.
Co-authored-by: multica-agent <github@multica.ai>
* docs: remove internal requirements proposal from the PR
Keep the PR to code and comments only, per reviewer request. The requirement
context lives in the tracking issue, not in this open-source code change.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
* fix(markdown): render inline data-URI images (MUL-3961)
Inline data:image/* URIs (QR codes, charts, base64 screenshots) were
stripped and rendered as broken images. Two gates dropped the src:
- rehype-sanitize's protocols.src only allowed http/https
- react-markdown's defaultUrlTransform blanks any data: URL to ''
Allow data:image/* through both gates, narrowed to image subtypes only
(non-image data URIs stay rejected) and leaving every other src form
unchanged. file-cards.ts data: rejection is intentional and untouched.
Co-authored-by: multica-agent <github@multica.ai>
* fix(markdown): allow data:image/* in ReadonlyContent too (MUL-3961)
Issue comments and other read-only surfaces render through ReadonlyContent,
which keeps its own sanitize schema + urlTransform separate from the base
Markdown component. Both still stripped data: URIs, so an agent inlining an
auth QR code in an issue comment still saw a broken image.
Apply the same image/*-narrowed data: allowance (protocols.src +
attributes.img + urlTransform) and add a regression test covering the
comment / readonly path. file-cards.ts data: rejection stays untouched.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: J <agent-j@multica.ai>
Co-authored-by: multica-agent <github@multica.ai>
* fix(views): preflight mermaid syntax before rendering
* refactor(views): use suppressErrorRendering instead of a pre-render parse
Enable Mermaid's suppressErrorRendering so render() throws on invalid
syntax without drawing its built-in error graphic into the DOM, letting
the existing catch fall back to the compact error state. This drops the
redundant mermaid.parse() pass over valid charts while still fixing the
orphaned error-SVG artifact from #3653.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: J <j@multica.ai>
Co-authored-by: multica-agent <github@multica.ai>
In a list item that cannot indent (first child / max depth), Tab was
returned unhandled, so the browser's native Tab moved focus out of the
editor onto adjacent controls. Decouple "swallow the key" from "did the
indent move anything": best-effort indent, then swallow whenever the
caret is inside the list (editor.isActive(name)) and only fall through
to focus navigation when not in a list. Covers bullet, ordered and task
lists via the shared keymap.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Stock prosemirror-schema-list `sinkListItem` returns false without
dispatching whenever `range.startIndex === 0`, so selecting a list from
the top and pressing Tab did nothing. Bullet, ordered, and task lists
all routed through the same command and were equally affected.
Wrap the shared Tab keymap (PatchedListItem + PatchedTaskItem) with
`sinkListItemRange`: try stock sink first, and when it bails on a
multi-item range whose first item is the list's first child, re-run the
stock command on a selection narrowed to start inside the second selected
item. The first item stays as an anchor and the rest nest under it
(Notion/GitHub nested-list behaviour), in a single undoable transaction.
Shift-Tab / liftListItem already handles ranges and the first-item case,
so it is unchanged. No schema change.
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
Selecting text in a readonly code block (comment/issue markdown) lost the
selection within seconds, making copy impossible, whenever the surrounding
view re-rendered — most reliably while a sibling agent task streamed over
WebSocket (a re-render roughly every ~100ms).
Root cause: the `code` renderer emits highlighted HTML via
`dangerouslySetInnerHTML={{ __html }}`, a fresh prop object every render. Each
unrelated parent re-render re-ran react-markdown, and React rewrote the
`<code>` innerHTML even though the HTML string was byte-identical, tearing down
and rebuilding all 161 hljs `<span>` nodes. The native selection is anchored to
those nodes, so it collapsed.
Fix: memoize the entire `<ReactMarkdown>` subtree on its only real inputs
(`processed` + `components`). A stable element reference lets React bail out of
the subtree on unrelated re-renders, so the code-block DOM is never rebuilt
while content is unchanged. Confirmed via an instrumentation probe: zero
`<code>` DOM mutations during streaming after the fix.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(editor): accept highlighted composer suggestion on Tab
Plain Tab now accepts the highlighted mention / slash-command suggestion,
matching Enter, across every composer built on the shared TipTap editor
(chat, issue description, comment/reply). A single shared isPickerAcceptKey
predicate centralizes the accept-key policy so the two picker lists stay in
sync instead of each re-deciding what counts as accept.
Shift+Tab and Ctrl/Cmd/Alt+Tab are intentionally NOT accept keys, so reverse
focus navigation and OS window switching are preserved. When no picker is
open, Tab keeps its existing behavior (list indent / focus traversal).
Adds unit coverage for both picker lists plus a plugin-order guard that fires
Tab through real ProseMirror dispatch with the caret inside a list item,
proving the suggestion layer outranks PatchedListItem's Tab -> sinkListItem.
Scope: web/desktop shared composer only; mobile and the generic combobox are
untouched.
Refs: MUL-3685
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
* test(editor): make Tab/list-item priority guard actually exercise sinkListItem
The guard placed the caret in the first (only) bullet item, where
sinkListItem is a no-op (no preceding sibling to nest under), so the test
passed regardless of whether the suggestion layer intercepted Tab. Rebuild
the fixture as a two-item list with the caret in the SECOND item, where
Tab -> sinkListItem can fire, and add a sanity control proving a bare Tab
(no picker) does sink that item — so the accept-wins assertion is meaningful.
