Two fixes for Emacs Round 2 findings:
1. Desktop TerminalPanel now sends `{type:"auth", payload:{token}}` as
the first WS frame in token mode and waits for `auth_ack`. Previously
it transitioned straight to "connected" without authing, so the
server's 10s first-frame deadline closed the connection and the
terminal never opened. Cookie-mode apps still skip the handshake.
Exposes ApiClient.getToken() so the panel can snapshot the bearer
without reaching into auth-store internals.
2. terminalBridge.handleOpen no longer treats `terminal.opened` as a
droppable control frame. sendFrame now returns bool; if the ack
can't be enqueued the bridge rolls the half-open session back
(cancel ctx, drop the route, sess.Close) so the spawned PTY doesn't
orphan until the idle sweep — which also pinned the GC active-env
mark for that window.
Regression coverage:
- TestTerminalBridge_OpenedEnqueueFailureClosesSession: drop-only
sender, spawn must happen exactly once, manager and bridge route map
must drain to zero.
- packages/views/issues/components/terminal-panel.test.tsx: asserts
the auth frame is sent in token mode and NOT sent in cookie mode.
Co-authored-by: multica-agent <github@multica.ai>
- Manager.openWith now fires OnSessionStart before sess.start(); if the
child exits before the hook ran, OnSessionStop would otherwise race
ahead and unmark an env root that was never marked, then OnSessionStart
would mark it forever. Added a regression test that uses a pre-closed
waitDone to force immediate exit; with a 20ms delay inside the start
hook it deterministically fails on the old ordering.
- terminalProxy.writePump now calls audit.RecordClose immediately on
terminal.exit, not just from run()'s defer. CloseTerminalSession is
idempotent (WHERE ended_at IS NULL) so the browser_disconnect fallback
becomes a no-op when fired later; without this, a client that keeps
the WS open after exit would leave ended_at NULL and `multica issue
runs` would render exited terminals as active.
- cli.APIClient.GetJSON / GetJSONWithHeaders now return *HTTPError
(matching PostJSON), so callers can errors.As to distinguish 404 from
other failures. runIssueRuns only swallows 404 (old server, no
terminal-sessions endpoint); 500 / auth / network failures now print
a warning to stderr instead of silently downgrading to "no terminal
rows".
Co-authored-by: multica-agent <github@multica.ai>
- terminal.Manager: OnSessionStart/OnSessionStop hooks fire around the
manager's register/deregister so callers can safely mark/unmark
external state (GC protection, audit rows) without racing Get/Done.
- daemon: wires the hooks to markActiveEnvRoot(filepath.Dir(workDir))
so an idle terminal on a done/cancelled issue can't have its workdir
reclaimed mid-session, and emits structured slog audit records
(user_id/task_id/duration; no keystrokes — RFC §Auth).
- server: persists every PTY open/close to a new terminal_sessions
table (migration 091) as the source behind the audit log and the
new `type=terminal` rows in `multica issue runs`. New endpoint
GET /api/issues/{id}/terminal-sessions.
- CLI: `multica issue runs` merges the agent task runs feed with the
terminal sessions feed, sorted by started_at; terminal rows render
with agent="terminal" and close_reason in the ERROR column. Old
servers without the endpoint degrade silently.
- server/handler/terminal_ws.go: pass parsed Cols/Rows query values
to sendOpenToDaemon so the first PTY frame matches the client's
viewport; post-open resize is now just a defensive patch (addresses
Emacs's Phase 3 non-blocking nit).
Co-authored-by: multica-agent <github@multica.ai>
Phase 3 of MUL-2295. Adds `multica issue terminal <issue-id>` which dials
the Phase 2 /ws/issues/{id}/terminal endpoint, performs first-frame auth
with the existing PAT/JWT, and runs an interactive PTY through the
daemon-side terminal manager from Phase 1. SIGWINCH on unix /
poll on windows pushes resize frames; ssh-style `<enter>~.` detaches.
Co-authored-by: multica-agent <github@multica.ai>
Address Phase 2 Round 1 review blockers + security item:
1. Fold daemonws teardown into a single cleanup defer so the order is:
cancel heartbeat → wait hbDone → clearWSWrites → close(writes) →
wait writer. The previous LIFO defer ordering let close(writes) run
before the terminal bridge tore down, so an in-flight terminal pump
could panic on send-to-closed-channel.
2. Replace silent drop of terminal.data on a saturated daemonws writer
with real backpressure: pump uses sendWSFrameCtx (blocking with ctx
escape) and bridge.closeAll now waits for every pump goroutine to
exit before returning, giving the wakeup loop a hard barrier before
close(writes).
3. terminalUpgrader now reuses realtime.CheckOrigin instead of
CheckOrigin: true. The terminal endpoint executes shells; it must be
at least as strict as the read-only realtime WS.
Tests:
- TestTerminalBridge_DataBackpressureNoSilentDrop pins that a full
writes channel never loses bytes.
- TestTerminalBridge_TeardownDoesNotPanicOnInFlightSend mirrors the
wakeup defer sequence (closeAll → close(writes)) and asserts no panic.
Co-authored-by: multica-agent <github@multica.ai>
Wires the Phase 1 terminal.Manager into the live daemonws transport so a
browser tab on the Multica Desktop app can attach to a PTY running in
the daemon's task workdir. End-to-end frame path:
Desktop (xterm.js) → /ws/issues/{id}/terminal (server proxy, cookie
or first-frame JWT auth, workspace membership
enforced before upgrade)
→ daemonws hub (new SendToRuntime + TerminalRouter routing
terminal.* frames back to the proxy by
request_id, then re-keyed on session_id)
→ daemon (new terminal_bridge.go owns one Manager per WS
connection, drains PtySession.Output() into
terminal.data frames, surfaces ExitC as
terminal.exit; closeAll on WS disconnect)
→ terminal.Manager.OpenWithInfo (new method) — server resolves
task.work_dir/issue_id/prior_session_id from its DB and embeds
them on the protocol; daemon trusts the server payload, no
daemon-local task cache needed.
Auth + ACL:
- Browser proxy enforces workspace membership before the upgrade.
- TerminalOpenPayload carries the resolved task info; cross-workspace
is structurally impossible at the bridge layer because both
OpenParams.WorkspaceID and TaskInfo.WorkspaceID come from the same
server-resolved field.
- terminal_bridge maps terminal.{Manager.OpenWithInfo} errors to the
protocol error codes (workspace_mismatch / task_not_found /
unsupported_os / spawn_failed / internal).
Resume:
- Server passes prior_session_id; daemon injects CLAUDE_SESSION_ID +
MULTICA_{WORKSPACE,ISSUE,TASK,USER}_ID into the PTY env per the RFC.
`claude --resume \$CLAUDE_SESSION_ID` continues the agent's session.
Lifecycle:
- Daemon installs a fresh terminalBridge per daemonws connection and
tears every PtySession down on disconnect; session_ids minted on one
WS cannot be reused on a reconnect because the server-side routing
registration is also gone.
- daemonws client.send buffer raised from 16 to 256 and read limit
from 4KB to 64KB so PTY traffic fits without evicting connections
used for heartbeat / wakeup hints.
Desktop UI:
- packages/views/issues/components/terminal-panel.tsx renders an
xterm.js console with FitAddon, base64 wire encoding, ResizeObserver
→ terminal.resize, reconnect button, and a clear web-only placeholder
when window.desktopAPI is absent.
- TerminalPanelSection wrapper hangs collapsed in the issue-detail
sidebar next to the execution log so bootstrap doesn't run for
every issue view.
Tests:
- terminal/manager_test.go: OpenWithInfo happy path + cross-workspace
reject (16 tests total, all -race clean).
- daemonws/terminal_test.go: TerminalRouter request_id → session_id
re-keying and unknown-session drop.
- daemon/terminal_bridge_test.go: server-supplied workdir round-trips
through OpenWithInfo, missing-workdir surfaces task_not_found and
never spawns, data+exit round trip.
- GOOS=windows go test -c clean for both daemon and daemon/terminal.
Phase 1 / 2 / 3 / 4 mapping unchanged: Phase 3 (CLI) and Phase 4
(execenv GC hook + issue runs entry + audit log) are still untouched.
Co-authored-by: multica-agent <github@multica.ai>
- Manager.Close concurrent re-entry now blocks late callers on closeDone
so every Close() return shares the "manager drained" guarantee.
- Open cleanup path on lost race with Close calls pty.Wait() to reap the
child synchronously (waitLoop never runs there).
- Tests: concurrent Close callers all observe drained state; Open cleanup
invokes pty.Wait at least once.