Production code unchanged.
Refs: MUL-3685
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
Add a dedicated checkbox/task-list toggle button to the editor bubble menu, between the List dropdown and Quote. It turns the current line(s) into a `- [ ]` task item or back to a paragraph in one tap, reusing the existing toggleTaskList() command and the same Toggle/Tooltip pattern as the neighboring controls. Adds bubble_menu.task_list locale keys for en/zh-Hans/ja/ko.
Co-authored-by: Lambda <lambda@multica.ai>
Co-authored-by: multica-agent <github@multica.ai>
The @mention popup tracked the highlighted row with a positional integer
(selectedIndex into displayItems), but rows are rendered in the re-bucketed
order produced by groupItems() (current → recent → search → users → issues).
An async server "search" result is appended to the END of displayItems yet
hoisted near the TOP on render, so the highlighted row and the committed
item pointed at different entries — you navigate to one target but mention
its neighbour. A separate useEffect also force-reset selectedIndex to 0 on
every displayItems change, snapping an active selection back to the first row
whenever async results landed.
Root cause: a slot index is not a stable target for a list whose order and
length change asynchronously. Track selection by item identity instead:
- Replace selectedIndex state with selectedKey (the item's `type:id`).
- Derive groups/orderedItems (the exact rendered order) and resolve the
numeric index from selectedKey against orderedItems; fall back to row 0
when the pinned item is gone or nothing is picked yet.
- Keyboard nav, Enter, clicks, highlight, and scroll all index orderedItems,
so the highlighted row always equals the committed item.
- Drop the force-reset effect; identity-based selection self-heals across
reorders and async arrival without resetting an active pick.
Adds a regression test asserting the highlighted row equals the committed
item when groupItems reorders the list, both initially and after ArrowDown.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The @mention popup navigated and committed by indexing the flat
`displayItems` array, but rendered rows in the re-bucketed order from
`groupItems()` (current → recent → search → users → issues). In chat
(context mode) the async server-search results are appended to the end
of `displayItems` yet tagged `group:"search"`, so `groupItems()` hoists
them near the top. The highlighted row and the committed item then point
at different entries — you select one target but mention its neighbour
(the reported "@bohan picks the next one" off-by-one).
Make the flattened grouped order (`orderedItems`) the single index space
for `selectedIndex`, arrow keys, Enter, and clicks, so the highlighted
row is always the committed item. Plain issue-comment mentions (default
mode) were already safe — no group tags means `groupItems()` is
order-preserving — and stay unchanged.
Adds a regression test asserting highlighted row == committed item when
the list is reordered by a hoisted search result.
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
In context mode (chat) a query merges context items (current page / recently
viewed) with search results into one popup. The old contextLayout made only the
"Recent" group scrollable (`min-h-0`) while every other group was `shrink-0`, and
the Recent `<section>` did not clip its own overflow. When the search groups
(Users/Issues) filled the height, the flex algorithm squeezed Recent toward zero
and its un-clipped rows painted on top of the groups below — the overlap users saw.
Collapse the two render branches into a single `overflow-y-auto` flex column so
every group just stacks and the whole popup scrolls; no group can collapse onto
another. The context variant only differs in width / max-height / chrome.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
#4288 swapped the chip cap from a fixed `max-w-72` to `max-w-[min(18rem,100%)]`.
A percentage max-width on a flex item is dropped while its flex-container wrapper
(`<a class="inline-flex">`) computes its own max-content size, so the wrapper
ballooned to the untruncated title width while the chip truncated to the cap —
leaving an empty, clickable strip after the visible chip.
Fix it as standard atomic-inline behavior instead of a fixed magic cap:
- IssueChip caps at `max-w-full` with the title truncating to an ellipsis, so it
wraps to the next line as a unit and only truncates once a whole line can't
hold it.
- Drop `inline-flex` from the editor NodeView `<a>` (issue + project) and the
markdown AppLink so the chip's only wrapper is a plain inline box with a
definite (line-based) percentage basis — no second flex container to balloon.
ProjectChip uses a definite `max-w-72`, so it never hit the gap; left as-is.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Extract three module-local helpers in content-editor.tsx and route the
duplicated call sites through them — no behavior change:
- normalizeMarkdown(md) / normalizeEditorMarkdown(editor): the single
definition of the "strip blob URLs + trimEnd" canonical form, replacing
the five editor-markdown sites and the one incoming-string site.
- hasUploadingNode(editor): replaces the two byte-identical document scans
(content-sync Guard 0 and hasActiveUploads).
The imperative getMarkdown() is deliberately left untrimmed; a safety-net
test pins its exact current return value.
Scope: WOR-59 upload-readiness review. file-upload.ts untouched; follow-up
items B1/F4/F5 are out of scope.
Co-authored-by: hzz <331380069@qq.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
* fix(markdown): don't auto-link bare filenames as external URLs
Agent comments that mention a project file like `plan.md` were turned into
clickable links to https://plan.md (dead external site). linkify-it fuzzy
detection matches `plan.md` as a domain because its extension is also a valid
TLD (md = Moldova; likewise io, sh, rs, py).
Suppress schemeless (fuzzy) linkify matches whose token is a bare filename
(single segment ending in a known source/config extension). Explicit schemes
(`https://plan.md`) and real domains (`example.com`) are unaffected. The file
extension list is now shared between the file-path and bare-filename detectors
so they can't drift.
Fixes#4222
Co-authored-by: multica-agent <github@multica.ai>
* docs(markdown): drop inaccurate .io example from bare-filename comment
io is not in the FILE_EXTENSIONS list, so .io domains are never suppressed.
Listing it as an example was misleading.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: J <j@multica.ai>
Co-authored-by: multica-agent <github@multica.ai>
* fix(editor): don't wipe in-flight uploads on external content sync
When a brand-new chat's first file upload triggers lazy session creation,
`setActiveSession(null → uuid)` flips ChatInput's draft key mid-upload, which
changes `defaultValue` to the new (empty) session draft. ContentEditor's
"sync external defaultValue" effect then ran `setContent` over a document that
still held the `uploading` image/fileCard node, wiping it — so the upload's
finalize could no longer find the node. The file vanished and the draft was
left with an empty `!file[name]()`.
The editor was never remounted (instance stays alive); the node was removed by
the content-sync effect. An uploading node is local state an external sync must
not overwrite, exactly like the existing dirty/focused guards. Add a guard that
bails the sync while any `uploading` node is present.
Pure frontend; affects only the first upload in a new chat (subsequent uploads
hit an existing session, so no draft-key flip).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* test(editor): cover the in-flight-upload content-sync guard
The content-sync effect now reads `editor.state.doc.descendants` on every run
to detect uploading nodes; the mocked editor didn't implement it, crashing all
ContentEditor tests. Add `descendants` (driven by `editorState.uploadingNodes`)
to the mock and a regression test asserting an external `defaultValue` change
does not setContent while an upload is in flight, and resumes once it settles.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(chat): migrate new-chat draft onto the session id on lazy create
The first file upload in a brand-new chat lazily creates the session, flipping
ChatInput's draft key from `__new__:agent` to the session id mid-upload. The
in-progress (empty-href) file-card markdown the editor had already written into
the `__new__:agent` draft was neither migrated nor cleared, so it stayed
stranded under that key — and resurfaced as a stale `!file[name]()` the next
time a new chat opened for the same agent (the send only cleared the
session-keyed draft).
Migrate the `__new__:agent` draft onto the new session id the moment the
session is created (upload path only — text send already clears the pre-flip
key via `keyAtSend`). Add a shared `newSessionDraftKey` helper so ChatInput and
ensureSession agree on the slot name, and a `migrateInputDraft` store action.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The two prior MUL-3254 fixes preserved draft/description state across a
modal close, but Desktop still could not RENDER the reopened image: in
CloudFront signed-URL mode every URL the renderer holds after reopen is
unloadable. The persisted record strips the expired signed download_url,
the raw CDN url is unsigned (403 on a signed distribution), and the
durable /api/attachments/<id>/download endpoint needs credentials that a
cross-site file:// <img> fetch cannot carry (web works via the same-site
session cookie, which is why the bug was desktop-only).
Two changes close the last mile:
- /api/config now reports cdn_signed when CloudFront signing is enabled,
and pickInlineMediaURL stops picking the raw (unsigned) CDN url in
that mode — it is a guaranteed 403.
- The Attachment renderer upgrades an auth-gated media URL to a freshly
signed one via authenticated GET /api/attachments/<id> (the same
re-sign the click-time download path already does), but only on
clients without a same-origin /api proxy (api.getBaseUrl() non-empty:
Desktop, mobile webview). Cached via TanStack Query with a 20-minute
staleTime, inside the server's 30-minute signed-URL TTL.
Old servers omit cdn_signed; the schema defaults it to false so behavior
is unchanged there. Non-CloudFront deployments return the API path again
from the metadata fetch and the renderer keeps the original URL.
Co-authored-by: J <j@multica.ai>
Co-authored-by: multica-agent <github@multica.ai>
* fix: flush issue description editor on close
Co-authored-by: multica-agent <github@multica.ai>
* fix: make unmount flush opt-in via flushPendingOnUnmount
The unconditional unmount flush re-emitted discarded content into
composers that clear their draft and then unmount (comment edit cancel,
create-issue / feedback submit), resurrecting the cleared draft.
- Add flushPendingOnUnmount prop (default false); only the issue-detail
description editor opts in.
- Cache the pending markdown in a ref at onUpdate time and emit that
cached copy on unmount, instead of reading the editor instance during
teardown.
- Regression tests: default drops the pending update on unmount, opt-in
flush emits the cached value even when the editor is already
destroyed, no double-emit after the debounce fired, and issue-detail
pins the opt-in wiring.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: J <j@multica.ai>
Co-authored-by: multica-agent <github@multica.ai>
remark-math defaults to singleDollarTextMath: true, so any paragraph
containing two dollar amounts (e.g. "costs $120/mo (~$85 net)") has
the text between them parsed as inline TeX and rendered by KaTeX in an
italic math font, with ~ treated as a non-breaking space. Disable
single-dollar parsing in both web render paths, matching GitHub's
behavior; explicit $$...$$ math still renders.