Co-authored-by: multica-agent <github@multica.ai>
Round 2 review fixes:
1. PtySession finalize sequence is now
ExitC -> close(output) -> onClose/deregister -> close(done)
so external waiters (bridge / GC hook / audit) can `<-Done()` and
immediately query the manager without a race window.
2. Manager.Close now waits for each session's Done() (not just Close())
so by the time it returns the registry is empty and every session
is fully finalized.
Adds TestSession_DoneFiresAfterDeregister (locks the ordering contract)
and TestManager_CloseWaitsForSessionFinalize (fakePTY.Wait delay proves
Manager.Close blocks through finalize).
Co-authored-by: multica-agent <github@multica.ai>
Wires in the four fixes Emacs flagged on the Phase 1 review:
1. Lifecycle: split stop/done with a WaitGroup. readLoop and idleLoop
exit via <-stop; waitLoop is the finalizer that waits on the WG
before closing output/done. Eliminates the "send on closed channel"
race when the output buffer is saturated. Adds a regression test
that fills output, calls Close, and verifies Done converges + ExitC
fires before output closes (the doc contract).
2. Errors: Manager.Open wraps spawner errors with double-%w so
errors.Is matches both ErrSpawnFailed and ErrUnsupportedOS. Adds a
test with a fake spawner that returns ErrUnsupportedOS.
3. Close path on unix: SIGHUP to the process group, 250ms grace,
SIGKILL, then close fd — comment now matches behavior. Skips the
signal+sleep work entirely when the child already exited naturally.
Manager.Close fans out per-session Close in parallel so the grace
period doesn't multiply by session count.
4. IdleTimeout semantics: removes the NewManager default that
silently rewrote 0 to 60min. Zero/negative now disables, per the
doc comment. Added DefaultIdleTimeout for daemon wiring to opt in
explicitly.
Verified: go test, go test -race, GOOS=windows go test -c.
Co-authored-by: multica-agent <github@multica.ai>
Daemon-side foundation for the Issue → Terminal feature. Manager owns
the lifecycle of all live PtySessions; sessions spawn a shell on a real
PTY via creack/pty (unix-only — Windows returns ErrUnsupportedOS until
ConPty support lands).
Open enforces the cross-workspace ACL — a client acting in workspace A
cannot attach to a task that belongs to workspace B. Each session
injects CLAUDE_SESSION_ID + MULTICA_{WORKSPACE,ISSUE,TASK,USER}_ID into
the child env so `claude --resume $CLAUDE_SESSION_ID` continues the
same session the agent run was using.
Adds the terminal.* WebSocket message types to server/pkg/protocol so
Phase 2 (daemonws routing) and Phase 3 (CLI) can land without touching
the manager.
Tests cover open, data round-trip, resize, explicit close, idle timeout
sweep, manager shutdown, cross-workspace rejection, and unknown task.
A fake Spawner backed by channels lets tests exercise lifecycle without
forking a real shell.
Co-authored-by: multica-agent <github@multica.ai>
Before: dispatchCreateIssue copied autopilot.created_by_type/id onto the
new issue's creator_type/creator_id, and the same fields were used as the
ActorType/ActorID of the issue:created event. Result: any issue spawned by
an autopilot was reported as created by the human who first configured
the autopilot, not by the agent that actually owns the work. Downstream
subscriber/activity/notification listeners inherited the same wrong actor.
After: creator and actor are both the autopilot's assignee agent
(creator_type=agent, creator_id=ap.assignee_id). The human owner is still
recoverable via origin_type=autopilot + origin_id.
Audited the other ap.created_by_* usages: analytics attribution
(autopilotActorID, task.go user-id), and the private-agent visibility
gate in shouldSkipDispatch — all correctly read the autopilot's owner,
not the executor, so they stay as-is.
Co-authored-by: multica-agent <github@multica.ai>
Extends the workspace /usage page Daily tokens chart toggle from
Tokens | Cost to Tokens | Cost | Time | Tasks, so users see daily
run-time and task-count trends alongside spend without leaving the page.
- New SQL `ListDashboardRunTimeDaily`: per-date totals from
agent_task_queue (terminal tasks only), scoped to workspace and
optionally project. Same time anchor as ListDashboardAgentRunTime
so day boundaries line up.
- New handler GET /api/dashboard/runtime/daily + TanStack Query option.
- New DailyTimeChart (single-series, smart h/m/s unit) and
DailyTasksChart (completed + failed stacked).
- Empty-state is per-metric so a workspace with tokens but no terminal
runs (or vice-versa) doesn't get a false "no data".
- i18n: en + zh-Hans daily.metric_time / metric_tasks + titles.
Co-authored-by: multica-agent <github@multica.ai>
* feat(daemon): force-stop hung agent runs via idle watchdog (MUL-2281)
A backend whose subprocess hangs on a stuck child process (e.g. claude
blocked on `docker ps` against a frozen dockerd) keeps the daemon's run
record at status="running" until the full DefaultAgentTimeout (2 h)
expires, because cmd.Wait() never returns and Session.Result is never
written. MUL-2225 spent 17+ minutes in this state in the wild.
Add a per-task idle watchdog around executeAndDrain:
- Wrap the caller's ctx so a single cancel propagates to the agent
subprocess (via the ctx passed to backend.Execute) AND the drain loop.
- Stamp lastActivityAt every time the drain loop receives a message.
- Tick at window/2; when idle_for >= window AND session.Messages buffer
is empty, set a fired flag and call cancel.
- Tag the resulting Result.Status as "idle_watchdog" so runTask routes
it through a dedicated failure_reason instead of "agent_error".
Default window is 5 min, configurable via MULTICA_AGENT_IDLE_WATCHDOG;
set to 0 to disable. Tests cover the activity-then-silence case, the
zero-message case, the disabled case, and the happy path.
Co-authored-by: multica-agent <github@multica.ai>
* fix(daemon): skip idle watchdog while a tool call is in flight
A legitimate long-running tool call (npm install, docker build, test
suite) can sit silent between tool_use and tool_result for many minutes.
Without this gate, the watchdog would yank the agent mid-build.
Track unmatched tool_use messages in an atomic counter; only let the
watchdog fire when the counter is zero. tool_result clamps non-negative
so a stray result with no matching use can't re-arm the watchdog one
call too early.
Adds two regression tests:
- DoesNotFireDuringInFlightToolCall: tool_use -> silence past
window -> tool_result -> completed (must NOT fire)
- FiresAfterToolResultIfBackendStaysSilent: tool_use -> tool_result
-> silence past window (MUST fire — backend really is stuck)
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
* feat(daemon): auto-update CLI when idle (MUL-2100)
Add a periodic poller that checks GitHub for a newer multica release
every hour and self-updates when the daemon is idle, reusing the same
brew-or-download upgrade path the Runtimes-page "Update" button already
runs.
- Refactor handleUpdate to call a shared runUpdate(target) helper so
both server-triggered and auto-triggered upgrades go through the same
brew detection + atomic replace + restart.
- New autoUpdateLoop gates each tick on: opt-out flag, Desktop launch
source, dev-build version, an in-flight update, and active tasks. The
idle gate guarantees we never interrupt a running agent — busy ticks
silently retry at the next interval.
- Config: MULTICA_DAEMON_AUTO_UPDATE=false to disable (also via
--no-auto-update), MULTICA_DAEMON_AUTO_UPDATE_INTERVAL to retune the
poll period.
- IsNewerVersion / IsReleaseVersion helpers in the cli package, with
tests covering patch/minor/major bumps, dev-describe strings, and
malformed input.
- Daemon-side tests cover every skip path (updating, active tasks,
fetch failure, no-newer) plus the success path that fires
triggerRestart while keeping the updating flag held to the end.
Co-authored-by: multica-agent <github@multica.ai>
* fix(daemon): close idle race + verify checksum in auto-update (MUL-2100)
Two issues raised in PR #2679 review:
1. The first idle check in tryAutoUpdate only ran before the release-metadata
fetch, so a poller that won the claim race during the fetch could end up
handing handleTask a task that triggerRestart was about to cancel via root-
ctx cancellation. Add a strict claim barrier: runRuntimePoller now
tryEnterClaim()s before ClaimTask, and tryAutoUpdate flips pauseClaims
under claimMu only after observing claimsInFlight + activeTasks == 0.
Pollers that were already mid-claim hold claimsInFlight > 0, so the barrier
refuses to engage and the update defers to the next tick.
2. The direct-download path replaced the running binary with whatever bytes
GitHub returned, without checking checksums.txt. Pull the manifest first,
buffer the archive, and reject on SHA-256 mismatch before extraction. The
GoReleaser config already publishes checksums.txt; we just consume it.