Co-authored-by: Matt Voska <voska@users.noreply.github.com>
* fix: keep issue draft attachment records
Co-authored-by: multica-agent <github@multica.ai>
* fix: avoid persisting signed draft attachment urls
Co-authored-by: multica-agent <github@multica.ai>
* fix: reuse resolved media url for draft previews
Co-authored-by: multica-agent <github@multica.ai>
* fix: address draft attachment review nits
- Backfill an empty caller download_url from the in-session upload on id
collision so a just-pasted image first-paints from the signed URL
instead of detouring through markdown_url.
- Prune draft attachments no longer referenced by the persisted
description when the create dialog reopens.
- Backfill EMPTY_DRAFT defaults on draft-store rehydrate so drafts
persisted before the attachments field existed get a stable shape.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: J <j@multica.ai>
Co-authored-by: multica-agent <github@multica.ai>
Table.configure had renderWrapper unset (defaults to false), so tables
rendered as bare <table> elements with no .tableWrapper div. The
overflow-x: auto rule in prose.css targets .tableWrapper and never
matched, so a wide table pushed the horizontal scrollbar onto the
issue detail's page-level scroll container instead of scrolling
within the table itself.
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Comment / issue / chat images uploaded inside the Desktop app rendered
as the broken-image fallback. The editor was persisting a site-relative
`/api/attachments/<id>/download` URL into markdown — that path only
resolves when the document origin proxies /api to the API host (apps/web
via Next.js rewrite). On Electron's file:// origin it never resolved.
Per GPT-Boy's plan, move the durable-URL choice from the client to the
server so the persisted shape is correct regardless of which client
performed the upload.
Server:
- AttachmentResponse gains a markdown_url field, computed by
buildMarkdownURL from the deployment policy:
• storage URL is already absolute + unsigned (public CDN, S3 public
bucket, LocalStorage with MULTICA_LOCAL_UPLOAD_BASE_URL on https) →
use it verbatim;
• CloudFront-signed mode → never expose the raw S3 URL (private
bucket); return cfg.PublicURL + /api/attachments/<id>/download so
the server can re-sign on every request;
• LocalStorage relative + cfg.PublicURL set → same prefixed API
endpoint;
• cfg.PublicURL unset → fall back to site-relative path so web's
Next.js rewrite still works.
- isDurablePublicURL helper rejects URLs carrying CloudFront / S3
signature query params, so a freshly-signed download_url can never
leak into persistence — the original MUL-3130 bug stays closed.
Frontend:
- Attachment type + AttachmentResponseSchema (and apps/mobile mirror)
carry markdown_url. Schema lenient-defaults to '' so a backend old
enough to predate this field doesn't break clients.
- useFileUpload picks markdownLink with three-layer fallback:
(1) att.markdown_url (modern server),
(2) attachmentDownloadPath(att.id) — legacy site-relative shape,
retained for backends old enough to omit markdown_url,
(3) att.url — no-workspace avatar branch with no attachment-row id.
- attachment.tsx keeps the relative→absolute absolutize pass, but
reframed as the legacy-compat fallback for already-persisted
/api/attachments/<id>/download or /uploads/<key> URLs in old
bodies. New content writes absolute URLs and skips this path.
- ContentEditor still tracks freshly-uploaded records into
AttachmentDownloadProvider so Quick Create's editor can swap the URL
via the resolver during the same session even before the server-side
binding lands.
Tests:
- server/internal/handler/file_test.go: 5 new buildMarkdownURL matrix
tests (public CDN passthrough, CloudFront-signed swap, relative
prefixing, PublicURL unset fallback, trailing-slash strip) + 15
table-driven isDurablePublicURL cases.
- packages/core/hooks/use-file-upload.test.ts: new file, 4 cases
covering modern server / legacy server / no-id avatar / oversize.
- packages/views/editor/attachment.test.tsx + content-editor.test.tsx:
10 cases for the absolutize matrix and in-session attachment merge.
- 6 existing test fixtures updated to include markdown_url.
Verification: 1236 @multica/views tests pass; 514 @multica/core tests
pass (4 new); server handler package tests pass for the new matrix
plus all pre-existing TestAttachmentToResponse* and TestDownload*
cases. Typecheck green for views/core/web/desktop. Lint clean on
touched files.
Quick Create attachment_ids binding (orphaned attachment relationship
on the resulting issue) is a follow-up — it requires a new --attachment-id
CLI flag and daemon prompt-template work and is intentionally scoped
out of this PR.
Refs: MUL-3192
Co-authored-by: Eve <eve@multica-ai.local>
Co-authored-by: multica-agent <github@multica.ai>
* MUL-3130: persist a stable attachment download URL in comment markdown
Comment image attachments rendered as broken placeholders ~30 minutes
after upload because the editor was persisting a short-lived
HMAC-signed URL into the comment body. After PR #3903 (MUL-3132)
hardened /uploads/* with auth, `attachmentToResponse` started signing
`attachment.url` as `/uploads/<key>?exp=<unix>&sig=<HMAC>` for
LocalStorage so token-auth clients could keep loading inline images.
The signature has a 30-min TTL by design — but `useFileUpload` was
returning that signed value as `link` and the editor was writing
`` straight into the markdown, so the comment
permanently captured a URL that stopped working as soon as the
signature expired.
The fix is to persist a stable per-attachment URL that the server can
re-sign on every request:
* `useFileUpload` now returns `link = /api/attachments/<id>/download`
(avatar uploads without an id still fall back to `att.url` so the
pre-attachment-row code paths keep working).
* `DownloadAttachment` self-resolves the workspace from the attachment
row instead of reading X-Workspace-Slug / X-Workspace-ID headers,
and the route is registered under the auth-only group so a native
browser <img>/<video> resource load (which cannot attach those
headers) succeeds. Membership is checked inside the handler with
a 404 deny shape so the route does not act as an IDOR oracle.
* A new `GetAttachmentByIDOnly` SQL query supports the workspace-
derivation step.
* `AttachmentDownloadProvider` now extracts the attachment id from
the stable URL when matching markdown refs to attachment records,
with a fallback to the existing url-equality check for legacy
comments (and S3/CloudFront markdown that points straight at the
CDN).
* `contentReferencesAttachment` covers both URL shapes for the
composer / standalone-list dedup paths so an attachment uploaded
before the fix and one uploaded after both deduplicate cleanly.
Tests:
- New unit tests for the URL helpers (16 tests, packages/core).
- Backend regression test: bare `<img src>`-style request without
workspace headers now succeeds for a member (200) and 404s for a
non-member, replacing the previous "400 without workspace context"
contract.
- Existing TestDownload*, TestServeLocalUpload*, TestAttachmentTo
Response* and the 1220 frontend views tests all pass.
Refs: MUL-3130, GitHub issue #3891
Co-authored-by: multica-agent <github@multica.ai>
* MUL-3130: address PR review — split markdown link from upload link, swap render src
Two follow-ups from GPT-Boy's review on PR #3937.
(1) Don't reroute every upload consumer through the workspace-gated
download endpoint.
The previous change made `useFileUpload`'s `link` field unconditionally
return `/api/attachments/<id>/download` whenever the upload had an id.
But `useFileUpload` is also used by avatar / logo pickers
(account-tab, workspace-tab, agents/avatar-picker, squads/squad-detail-page)
that persist `result.link` directly into `avatar_url`. Avatars are
referenced cross-workspace (mention chips, member lists, inbox
items), so binding their URL to a workspace-membership-gated
endpoint would silently break cross-workspace avatar visibility.
The fix splits the URL into two semantically distinct fields:
- `link` — same as `att.url` (legacy contract). Avatar /
logo callers continue to use this and remain
on whatever URL semantics the storage backend
dictates.
- `markdownLink` — the stable per-attachment URL
`/api/attachments/<id>/download`. Only the
editor's markdown-persisting flow consumes
this. Falls back to `link` for the
no-workspace upload branch (where there is
no attachment-row id to address).
`editor/extensions/file-upload.ts` switches `image.src` and
`fileCard.href` to `markdownLink ?? link` so comment markdown gets
the stable shape while avatar callers stay on `link` unchanged.
(2) Make the render-time img src loadable for token-mode clients.
Persisting the stable `/api/attachments/<id>/download` URL fixes the
expiry problem but the path itself sits behind `middleware.Auth`,
which expects either a `multica_auth` cookie or a Bearer token in
`Authorization`. Native `<img>`/`<video>` resource loads from
token-mode clients (Electron's default mode, the mobile app,
legacy-token web sessions) cannot attach the Authorization header,
so the bare URL would 401 immediately rather than 30 minutes later.
`Attachment.normalize` now runs the resolved record through a new
`pickInlineMediaURL` helper that returns:
- `record.download_url` when it's an absolute URL with a
recognised CDN signature query (CloudFront-signed
`Signature` / `Expires` / `Key-Pair-Id`, or
`X-Amz-Signature` for raw S3 presigns) — these load as
native resource src in any client.
- else `record.url`, which on the LocalStorage backend carries
a freshly-minted `/uploads/<key>?exp&sig` query whose
signature IS the auth (token-mode-loadable). On non-CF S3
backends this is the raw stored URL — same behaviour as
today.
- else the original input URL (legacy / unresolved markdown
keeps its existing path).
This gives the same effect for both `kind: "record"` and
`kind: "url"` attachment inputs: once a record is in hand, the
rendered media src is whichever URL the current backend exposes
a working signature on.
Tests:
- New `file-upload.test.ts` regression pinning that `markdownLink`
is what lands in the markdown body when the upload result returns
both a short-lived storage URL and a stable download path.
- Updated `attachment.test.tsx` to reflect the new render-time
swap (the rendered img src now follows the freshly signed URL,
not the raw storage URL) and added a record-mode regression
pinning the LocalStorage default — when `download_url` is the
bare /api/attachments/<id>/download path, the renderer must fall
through to the signed `record.url`.
- Updated `chat-input.test.tsx` makeUpload helper for the new
`markdownLink` UploadResult field.
- 1222 frontend views tests + 507 core tests + typecheck across
@multica/{core,ui,views} all pass.
Refs: MUL-3130, GitHub issue #3891. Builds on a740f7a35.
Co-authored-by: multica-agent <github@multica.ai>
* MUL-3130: chat upload map keys on persisted markdownLink, not the short-lived link
GPT-Boy's second-round review on PR #3937 caught a chat-only blocker
left over from the previous fix.