Also tighten parseReleaseVersion so it stops accepting dev-describe shapes
like "v0.1.13-5-gabcdef0" through the patch trim, matching its docstring.
The auto-update loop already guards on IsReleaseVersion, but the lenient
parser was a footgun and the existing test name even said "not newer" while
asserting the opposite.
Tests:
- TestTryAutoUpdate_DefersWhenClaimInFlightAtBarrier (new race coverage)
- TestTryAutoUpdate_HoldsBarrierAcrossRestart / ReleasesBarrierOnUpgradeFailure
- TestTryEnterClaim_RespectsBarrier
- TestFindChecksumManifestAsset / TestParseChecksumManifest / TestVerifyAssetSHA256
- TestIsNewerVersion: dev-describe cases now expect false (matches docstring)
Co-authored-by: multica-agent <github@multica.ai>
* chore(daemon): default auto-update poll interval to 6h (MUL-2100)
1h was overly chatty for a release that lands at most a few times a week.
Operators who want a different cadence can still set
MULTICA_DAEMON_AUTO_UPDATE_INTERVAL or --auto-update-interval.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
* docs(squad): address plan-review feedback for archive + role plan
Resolve the 4 items the reviewer raised on MUL-2265:
1. TS schema: declare `active_issue_count` as optional (`number | null | undefined`)
so list/create/update Squad responses don't lie about their shape; only
`getSquad` parses through SquadSchema.
2. Archive semantics: restrict TransferSquadAssignees to active issues
(status NOT IN done, cancelled) so dialog count and SQL operate on one set
and terminal-state issues keep their historical assignee.
3. Index assumption: corrected — `idx_issue_assignee (assignee_type,
assignee_id)` exists and is sufficient at realistic squad cardinality;
no new index needed.
4. Fixed `*int64` test comparison and added `.loose()` to SquadSchema per
the local schemas.ts convention.
Co-authored-by: multica-agent <github@multica.ai>
* docs(squad): plan v3 — revert to count-all/transfer-all on archive
Reviewer round 2 surfaced two structural problems with plan v2's
active-only carve-out:
1. useActorName resolves squad names via ListSquads, which filters
archived_at IS NULL. A closed issue with an archived-squad assignee
would render as "Unknown Squad".
2. The status-only update path in UpdateIssue skips validateAssigneePair,
so a done/cancelled issue with an archived-squad assignee could be
reopened to in_progress, violating the "no active issue on an archived
squad" invariant enforced elsewhere.
Both problems disappear by reverting to count-all + transfer-all: after
ArchiveSquad runs, no issue points at the archived squad, so neither
case can occur. The product trade-off is that closed historical issues
now show the leader agent instead of the archived squad in their
"Assigned to" badge — consistent with existing agent-level reassignment
behavior elsewhere in the product.
Field rename: active_issue_count -> issue_count.
TransferSquadAssignees SQL is unchanged (already transfers all).
Co-authored-by: multica-agent <github@multica.ai>
* docs(squad): add Task 2b — wrap DeleteSquad transfer + archive in one tx
Reviewer round-3 flagged that the v3 invariant ("after archive no
issue points to the squad") was asserted on the happy path only.
DeleteSquad's current best-effort impl breaks it two ways:
- transfer failure → slog.Warn but archive proceeds (Unknown Squad,
reopen-into-archived-squad bugs reappear)
- archive failure after a committed transfer → 500 with squad still
active but emptied
Task 2b rewrites DeleteSquad to run TransferSquadAssignees +
ArchiveSquad inside one pgx tx, mirroring the project.go:266-314
pattern. Publish moves below Commit. Adds two regression tests that
lock both partial-write failure modes.
Co-authored-by: multica-agent <github@multica.ai>
* feat(squad): replace native confirm() with AlertDialog and rewrite role editor as combobox
Backend:
- Add CountIssuesForSquad sqlc query (counts every issue assigned to a squad,
no status filter — matches the existing transfer-all archive semantics).
- Extend SquadResponse with optional `issue_count` (`*int64` + omitempty,
populated only by GetSquad to avoid an N+1 in the list endpoint).
- Wrap DeleteSquad's transfer + archive in a single pgx transaction so the
v3 invariant ("after archive, no issue points to the squad") is durable
rather than best-effort. Promote slog.Warn to slog.Error and check the
parseUUIDOrBadRequest ok flag (silent zero-UUID was a #1661-class latent
bug). Publish only after Commit so realtime never sees rolled-back state.
- Tests cover happy path (count, transfer-all including terminal statuses)
and both rollback directions (transfer fail / archive fail) via a
fault-injecting tx wrapper.
Frontend:
- Extend Squad TS type with `issue_count?: number | null` (optional —
list/create/update legitimately omit it). Add SquadSchema with `.loose()`
and wrap getSquad with parseWithFallback so older servers and count-error
responses degrade to the dialog's "no count" copy variant.
- Replace `window.confirm()` with shadcn `ArchiveSquadConfirmDialog`
(destructive variant, leader name + count + closed-issue caveat in the
copy, Loader2 while pending). i18n keys added under squads.archive_dialog.
- Rewrite RoleEditor as a Popover + Command combobox: Pencil affordance is
always visible, suggestions aggregate other members' roles, commit only
on Enter or selecting a suggestion (blur discards), per-member savingId
drives Loader2 so the spinner only renders on the row being saved.
Co-authored-by: multica-agent <github@multica.ai>
* fix(squad): discard RoleEditor draft on close and no-op blank Enter
Two reviewer findings on e0d754bf:
1. Closing the Popover (outside click, Esc, trigger re-click) left `query`
in state, so reopening + Enter would commit the stale draft. Clear
`query` on every non-saving close path.
2. With an existing role, opening the editor and pressing Enter on an
empty input committed "" — `commit` only no-op'd when trimmed matched
value. Treat blank Enter as a no-op; clearing a role would need an
explicit clear action that doesn't exist yet.
Add two regression tests:
- close (via outside click) → reopen surfaces a clean input; Enter does
not commit the stale draft
- blank Enter on an existing role does not call onSave
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
* fix(squad): add explicit Clear button to RoleEditor
Role is optional, but the previous fix turned blank Enter into a no-op
without exposing any other way to clear an existing role — that broke a
valid terminal state. Keep blank Enter as no-op; add a "Clear role"
button at the bottom of the popover that only renders when value is
non-empty and routes through onSave("").
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
* feat(task): wire claim lease queries into TaskService and sweeper (MUL-2246)
- ClaimTask now uses ClaimAgentTaskWithLease (generates claim_token + lease)
- StartTask accepts optional claim_token for token-verified start
- AgentTaskResponse includes claim_token for daemon to use
- Daemon client sends claim_token in StartTask body
- Sweeper calls RequeueExpiredClaimLeases each tick
- Legacy daemons without claim_token still work (graceful fallback)
Co-authored-by: multica-agent <github@multica.ai>
* fix(task): address PR #2662 review blockers (MUL-2246)
1. ClaimAgentTaskForRuntime: push runtime_id into atomic SQL WHERE clause
so runtime A cannot claim tasks queued for runtime B under the same agent.
2. Legacy StartAgentTask: add claim_token IS NULL guard so leased rows
cannot be started without token verification. Handler rejects malformed
tokens with 400 instead of silently degrading to legacy path.
3. StartAgentTaskWithClaimToken: validate claim_expires_at >= now(),
preserve claim_token until terminal state (only clear claim_expires_at),
use CTE + UNION ALL for idempotent retry when daemon resends after a
lost StartTask response. Return 409 Conflict on token mismatch/expiry.
Co-authored-by: multica-agent <github@multica.ai>
* fix(daemon): StartTask 409 handling, transport retry, claim_token on FailTask (MUL-2246)
- StartTask 409 (claim superseded): release slot, don't call FailTask
- StartTask transport timeout/5xx: retry once with same token, then
check task status before failing
- FailTask now sends claim_token; server-side FailAgentTask SQL adds
AND (claim_token IS NULL OR claim_token = @claim_token) guard so
stale daemons cannot fail tasks that have been re-claimed
Co-authored-by: multica-agent <github@multica.ai>
* fix(task): close FailTask token bypass and RequeueExpiredClaimLeases liveness gap (MUL-2246)
Blocker 1 - FailTask token validation:
- SQL: change (param IS NULL OR claim_token = param) to
(param IS NULL AND claim_token IS NULL) OR claim_token = param
so tokenless requests can only fail legacy (tokenless) rows.