After the previous commit split `UploadResult.link` into `link`
(legacy avatar/logo URL) and `markdownLink` (stable per-attachment
URL persisted into markdown), the comment editor's image src + file
card href correctly switched to `markdownLink ?? link`. But chat
input still kept the upload-map key on the old `link`:
uploadMapRef.current.set(result.link, result.id)
…
if (content.includes(url)) activeIds.push(id)
In the LocalStorage backend `link` is the short-lived
`/uploads/<key>?exp=&sig=` URL. The editor persists the stable
`/api/attachments/<id>/download` URL into the message body, so
`content.includes(url)` never matches and the send call drops
`attachment_ids`. The attachment ends up bound only to the chat
session, not to the message — agents reading message-level metadata
see no attachments.
Fix: key the upload map on the same value the editor actually wrote
into the markdown body (`markdownLink || link`). The
`content.includes(url)` check then matches and the attachment id is
correctly forwarded on send.
Tests:
- Updated the chat-input mock editor to insert `markdownLink || link`
into its value, mirroring the real editor's persisted-URL choice
(uploadAndInsertFile in editor/extensions/file-upload.ts). Without
this the mock would silently paper over the bug.
- Added a regression test where the upload result returns a
short-lived `link = /uploads/...?exp&sig` and a stable
`markdownLink = /api/attachments/<id>/download`. Asserts (a) the
message body carries the stable URL and never the signed query,
and (b) the bound `attachment_ids` includes the attachment id.
All 1223 frontend views tests pass (was 1222, +1 new regression).
Typecheck and 507 core tests still green.
Refs: MUL-3130, PR #3937 review by GPT-Boy. Builds on f66a522d0.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: Eve <eve@multica-ai.local>
Co-authored-by: multica-agent <github@multica.ai>
* feat(comments): skip agent triggering on /note-prefixed comments
A comment whose first token is the reserved /note prefix (case-insensitive)
is stored like any other comment but never wakes an agent. The guard sits at
the top of triggerTasksForComment, the single chokepoint, so it covers all
three trigger paths — assignee, squad leader, and @mentioned agents. Gating
only shouldEnqueueOnComment (as originally proposed) would still let
"/note @agent ..." through the mention path.
Lets members leave human-only tips/notes on agent-assigned issues without
burning an agent run. MUL-3115, closes#3649.
Co-authored-by: multica-agent <github@multica.ai>
* feat(editor): add /note built-in slash command to comment composer
Enable the `/` menu in the issue comment and reply composers in a new
"command" mode that lists fixed built-in commands instead of the chat
skill picker. Currently one command, /note, which marks a comment as a
human-only note that won't trigger the assigned agent.
Selecting it inserts the plain-text "/note " prefix (not a rich node), so
a menu pick and a hand-typed command are byte-identical and the backend
detects either with a simple prefix match. The command menu renders nothing
on a non-matching `/` (hideOnEmpty) so typing a date like 6/8 isn't noisy.
The chat skill picker is unchanged. MUL-3115.
Co-authored-by: multica-agent <github@multica.ai>
* refactor(editor): match /note by label prefix and localize its description
Address PR review feedback:
- buildBuiltinCommandItems now matches the command label as a prefix only,
dropping the description substring match copied from the skill picker. With
one command this keeps the menu predictable (/no surfaces note; /deploy or a
description word like /agent shows nothing) and avoids Enter selecting note
unexpectedly.
- The command description is now a localized UI string: added
slash_command.commands.note to all four editor locales (en/ja/ko/zh-Hans)
and the menu renders it via the typed translator. The /label itself stays
literal since it's the typed token the backend matches.
MUL-3115.
Co-authored-by: multica-agent <github@multica.ai>
* fix(editor): shorten /note command description to avoid truncation
The slash menu item is single-line (truncate, w-72), so the longer copy was
cut off. Shorten to "won't trigger any agents" across all four locales — also
more accurate, since /note skips assignee, squad leader, and @mentioned agents,
not just the assigned one.
MUL-3115.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: J <j@multica.ai>
Co-authored-by: multica-agent <github@multica.ai>
@tiptap/markdown parses via marked, whose tokenizer is O(n²) in document
length. Opening a large markdown doc (issue description, agent
instructions, …) froze the UI for tens of seconds: a 533KB plain-text doc
took 61.8s to parse while the subsequent ProseMirror setContent was only
40ms. Upgrading marked doesn't help — already on 17.0.5, whose fix only
covers `_`/`*` delimiter runs, not general prose.
Parse large markdown in chunks instead of in one shot: split on blank
lines outside fenced code blocks, parse each chunk independently, then
concatenate the resulting docs. This drops marked's cost to O(n²/k) while
producing a byte-identical document. Applied transparently at
ContentEditor's two parse entry points (mount + WS-driven re-parse), gated
at 50KB so normal small docs stay on the single-parse fast path.
533KB: parse 61.8s -> 0.95s (65x), open 100s -> 3.2s (31x).
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
navigator.clipboard is only exposed in a secure context (https or
localhost). On self-hosted instances served over plain http:// it is
undefined, so every copy / "copy all" / export button silently failed and
left the clipboard empty (GitHub #3781).
Add a shared copyText(text): Promise<boolean> helper in
@multica/ui/lib/clipboard that prefers the async Clipboard API and falls
back to a hidden <textarea> + document.execCommand('copy') for non-secure
contexts. Migrate all direct navigator.clipboard.writeText call sites
(code blocks, agent transcript copy-all, token / webhook / issue-link
copy, etc.) to it, gating success side-effects on the returned boolean,
and remove the now-redundant copyMarkdown wrapper. Secure-context users
keep the native path unchanged.
MUL-3068
Co-authored-by: J <j@multica.ai>
Co-authored-by: multica-agent <github@multica.ai>
* fix editor image upload caret placement
Co-authored-by: multica-agent <github@multica.ai>
* feat(editor): reserve image box via intrinsic dimensions to kill paste layout shift (#3803)
Capture an image's intrinsic width/height on upload and render them as
<img width height> so the browser reserves the box before the image
decodes. Removes the layout shift that pushed the caret out of view after
a pasted-image insert, making the post-insert scrollIntoView correct.
- Add width/height node attrs to ImageExtension (render-only; not
serialized to markdown, so round-trips stay clean).
- Measure dimensions off-thread via createImageBitmap and patch the node
after insert. Fire-and-forget so the synchronous-insert contract
(instant preview) is preserved; degrades to no-box when the API is
unavailable (jsdom). The src swap keeps width/height via attr spread.
- Thread width/height through ImageView -> Attachment -> <img>.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: multica-agent <github@multica.ai>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(editor): support markdown checkbox task lists (#3593)
Render `- [ ]` / `- [x]` as interactive checkboxes in the issue content
editor, matching GitHub / Notion.
- Register TaskList + a patched TaskItem in the shared extension factory.
Both ship their own markdown tokenizer / renderMarkdown, input rules, and
a checkbox NodeView; the taskList tokenizer is consulted before marked's
built-in list tokenizer, so `- [ ]` becomes a task while a plain `- ` still
falls through to the bullet list.
- Patch TaskItem's keymap to share PatchedListItem's split -> lift Enter
chain (double-Enter on an empty item exits the list); nested: true enables
sub-tasks and nested round-trips.
- Add a "Task list" entry to the bubble-menu list dropdown (+ i18n for en /
zh-Hans / ja / ko).
- Style task lists in prose.css for both the editor ([data-type="taskList"])
and the readonly remark-gfm output (.contains-task-list); completed items
render muted.
Readonly already rendered task lists via remark-gfm; this brings the editable
view to parity. Adds markdown round-trip and readonly checked-state tests.
MUL-2926
Co-authored-by: multica-agent <github@multica.ai>
* fix(editor): keep readonly nested task lists block-laid-out (#3593)
The shared `display: flex` rule on task-list items broke nested task lists in
the readonly view. remark-gfm renders a task item as
`<li><input> text <ul>…</ul></li>` — no body wrapper — so a nested list is a
direct sibling of the checkbox and text, and flex pulled it onto the same row.
The editor's Tiptap NodeView wraps the body in a `<div>`, so it was unaffected.
Split the task-list CSS into separate editor and readonly blocks: the editor
keeps the flex row; readonly stays a block list item with an inline checkbox so
a nested `<ul>` drops below and indents under its parent. Adds a readonly test
that pins the nested DOM shape (nested `<ul>` inside the parent `<li>`), so a
future remark-gfm change that wraps the body fails loudly.
MUL-2926
Co-authored-by: multica-agent <github@multica.ai>
* feat(editor): convert `- [ ] ` typing into a task list (#3593)
TaskItem's built-in input rule only converts `[ ] ` / `[x] ` typed at the start
of a plain paragraph. When the user types the GitHub-style `- [ ] ` the leading
`- ` first turns the line into a bullet, and the built-in rule no longer fires —
so `[ ]` stayed as literal text and nothing became a checkbox.
Add an input rule on PatchedTaskItem that catches the checkbox token when it is
the entire content of a freshly-typed list item (bullet or ordered) and converts
just that item into a task item (deleteRange → liftListItem → toggleList). The
anchored regex means it only fires on an item whose whole content is `[ ] ` /
`[x] `, so sibling items in the same list are left untouched.
Adds typing-level tests (real input-rule simulation) covering `[ ] `, `[x] `,
`- [ ] `, `- [x] `, the mixed-list split case, and the plain-bullet no-op.
MUL-2926
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: J <j@multica.ai>
Co-authored-by: multica-agent <github@multica.ai>
Fix attachment download for self-hosted deployments using private S3-compatible buckets without CloudFront. Closes#3721.
**Server**
- New unified `GET /api/attachments/{id}/download` endpoint that picks CloudFront / S3 presign / server proxy at request time.
- `ATTACHMENT_DOWNLOAD_MODE=auto|cloudfront|presign|proxy` and `ATTACHMENT_DOWNLOAD_URL_TTL` env knobs; `auto` routes Docker hostnames / localhost / private IPs through the proxy and public S3 endpoints through presign.
- `Storage.PresignGet` capability; S3 implementation generates presigned GET URLs.
- `attachmentToResponse` returns the unified relative endpoint instead of leaking raw unsigned S3 URLs when CloudFront is not configured. Proxy path streams via `io.Copy` with `Content-Disposition` / `Content-Length` / `Cache-Control: no-store` / `X-Content-Type-Options: nosniff`.