- task.go: malformed claim_token now returns ErrInvalidClaimToken (400)
instead of being silently dropped to NULL.
- Handler: maps ErrInvalidClaimToken→400, ErrClaimTokenInvalid→409.
- Service: when UPDATE returns no rows but task is still active,
return ErrClaimTokenInvalid (token mismatch) instead of silent success.
Blocker 2 - RequeueExpiredClaimLeases runtime liveness:
- SQL: JOIN agent_runtime, only requeue tasks where runtime is 'online'.
Dead/offline runtime tasks stay dispatched for FailTasksForOfflineRuntimes.
- FOR UPDATE → FOR UPDATE OF atq (required with JOIN).
Regression tests:
- task_claim_token_test.go: malformed, tokenless-on-tokened, wrong-token
- requeue_lease_test.go: SQL must JOIN agent_runtime with online filter
Co-authored-by: multica-agent <github@multica.ai>
* fix(task): move expired lease requeue to ClaimTaskForRuntime preflight, add heartbeat freshness backstop (MUL-2246)
- Add RequeueExpiredClaimLeasesForRuntime: per-runtime preflight self-requeue
in ClaimTaskForRuntime. Runtime proves liveness by actively claiming, so no
heartbeat check needed.
- Update global RequeueExpiredClaimLeases to require ar.last_seen_at freshness
(stale_threshold_secs param). Prevents requeuing to a dead runtime in the
90s gap between lease expiry (60s) and offline detection (150s).
- Add regression tests verifying the heartbeat freshness check and that the
preflight query does not join agent_runtime.
Co-authored-by: multica-agent <github@multica.ai>
* fix(task): use LivenessStore for global requeue, move preflight before empty-cache (MUL-2246)
Blocker 1: Global RequeueExpiredClaimLeases now uses LivenessStore.IsAliveBatch
to verify runtimes are truly alive before requeuing expired leases. When
LivenessStore is unavailable (no Redis), global requeue is skipped entirely —
the preflight self-requeue in ClaimTaskForRuntime handles live runtimes. This
closes the 60-150s gap where a dead runtime still appears online in DB.
Blocker 2: Moved RequeueExpiredClaimLeasesForRuntime BEFORE EmptyClaim.IsEmpty
fast-path in ClaimTaskForRuntime. Expired leases are now requeued (which bumps
the empty cache via notifyTaskAvailable) before the empty check can
short-circuit the claim path.
Also adds ListRuntimesWithExpiredClaimLeases SQL query and LivenessChecker
interface on TaskService.
Co-authored-by: multica-agent <github@multica.ai>
* fix(task): wire EmptyClaimCache into backend taskSvc for backstop requeue (MUL-2246)
The backend taskSvc used by the sweeper only had Liveness wired but not
EmptyClaim. When global backstop requeue called notifyTaskAvailable,
s.EmptyClaim.Bump() was a nil no-op — the handler's empty-cache was never
invalidated, so the daemon's next claim hit a stale empty verdict.
Fix: wire the same Redis-backed EmptyClaimCache into the backend taskSvc
in main.go (same Redis keys as router.go:139 handler instance).
Add regression test verifying backstop requeue invalidates the handler's
empty-cache.
Co-authored-by: multica-agent <github@multica.ai>
* fix(task): global backstop must not requeue — alive runtimes use preflight, dead stay dispatched (MUL-2246)
- RequeueExpiredClaimLeases is now a no-op (returns 0 always)
- Alive runtimes self-requeue via ClaimTaskForRuntime preflight
- Dead runtimes stay dispatched for FailTasksForOfflineRuntimes
- Rewriting to queued on dead runtime creates 2h blackhole (offline
sweeper only handles dispatched/running)
- Test actually calls RequeueExpiredClaimLeases and asserts 0 in all cases
Co-authored-by: multica-agent <github@multica.ai>
* fix(daemon): remove duplicate usage reporting block after merge conflict (MUL-2246)
The merge resolution introduced a second ReportTaskUsage call after the
status check, duplicating the usage-before-early-return block that already
runs right after runner.run. Remove the duplicate and add a regression test
asserting /usage is called exactly once on the normal completion path.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
Add claim_token + claim_expires_at columns to agent_task_queue and three
new SQL queries for the claim lease protocol:
- ClaimAgentTaskWithLease: generates a UUID token and sets a lease expiry
when claiming a task, so the daemon must prove it received the response
- StartAgentTaskWithClaimToken: validates the token on StartTask, preventing
stale daemons from starting requeued tasks
- RequeueExpiredClaimLeases: moves dispatched tasks with expired leases back
to queued for re-claim
This closes the reliability gap where a claim response lost in transit
leaves a task stuck in dispatched until the 60s dispatch timeout fires.
Co-authored-by: multica-agent <github@multica.ai>
* docs(email): clarify 888888 is opt-in via MULTICA_DEV_VERIFICATION_CODE; document SMTP option in self-host docs
The startup log line, .env.example, and SELF_HOSTING_ADVANCED.md still
implied that the dev master code 888888 is auto-active whenever
APP_ENV != "production". That has not been true since the master code
was gated behind MULTICA_DEV_VERIFICATION_CODE — the fixed code is
disabled by default and must be opted in explicitly.
Also extend the docs site with the SMTP relay backend added in #1877:
auth-setup, environment-variables, and self-host-quickstart now cover
both Resend and SMTP options in EN and ZH.
Co-authored-by: multica-agent <github@multica.ai>
* docs(email): treat SMTP as an email backend in self-host docs and startup warning
Address review feedback on #2666:
- server: startup warning now fires only when both RESEND_API_KEY and SMTP_HOST
are empty, since either one is a valid email backend. Otherwise the log
mis-tells SMTP-only operators that verification codes go to stdout.
- self-host-quickstart (EN/ZH): tell readers to fetch the verification code
from whichever backend they configured (Resend or SMTP); fall back to
stdout only when neither is configured.
- auth-setup (EN/ZH): \"without Resend\" → \"without any email backend
configured\" so the wording stays correct now that SMTP is a first-class
option.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
* fix(realtime): include actor_type in WebSocket broadcast messages
The WS broadcast message format was {type, payload, actor_id} but missing
actor_type. This meant the web UI could not distinguish agent from human
operations in real-time events at the top level.
While payload data for comments (author_type) and activities (entry.actor_type)
already included the type, the top-level message did not — causing the web UI
to display agent CLI operations as human operations when relying on the
broadcast actor identity.
Changes:
- server/cmd/server/listeners.go: add actor_type to all broadcast messages
- packages/core/types/events.ts: add actor_type to WSMessage interface
- packages/core/api/ws-client.ts: pass actor_type to event handlers
- packages/core/realtime/hooks.ts: update EventHandler type signature
- packages/core/realtime/provider.tsx: update EventHandler type signature
Fixes MUL-2260
Co-authored-by: multica-agent <github@multica.ai>
* test: add frame-shape unit test asserting actor_type in WS frames
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
* feat(email): add SMTP relay as alternative to Resend
Self-hosted deployments often run behind a corporate firewall with an
existing SMTP relay (Exchange, Postfix, sendmail) and no access to
external SaaS APIs. Resend requires a public domain, an API key, and
outbound HTTPS to api.resend.com — all unavailable in air-gapped or
private-network setups.
This adds a second email delivery path using Go's stdlib net/smtp,
activated when SMTP_HOST is set. Priority order:
1. SMTP relay (SMTP_HOST set)
2. Resend API (RESEND_API_KEY set)
3. DEV stdout (neither set)
New env vars (all optional, no breaking change):
SMTP_HOST — SMTP server hostname
SMTP_PORT — port, default 25
SMTP_USERNAME — for authenticated SMTP; empty = unauthenticated relay
SMTP_PASSWORD — used only when SMTP_USERNAME is set
SMTP_TLS_INSECURE — set to "true" to skip TLS cert verification
(for private CA / self-signed certs)
The implementation:
- Dials TCP, creates smtp.Client manually (avoids smtp.SendMail which
does not expose TLS config)
- Tries STARTTLS if advertised; uses InsecureSkipVerify only when
SMTP_TLS_INSECURE=true (opt-in, nolint:gosec annotated)
- Applies PlainAuth only when SMTP_USERNAME is non-empty
- Wraps all errors with context for easier debugging
- Reuses existing HTML templates from buildInvitationParams for
invitation emails (no template duplication)
Also updates .env.example and docker-compose.selfhost.yml with the
new variables and inline documentation.
* fix(email): add dial timeout, session deadline, RFC headers for SMTP path
Address review blockers from multica-eve and Bohan-J (PR #1877):
- net.Dial → net.DialTimeout(10s) + conn.SetDeadline(30s) so a blackholed
SMTP relay cannot hang SendVerificationCode (called synchronously from the
auth handler) or leak goroutines in the invitation path.
- Add Date, Message-ID, and proper Content-Transfer-Encoding headers.
Date is required by RFC 5322; many strict relays reject messages without it.
Message-ID aids deliverability and threading.
- MIME-encode Subject via mime.QEncoding so non-ASCII workspace/inviter names
(CJK, emoji) survive without corruption across any RFC 2047-conformant relay.
- Probe 8BITMIME after (possible) STARTTLS: use Content-Transfer-Encoding 8bit
when the relay advertises 8BITMIME, quoted-printable otherwise — safe for
all relay configurations without forcing base64 overhead.
- Update SELF_HOSTING_ADVANCED.md to document Option B (SMTP relay) alongside
the existing Resend section, including all five env vars and a note that
port 465/SMTPS is not yet supported.
* fix(email): correct has8Bit assignment order (bool is first return of Extension)
handleTask had two early-return paths that ran before ReportTaskUsage:
the cancelledByPoll select and the post-run GetTaskStatus check. Both
silently discarded any usage accumulated by the agent — and both
claude.go and codex.go populate Result.Usage even when runCtx is
cancelled mid-run, so cancelled tasks consistently under-reported tokens.
Hoist ReportTaskUsage to run immediately after the runner returns,
before any early-return path. Add a taskRunner interface seam and a
cancelPollInterval field so tests can inject a fake runner and trigger
the poll-cancellation path on a 10ms ticker without spawning real agents.
Two regression tests cover both leak windows:
- TestHandleTask_ReportsUsageBeforeCancel: post-run /status returns
"cancelled"; usage must be reported before the status check.
- TestHandleTask_ReportsUsageWhenCancelledByPoll: poll goroutine fires
first and cancels runCtx; runner returns usage on Done; assert
poll-status precedes usage (proving the cancelledByPoll branch was
the one exercised, not the post-run path).
Sanity-checked: reverting only the ReportTaskUsage hoist fails both
tests with the original "tokens lost" message.
MUL-2258
Co-authored-by: Jiang Bohan <bhjiang@outlook.com>
Co-authored-by: multica-agent <github@multica.ai>
Without the full [@Name](mention://<type>/<UUID>) syntax, the platform
does not trigger the target agent. Add an explicit, strongly-worded
hard rule at the top of the list so the leader model never forgets.
Co-authored-by: multica-agent <github@multica.ai>
* feat(squad): accept avatar_url on CreateSquad
Threads avatar_url through the SQL query, sqlc-generated code, and the Go
handler so the create-squad flow can persist an avatar at creation time
instead of forcing a follow-up PATCH.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
* feat(squad): add avatar_url to CreateSquadRequest
Extends the TS contract for the new backend field so the frontend can pass
an uploaded avatar URL through api.createSquad.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
* feat(squads): rework Create Squad modal to match CreateAgentDialog (MUL-2233)
Replaces the cramped small-dialog flow with the same large-dialog shape used
by Create Agent: identity row (AvatarPicker + name + description with char
counter), grouped Leader picker (My Agents first, then Workspace Agents),
and a new multi-select Additional Members picker covering agents and
workspace members. The members trigger collapses to "+N" once more than
three are selected; promoting an agent to leader auto-drops it from the
additional-members list.
After createSquad, additional members are attached via Promise.allSettled
so a single failure surfaces a warning toast without blocking navigation —
the squad still exists and the user can retry from the Members tab.
Adds packages/views/modals/create-squad.test.tsx covering identity binding,
leader-group ordering, leader/member conflict sanitization, the empty- and
partial-failure success paths, and the create-failure recovery path.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
* fix(squads): valid trigger HTML + drop conflicted leader from members
Two issues from PR #2645 review:
1. AdditionalMembersPicker's PopoverTrigger was a <button> containing
MemberChip's remove <button>, which React/HTML flags as nested
interactive content (hydration + a11y warning). Render the trigger as
a <div role="combobox"> via Base UI's render prop so the chip's
remove button is valid.
2. sanitizedMembers only hid the leader from rendered/submitted output,
so promoting an additional member to leader then switching leader
away resurrected the hidden pick. Drop it from selectedMembers at
the moment of promotion via handleLeaderChange; sanitizedMembers is
no longer needed.
Adds a test that promotes → switches leader and asserts the member is
not resubmitted.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
* fix(projects): accept SSH repo URLs for github_repo resources (#2484)
The project resource validator rejected anything that wasn't http(s), so
workspace repos configured with an SSH remote (ssh:// or the scp-like
`git@host:owner/repo.git` shorthand) could not be attached to a project.
Both forms are valid git remotes and the daemon hands the URL straight to
`git clone`, so the API has no reason to require https specifically.
Relax the validator to accept http/https/ssh/git schemes and the scp-like
shorthand, while still rejecting pasted garbage (no scheme, missing host,
missing path, ftp://, file://, etc.).
Co-authored-by: multica-agent <github@multica.ai>
* fix(projects): reject scp-like URLs with '@' after ':' to avoid panic
isValidGitRepoURL indexed '@' and ':' independently, then sliced
s[at+1 : colon]. For inputs without '://' where '@' appears after the
first ':' (e.g. `host:org/repo@branch`), `at+1 > colon` triggered a
slice-bounds panic instead of a 400. Guard the slice and treat such
inputs as malformed.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
* fix(daemon): disable Claude AskUserQuestion in non-interactive mode (MUL-2244)
GitHub #2588: when Claude Code calls its built-in AskUserQuestion tool
inside the daemon's stream-json runtime, the question never reaches the
user — there's no UI to render it — so the SDK returns an empty answer
and the agent silently "infers" and continues. From the issue's
perspective, execution looks stuck while the agent is actually charging
ahead on its own guess.
Two-part fix:
- `buildClaudeArgs` now passes `--disallowedTools AskUserQuestion` so
the tool is not exposed to the model at all.
- The Claude-specific runtime brief tells the agent to use a `blocked`
issue comment for genuine clarification, or to state an explicit
assumption and proceed.
Adds a regression test that pins both: AskUserQuestion is forbidden in
CLAUDE.md and is NOT mentioned in the AGENTS.md emitted for non-Claude
providers (the tool is Claude-specific).
Co-authored-by: multica-agent <github@multica.ai>
* refactor(daemon): drop CLAUDE.md AskUserQuestion guidance, rely on --disallowedTools
The --disallowedTools flag already prevents Claude from invoking
AskUserQuestion, so duplicating the rule in the runtime brief just bloats
the prompt without changing behavior. Removes the section and its
regression test; the argv-level test in pkg/agent already pins the flag.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
* feat(execenv): native OpenClaw skill discovery via per-task config
MUL-2213 stopped lying about native discovery and routed openclaw skills
to .agent_context/skills/ — a path openclaw's scanner never reads.
Multica skills attached to openclaw-backed agents were still invisible to
the runtime; the AGENTS.md fallback was only a documentation patch.
OpenClaw's skill scanner walks <workspaceDir>/skills/ (plus a few other
roots), and workspaceDir is resolved from the openclaw config file —
specifically agents.list[id].workspace → agents.defaults.workspace →
~/.openclaw/workspace. There is no CLI flag or env var override on the
agent runtime; the only knob is the config file.
This change wires a per-task synthesized config:
1. execenv.prepareOpenclawConfig deep-copies the user's existing
openclaw.json (priority: $OPENCLAW_CONFIG_PATH, else
~/.openclaw/openclaw.json), rewrites agents.defaults.workspace AND
every agents.list[].workspace to the task workdir, and writes the
result to {envRoot}/openclaw-config.json. Provider sections,
registered agents, model providers, gateway settings — everything
openclaw needs to actually start — are preserved as-is.
2. resolveSkillsDir for "openclaw" now points at {workDir}/skills/,
which is the first path openclaw scans under workspaceDir. Skills
written here are picked up natively.
3. daemon.go exports OPENCLAW_CONFIG_PATH={env.OpenclawConfigPath} on
the openclaw subprocess and adds OPENCLAW_CONFIG_PATH to the
custom_env blocklist so users cannot accidentally override it.
4. buildMetaSkillContent now lists openclaw alongside the
"discovered automatically" providers; the .agent_context/skills/
fallback line stays for gemini/hermes.