**Clients**
- CLI / Desktop / Mobile resolve relative `download_url` values against the configured API base. Desktop covers the Electron native download bridge and the media preview modal; Mobile covers `Linking.openURL`, the markdown image RN loader, and the composer's completed non-image file chip.
- Mobile gains a minimal Node-environment vitest lane wired into `mobile-verify.yml`.
**Docs**
- `.env.example`, `docker-compose.selfhost.yml`, `SELF_HOSTING_ADVANCED.md`, and the `environment-variables` doc set updated with the new env keys and the `ATTACHMENT_DOWNLOAD_MODE=proxy` recommendation for Docker / VPC-internal object stores.
**Tests**
- `internal/storage`, `internal/cli`, `internal/handler` (download endpoint, mode selection, proxy header, `/content` non-regression), `cmd/server` (trusted proxy parser).
- `packages/views/editor/use-download-attachment.test.tsx` and `attachment-preview-modal.test.tsx` exercise relative URL resolution + absolute pass-through.
- `apps/mobile/lib/attachment-url.test.ts` covers every helper branch plus the composer non-image chip case.
* feat(editor): support text highlight (==text==) in description & comments
Adds a single-color (yellow) text highlight mark to the shared rich-text
editor, round-tripped through stored Markdown as ==text==.
- HighlightExtension: @tiptap/extension-highlight + @tiptap/markdown hooks
(markdownTokenizer/parseMarkdown/renderMarkdown) so ==text== <-> <mark>
round-trips; inner inline formatting preserved via inlineTokens.
- Bubble menu: highlight toggle button (Mod-Shift-H), i18n in 4 locales.
- Read-only renderer: highlightToHtml lowers ==text== -> <mark> (skips code
and math); rehype-sanitize schema whitelists <mark>. Nested Markdown inside
a highlight still parses via the existing rehype-raw step.
- prose.css: single yellow <mark> style, legible in light/dark.
Pinned @tiptap/extension-highlight to exact 3.22.1 to match @tiptap/core
(>=3.23 expects a getStyleProperty export core 3.22.1 doesn't have).
Web/desktop only. Mobile (native md4c, no == syntax, no custom renderers)
is tracked as a follow-up. MUL-2934.
Tests: editor round-trip (cross-process serialization protocol), readonly
<mark> rendering + sanitize, and the ==->mark transform incl. code-skip.
Co-authored-by: multica-agent <github@multica.ai>
* fix(editor): align highlight boundary rules across editor & readonly
Addresses two boundary bugs from review (PR #3661):
1. A == inside inline code/math could close a highlight when the opening
== was outside the literal span (e.g. ==a `b==c` d== wrongly became
<mark>a `b</mark>c` d==). Both the editor tokenizer's lazy regex and the
readonly transform only guarded the opening fence, not the closing one.
2. The readonly transform matched across blank lines (==a\n\nb==) while the
editor lexes those as two literal paragraphs — a storage↔editor↔readonly
mismatch.
Fix: extract one shared matcher (utils/highlight-match.ts) used by BOTH the
editor tokenizer and the readonly lowering, so the rules can't drift. It skips
fences that fall inside code/math literal ranges (open or close) and caps the
inner span at the first blank line.
Tests: shared-matcher unit tests + both repros covered on the editor
(round-trip/HTML) and readonly (transform + rendered DOM) sides.
Co-authored-by: multica-agent <github@multica.ai>
* fix(editor): handle CRLF in highlight blank-line boundary
BLANK_LINE_RE only matched LF, so a CRLF blank line (==a\r\n\r\nb==) was not
recognized as a block boundary and got highlighted. Widen to \r?\n[ \t]*\r?\n.
Tests: CRLF blank-line (no highlight) + CRLF soft-break (still highlights) on
the matcher, readonly transform, and editor sides.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: Lambda <lambda@multica.ai>
Co-authored-by: multica-agent <github@multica.ai>
* feat(editor): add / slash-command palette for invoking agent skills
Adds a `/` trigger in the chat box that opens a popover listing the active
agent's skills. Selecting an item inserts a `[/label](slash://skill/<id>)`
token; the daemon extracts those IDs in `buildChatPrompt` and emits an
"Explicitly selected skills:" block using the canonical names from the
agent's skill registry — labels are display-only and never trusted.
Built on Tiptap's `Mention` extension so the suggestion lifecycle,
keyboard routing, and IME handling mirror the existing `@` mention UX.
Item list is sourced from the React Query workspace cache (no per-keystroke
fetch). Gated behind a new `enableSlashCommands` prop so only `chat-input`
opts in; other `ContentEditor` consumers (issue editor, comments) are
unaffected. Read-only markdown surfaces render the token as a `.slash-command`
pill via a custom link renderer + sanitize-schema/url-transform allowlists.
Closes#3108
* fix(i18n): add slash_command editor copy for ko/ja
The PR added slash_command popover empty-state keys to en + zh-Hans only;
locales/parity.test.ts requires every locale to cover every EN key, so ko
and ja failed CI. Add the two keys (no_skills_configured, no_results)
matching existing skill terminology (스킬 / スキル).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Naiyuan Qing <145280634+NevilleQingNY@users.noreply.github.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