The new regression test TestPrepareOpenclawSkillWriteMatchesScanPath is
the one MUL-2219's DoD calls out: it resolves the workspaceDir the way
openclaw does (reading agents.defaults.workspace out of the synthesized
config) and proves {workspaceDir}/skills/<name>/SKILL.md is what Multica
actually wrote. The pre-MUL-2219 fix asserted "we wrote a file" without
checking the scanner would ever see it — which is how the dead drop into
.openclaw/skills/ landed in #2621's first commit.
Verified locally: minimum-viable synthesized config validates via
`openclaw config validate`, and `OPENCLAW_CONFIG_PATH=<path> openclaw
config get agents.defaults.workspace` returns the task workdir as
expected. MUL-2219
Co-authored-by: multica-agent <github@multica.ai>
* fix(execenv): delegate openclaw config parsing to CLI and fail closed
Address Elon's must-fix on PR #2628: the previous implementation parsed
~/.openclaw/openclaw.json with encoding/json, which cannot read JSON5
or follow $include — the OpenClaw spec's actual format. When parsing
failed, prepareOpenclawConfig silently emitted a minimal config, which
could boot OpenClaw without the user's registered agents, model
providers, or API keys.
Two changes:
1. Delegate active-config-path resolution and config reading to the
openclaw CLI itself. `openclaw config file` locates the active
config (covering OPENCLAW_CONFIG_PATH / OPENCLAW_STATE_DIR /
OPENCLAW_HOME / default and the legacy chain), and the wrapper we
write uses $include to point at it so OpenClaw's own loader handles
JSON5, $include nesting, env-substitution, and secret refs. We read
only agents.list via `openclaw config get --json` to rewrite each
entry's workspace — secrets, comments, and includes in the user
config are never touched.
2. Remove the silent minimal-config fallback. Any CLI failure,
malformed output, or write error now surfaces as a hard error from
Prepare / Reuse. The only "synthesize minimal" path left is a fresh
install (CLI reports a path but the file doesn't exist), where
there is no user data to lose.
The per-task override still rewrites every agents.list[].workspace,
not just agents.defaults.workspace — this is intentional task
isolation, documented in prepareOpenclawConfig and the PR body. A
host-scope per-agent workspace would otherwise silently route the
scanner back to the user's shared workspace.
Cleanups Elon flagged in the same review:
- daemon.go inline-system-prompt comment no longer claims openclaw
ignores the task workdir; it does load it now, and the inline brief
is a belt-and-suspenders carryover for older releases.
- execenv.go openclaw block no longer references "skill file paths in
the inline brief" — the brief uses "discovered automatically".
Reuse() switches to a ReuseParams struct so the openclaw binary path
threads through alongside CodexVersion without a 6th positional arg.
MUL-2219
Co-authored-by: multica-agent <github@multica.ai>
* fix(execenv): grant OpenClaw $include cross-dir confinement for per-task wrapper
The per-task wrapper at envRoot/openclaw-config.json $includes the user's
active config (typically ~/.openclaw/openclaw.json), but OpenClaw confines
$include resolution to the wrapper file's directory unless the target's
parent is granted via OPENCLAW_INCLUDE_ROOTS. Without this, OpenClaw refuses
to follow the link at runtime and the wrapper boots with no user-registered
agents.
prepareOpenclawConfig now returns dirname(activePath) as IncludeRoot, and
the daemon prepends it to whatever the user already has in
OPENCLAW_INCLUDE_ROOTS via the new composeOpenclawIncludeRoots helper
(dedupes, drops empty segments, preserves user-configured roots). Fresh
install emits no $include and leaves the env var untouched.
Adds OPENCLAW_INCLUDE_ROOTS to the custom_env blocklist so a per-agent
override cannot strip the granted root.
Regression tests:
- TestPrepareOpenclawConfigWrapperLoadableUnderIncludeConfinement asserts
every $include target's dirname is covered by the IncludeRoot we surface.
- TestPrepareEnvironmentOpenclawWiresIncludeRoot covers the non-fresh-install
Environment wiring.
- TestComposeOpenclawIncludeRoots covers the daemon-side env composition
(preserve, dedupe, drop empties).
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
* fix(squad): wake leader when dual-role agent posts as worker (MUL-2218)
The squad-leader self-trigger guard skipped a comment whenever the
author equalled the squad's leader id, regardless of the role the agent
was acting in. For an agent that holds both leader and worker roles in
the same squad, this meant the leader role never reacted to its own
worker output and the issue stalled.
Tag each enqueued task with is_leader_task and consult the agent's
most recent task on the issue from both self-trigger guards (comment
path + @squad mention path) — skip only when that task was itself a
leader task.
Co-authored-by: multica-agent <github@multica.ai>
* fix(squad): inherit is_leader_task on retry task clone (MUL-2218)
CreateRetryTask cloned a parent task into a fresh queued attempt but
omitted is_leader_task from the column list, so the child silently fell
back to the column default (false). For a leader task that hit auto-retry
through MaybeRetryFailedTask, the retried task posed as a worker task —
the self-trigger guard then no longer recognised the leader's own
comments, re-opening the very loop MUL-2218 closes.
Inherit p.is_leader_task in the clone and add a query-level test that
covers both leader and worker retries.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
* MUL-2215: fix(daemon): close handleRuntimeGone success/straggler race
handleRuntimeGone coalesced concurrent recoveries with a per-workspace
`reregisterNextAttempt` slot that was deleted immediately on success. A
late-arriving goroutine whose `removeStaleRuntime` was delayed by mutex
contention could reach the coalesce gate after the winner cleared the
slot, observe no slot, re-claim, and double-register — the source of the
intermittent `register endpoint called 2 times under stampede, want 1`
failure on PR #2348.
The slot delete on success is intentional (a genuinely later distinct
deletion in the same workspace must register again, validated by
TestHandleRuntimeGone_DistinctDeletionsWithinCoalesceWindowBothRecover),
so we can't just extend the slot's lifetime.
Add a second per-workspace gate: `reregisterLastCompletedAt`. Every call
captures `entryAt` at the top of handleRuntimeGone; at the coalesce gate
a caller bails if `lastCompletedAt >= entryAt`, i.e. a peer's register
completed AFTER we entered the function. Same-wave stragglers bail
deterministically; distinct later events have `entryAt > lastCompletedAt`
and proceed.
Extracted the gate into `tryClaimRegisterSlot` / `recordRegisterCompletion`
so the race can be exercised deterministically with synthetic timestamps
instead of relying on `-count=N` to win the scheduling lottery.
- TestHandleRuntimeGone_CoalescesConcurrentCallers: -count=500 -race
clean (previously intermittent).
- New unit tests cover the straggler bail, the distinct-later-event
claim, failure backoff suppression, and peer-holds-slot coalescing.
Co-authored-by: multica-agent <github@multica.ai>
* MUL-2215: narrow completion stamp to success path
Second review caught that recordRegisterCompletion stamped
lastCompletedAt on both success and failure. A failed register has not
covered any workspace state, so a same-wave straggler whose entryAt
predates the failure must be allowed to retry once the failure backoff
expires — the previous behavior would let the failure-time stamp also
hide that straggler. workspaceSyncLoop only retries when a workspace's
runtimeIDs fully drain, so partial-deletion recovery has to come from
the straggler path.
Failure path now only updates reregisterNextAttempt; success path keeps
its existing stamp + slot clear. Add a regression test covering the
entryAt-before-failed-completion / arrival-past-backoff edge.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
* fix(execenv): write OpenClaw skills to .openclaw/skills/ for native discovery
The OpenClaw provider was missing a case in resolveSkillsDir, so workspace
skills attached to OpenClaw-backed agents fell through to .agent_context/
skills/ — a path the openclaw CLI never inspects. The result: agents
created against the OpenClaw runtime saw zero of their loaded Skills in
chat or task runs, even though the meta AGENTS.md content advertised
them as auto-discovered.
Mirrors the same per-provider mapping already in place for OpenCode,
Copilot, Pi, Cursor, Kimi, Kiro. Also adds .openclaw to the repocache
git-exclude list so the per-task skills directory does not pollute
checked-out repos. MUL-2213
Co-authored-by: multica-agent <github@multica.ai>
* fix(execenv): drop .openclaw/skills dead-drop write; flag openclaw as non-auto-discovery
Reviewer (Elon) pointed out that {workDir}/.openclaw/skills/ is not in any
OpenClaw skill discovery path. Confirmed by reading openclaw upstream
(src/agents/skills/refresh.ts, src/agents/agent-scope-config.ts,
src/cli/program/register.agent.ts):
- OpenClaw scans <workspaceDir>/skills, <workspaceDir>/.agents/skills,
~/.openclaw/skills, ~/.agents/skills, bundled, and config
skills.load.extraDirs.
- workspaceDir is resolved from the openclaw config (per-agent
workspace -> agents.defaults.workspace -> ~/.openclaw/workspace).
It is NOT the cwd of the openclaw process.
- There is no --workspace CLI flag on 'openclaw agent', and no
OPENCLAW_WORKSPACE env var consumed at runtime. The only knob is the
config file.
So {workDir}/.openclaw/skills/ written by Multica is never seen by the
openclaw runtime, and the meta AGENTS.md was lying to the agent by
claiming auto-discovery. Reverts:
- resolveSkillsDir: drop the openclaw case; falls back to
.agent_context/skills/ (same path as hermes).
- agentGitExcludePatterns: drop .openclaw; nothing is written there now.
Also updates the openclaw branch in buildMetaSkillContent to point the
agent at .agent_context/skills/ explicitly (alongside gemini/hermes), so
loaded skills are at least referenced by path in the AGENTS.md context.
The openclaw native loader still won't see them as installed skills.
Native auto-discovery for openclaw needs per-task workspace integration
(e.g. synthesized per-task config via OPENCLAW_CONFIG_PATH that overrides
agents.defaults.workspace, or resolving the agent's actual configured
workspace at exec time) — tracked as follow-up. MUL-2213
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
* fix(daemon): resolve agent CLIs via login shell when daemon PATH misses them
GUI-launched daemons on macOS/Linux do not inherit the user's interactive
shell PATH, so fnm/nvm/volta multishells and the Anthropic native installer
silently disappear during onboarding even though `claude --version` works
in Terminal. Fall back to `$SHELL -ilc` to ask the login shell for the
canonical absolute path, then verify it with exec.LookPath before trusting
it. Symlinks (fnm/nvm prefix dirs) are resolved while the helper shell is
still alive so per-session paths get canonicalised before they vanish.
Refs MUL-2167, multica-ai/multica#2512.
Co-authored-by: multica-agent <github@multica.ai>
* fix(daemon): strip alias shadowing, harden timeout, lazy-resolve via login shell
Three follow-ups from the PR #2620 review (Elon):
1. Alias shadowing — `command -v claude` in zsh/bash returns the alias
definition, not the binary, and the absolute-path filter then rejects it.
The script now `unalias`/`unset -f` the name before lookup so `command -v`
falls through to the real PATH binary. This is the exact case behind
#2512.
2. Hard timeout — `CommandContext` kills only the shell process. Rc files
that background processes inheriting stdout (`direnv hook`, `nvm` shims,
plain `&`) keep the pipe open and `cmd.Output()` would block for as long
as the survivors live. `Cmd.WaitDelay` forcibly closes the pipes once
the cap elapses, so total startup penalty is bounded by
`timeout + waitDelay` regardless of rc-file content.
3. Lazy fallback — the resolver no longer runs on every daemon start.
`getShellResolved` is `sync.Once`-guarded and only fires when a bare
command name actually misses `exec.LookPath`. Users whose PATH already
contains every agent never pay the rc-file load cost.
Tests: - `TestResolveAgentsViaLoginShell_StripsAliasShadowing` — rc declares
`alias fakeclaude=...`, real binary lives on PATH, resolver must
return the binary, not the alias text.
- `TestResolveAgentsViaLoginShell_HardTimeoutOnBackgroundedStdout` —
rc backgrounds a 60s sleeper holding stdout; resolver must return
inside `timeout + waitDelay + slack`, not 60s.
- `TestLoadConfig_SkipsLoginShellWhenLookPathSucceeds` — when
exec.LookPath finds every agent, SHELL (a marker-writing sentinel)
must not be invoked.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
When the user opens quick-create with a squad selected, the task is
enqueued against the squad's leader agent — but the squad, not the
leader, is the expected owner. The prompt previously instructed the
leader to "default to YOURSELF" using its own agent UUID, hiding new
issues from the squad's delegation flow.
Surface the squad's id + name on the claim response and branch the
default-assignee instruction in buildQuickCreatePrompt: when SquadID is
present, point --assignee-id at the squad UUID and explicitly forbid
self-assignment.
MUL-2203
Co-authored-by: multica-agent <github@multica.ai>
Exposes the existing /api/tasks/{id}/cancel backend endpoint as a CLI
command. Combined with upstream #2107 (cancel running agent on
server-side task delete), this gives operators a way to interrupt a
runaway agent push-storm without resorting to admin-bypass on the
downstream PR.
Use cases:
- Titan / DevBot iterating beyond its boundary (e.g. push-skip loops)
- Codex turn that locked in tool-call spam
- Manual recovery when a long-running task needs to stop NOW
Symmetric with 'issue rerun': accepts the short ID prefix shown by
'issue runs', supports --issue scoping, and reuses resolveTaskRunID
for ambiguity handling.
Refs: PR#19 octo-server post-mortem (2026-05-13)
Co-authored-by: yujiawei <yujiawei@mininglamp.com>
Text/code attachments (markdown, JSON, .ts, .log, …) need an attachment id
to render through `/api/attachments/{id}/content`. The composer pipeline
was dropping that id at the upload-hook boundary, so the Eye preview gate
only fired for media (PDF / video / audio via filename fallback).
- `useFileUpload` now returns the full `Attachment` (with `link` kept as a
`url` alias) so editor providers can resolve content-type and id.
- New-comment and reply composers hold a `pendingAttachments` state and
feed it to `ContentEditor`; the active subset (those still referenced in
the markdown) is sent on submit as before.
- Comment edit modes (CommentRow + CommentCardImpl) merge pending uploads
with `entry.attachments` for the editor and pipe `attachment_ids` into
`onEdit` so newly uploaded files actually bind to the comment.
- Issue description editor pushes pending `attachment_ids` on every
debounced save and invalidates `issueKeys.attachments` so the preview
Eye survives a refresh.
- `UpdateComment` and `UpdateIssue` handlers accept `attachment_ids` and
call the existing `linkAttachmentsByIDs` / `linkAttachmentsByIssueIDs`
helpers; the bind is idempotent so re-sending an existing id is safe.
Closes MUL-2153.
Co-authored-by: multica-agent <github@multica.ai>
* fix: trigger squad leader agent run when squad is @mentioned in comment
Previously, enqueueMentionedAgentTasks only processed m.Type == "agent"
mentions, skipping squad mentions entirely. The shouldEnqueueSquadLeaderOnComment
path only fires when the issue is already assigned to a squad.
This adds handling for m.Type == "squad" in enqueueMentionedAgentTasks:
when a squad is @mentioned, look up the squad's leader agent and enqueue
a task for them (with the same dedup/self-trigger/archived guards as
direct agent mentions).
Co-authored-by: multica-agent <github@multica.ai>
* fix: add canAccessPrivateAgent gate to squad mention branch
Closes the P1 permission vulnerability where a plain workspace member
could trigger a private squad leader by @mentioning the squad, bypassing
the private-agent access check that the direct @agent mention path
enforces.
Adds regression test TestCreateComment_SquadMentionPrivateLeaderBlocksPlainMember.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
* feat(agents): rewrite template catalog as 25 lightweight starters
Replaces every Phase-1 template with a curated set built around the
"persona + intake + scaffold + hard negatives" instruction shape. Cross-
platform survey (Cursor / Cline / Roo / Continue / Custom GPTs) showed
the industry baseline for starter agents is "few but sharp" — single
intent, no methodology buy-in, mostly prompt-only. The original catalog
went the opposite direction (avg 2.5 skills, six-skill Full-stack
methodology stack) and felt heavy for first-time use.
Catalog shape:
- 25 templates across 7 categories: Engineering (8), Product (4),
Writing (5), Design (3), Communication (2), Team (1), Productivity (2).
New Product / Design / Communication / Team domains fill gaps the old
Eng-heavy catalog ignored.
- 16 / 25 are prompt-only (no skill fan-out). Avg 0.56 skill per template
vs. 2.5 prior. Heaviest is 2 skills, only for templates whose intent
cannot be expressed in instructions alone (Playwright runner, single-
file HTML bundlers, design + UX-guidelines pair).
- Universal top-frequency intents that the old catalog missed are now
covered: Code Explainer (intent #1 across every platform surveyed),
Translator (中英), Summarizer, Writing Critic, PRD Drafter/Critic,
RCA Writer, ADR Writer, PR Description Writer, Commit Message Writer.
Loader allows 0-skill templates:
- server/internal/agenttmpl/loader.go drops the "must declare at least
one skill" validation; comment explains the picker's "Prompt only"
rendering path.
- loader_test.go: removed the corresponding negative case, added
TestLoadFromFS_PromptOnlyTemplate as a regression guard.
- agent_template.go handler is unchanged — every len(tmpl.Skills) call
site was already 0-safe (empty fan-out short-circuits the fetch phase
and the in-tx loop both skip cleanly).
Frontend:
- template-picker.tsx: 18 new lucide icons (BookOpen, Bug, GitPullRequest,
GitCommit, AlertTriangle, Scale, ClipboardList, Microscope, UserRound,
Target, Highlighter, Languages, AlignLeft, GraduationCap, Lightbulb,
Type, MessageSquare, Briefcase). Card renders a "Prompt only" badge
when skills.length === 0 instead of "0 skills".
- template-detail.tsx: skill list section is hidden entirely for prompt-
only templates — a header reading "Includes 0 skills" above an empty
list was just visual noise. Instructions section below carries the
agent's identity for these.
- locales/en + zh-Hans agents.json: new create_dialog.template_card.
prompt_only key ("Prompt only" / "纯指令").
Verification:
- go test ./internal/agenttmpl/ — 9/9 pass, including
TestLoad_RealTemplates which fails closed if any new JSON is malformed.
- pnpm typecheck — all 6 packages clean.
- pnpm --filter @multica/views test — 482/482 pass.
- pnpm lint — 0 errors.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(agents): add category filter pills to template picker
25 templates across 7 categories made the picker scroll-heavy on first
open. Add a single-select category filter row above the grid so a PM
can isolate Product templates in one click, an engineer can jump
straight to Engineering, etc.
Visual reuses the IssuesHeader scope-toggle pattern verbatim — Button
variant="outline" + active class swap (bg-accent / text-muted-foreground)
— so the affordance reads the same as the existing filter pills in
issues / squads / runtimes / my-issues. flex-wrap keeps the 8 pills
(All + 7 categories) honest on narrow widths.
Counts are inlined into the label ("Engineering (8)") rather than
shown as a separate badge — single-line-tall pills look right next to
the picker grid, and surfacing the per-category density up front
doubles as a hint at the catalog's "less but sharper" intent.
When a specific category is active, the grid renders flat (no
section headers) — the active pill already names what's on screen,
and a header reading "Engineering" above an only-Engineering grid is
visual duplication. "All" falls back to the prior grouped layout.
State is component-local (no URL sync, no persistence) since the
picker is dialog-internal transient state — closing the dialog
naturally resets the filter, which is the expected behaviour for a
"choose from a catalog" surface.
i18n: new `create_dialog.template_picker.filter_all` key in en + zh.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The per-turn prompt in buildCommentPrompt() only injected the squad
leader no_action prohibition inside the 'if TriggerAuthorType == agent'
block. When a member (human) posted a comment like 'LGTM', the squad
leader was triggered but the per-turn prompt did NOT include the
prohibition, causing the model to post noise comments like 'LGTM is a
pure acknowledgment — no reply needed. Exiting silently.'
Fix: move the squad leader no_action rule outside the agent-only block
so it fires for ALL trigger types (agent and member).
Fixes: MUL-2168
Co-authored-by: multica-agent <github@multica.ai>
PR #2564 only added IsSquadLeader handling to the assignment-triggered
workflow path and the Output section. When a squad leader is triggered by
a comment (the common case for re-evaluation), the comment-triggered
workflow path had NO squad leader special handling, so the model still
posted comments announcing no_action/silence.
Changes:
- runtime_config.go: Add IsSquadLeader check to comment-triggered step 4
with explicit prohibition against posting no_action announcement comments
- runtime_config.go: Strengthen Output section from 'may exit silently' to
'MUST exit without posting any comment' with explicit DO NOT examples
- runtime_config.go: Strengthen assignment-triggered step 5 similarly
- prompt.go: Add squad leader no_action rule to per-turn comment prompt
when trigger author is an agent and agent instructions contain the
Squad Operating Protocol marker
- Add tests for both the per-turn prompt and CLAUDE.md generation
Fixes MUL-2168
Co-authored-by: multica-agent <github@multica.ai>
* fix(squad): skip leader on comment when a member @mentions any agent (MUL-2170)
When a human commenter routes an issue directly at a specific agent via
[@Name](mention://agent/<id>), the squad leader was still being woken up
to evaluate the same comment. The leader's only real options were to
re-delegate to the agent the member already named or to record
no_action — both of which produce queue noise without changing the
outcome.
This skips the leader-enqueue path entirely when:
- the assignee is a squad,
- the comment author is a member, AND
- the comment body contains at least one agent mention.
Agent-authored comments are intentionally exempt: when an agent posts
an update that @mentions another agent, the leader still needs to
coordinate the thread. The existing leader-self-trigger guard is
preserved. Only the current comment's body is inspected — parent
(thread root) mentions are not inherited here.
Tests cover the helper (mentions parsing) plus the integration matrix:
member plain / member @member / member @non-leader-agent /
member @leader / agent @agent / leader-self.
Co-authored-by: multica-agent <github@multica.ai>
* test(squad): exercise full CreateComment path for leader-skip rule (MUL-2170)
Adds an integration test that drives the HTTP-layer CreateComment handler
(not just the helper) to lock the call-site wiring: a member top-level
comment with an @agent skips the squad leader, and a subsequent plain
reply in the same thread DOES wake the leader — the parent's @agent
mention must not be inherited into the leader-skip decision.
Picks up a non-blocking review note on PR #2569.
Co-authored-by: multica-agent <github@multica.ai>
* fix(squad): skip leader on any explicit member mention, not only @agent (MUL-2170)
Broaden the leader-skip rule for squad-assigned issues: a member comment
that explicitly @mentions anyone — @agent, @member, @squad, or @all —
counts as deliberate routing and the squad leader stays out. Issue
cross-references (mention://issue/...) are not routing and still trigger
the leader as before.
Per Bohan's follow-up on MUL-2170 — @member should suppress the leader
for the same reason @agent does: the human has already pointed at a
specific recipient, so a leader turn would just be observation noise.
Helper renamed commentMentionsAnyAgent → commentMentionsAnyone with
explicit handling of all four routing mention types. Existing call-site
wiring (current-comment-only, agent-author exemption, leader self-trigger
guard) is unchanged.
Tests updated and extended to cover the full routing matrix:
@member / @squad / @all / @issue (cross-ref) plus the @agent variants
already covered.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
The runtime prompt's Output section unconditionally required all tasks to
post a comment via 'multica issue comment add', which conflicted with the
squad leader protocol that says to 'exit silently' on no_action.
Changes:
- Add IsSquadLeader bool to TaskContextForEnv (detected via Squad Operating
Protocol marker in agent instructions)
- Relax the Output section and assignment-triggered workflow step 5 to
allow squad leaders to exit with only a 'multica squad activity' call
when the outcome is no_action
Fixes MUL-2168
Co-authored-by: multica-agent <github@multica.ai>
* fix(cli): resolve squad assignees in issue create/update/assign (MUL-2165)
The CLI assignee resolver only searched workspace members and agents, so a
quick-create input like "assign to <SquadName>" silently fell through to
"Unrecognized assignee: <SquadName>" in the issue description — even though
squads are first-class assignees server-side and the prompt's whole point was
to route the work for the user.
Extend resolveAssignee / resolveAssigneeByID to also fetch /api/squads, teach
the actor display lookup to render squad names in table output, update the
quick-create prompt and runtime-config command listing to mention
`multica squad list` alongside members and agents, and lock in the new
behavior with tests.
Co-authored-by: multica-agent <github@multica.ai>
* fix(cli): gate squad assignee resolution behind an allowed-kinds set (MUL-2165)
The earlier MUL-2165 fix taught resolveAssignee / resolveAssigneeByID to also
return (squad, ...), but those helpers are shared. Project lead and issue
subscriber callers were still using them, and their target schemas reject
squads — project.lead_type has a DB CHECK constraint
(server/migrations/034_projects.up.sql:10) and the subscriber handler's
isWorkspaceEntity switch only knows member/agent
(server/internal/handler/handler.go:414). So
`multica project create --lead "<SquadName>"` and
`multica issue subscriber add --user "<SquadName>"` would resolve to
(squad, ...) and surface as a 500/403 server-side instead of a clean
CLI-side resolution error.
Thread an assigneeKinds set through the resolver and the pickAssigneeFromFlags
helper. Issue create/update/assign/list pass `issueAssigneeKinds` (all three);
project lead and subscriber pass `memberOrAgentKinds`. The squads fetch is
skipped entirely when not allowed, and the not-found / no-match error wording
adapts to the allowed kinds so it never mentions a type the caller cannot use.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
