Agent-native parity: users can list workflows and select one for a task in
the dashboard, so agents should be able to as well. Adds two task-session
tools — fn_workflow_list (read: built-ins + user definitions) and
fn_workflow_select (assign a workflow to a task, defaulting to the current
one) — wired into the executor's customTools and exported from the engine
index. Covered by unit tests against mock stores.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The node "Auto-approve requests" toggle was captured but unused. The only
human-approval pause reachable from a custom node is the CLI first-run
trust-on-first-use gate (review-style nodes run as ephemeral readonly agents
with no permission gate), so autoApprove now bypasses that pause — a superset
of the CLI-specific cliSkipApproval flag. The inspector explains the effect
when the toggle is on.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
In-review tasks showed other tasks' files in their "files changed"
list. Task branches fork from local main, but the base capture
measured merge-base(HEAD, origin/main) — when local main carried
merged-but-unpushed task commits, the recorded base rewound past
them, and after the post-merge rebase-and-push rewrote those SHAs,
baseCommitSha..HEAD permanently swept the predecessors' files into
the new task's diff.
Extract the capture into base-commit-capture.ts, measure local main
first (origin/main fallback) to match the contamination-base sites,
and add a real-git regression suite covering local-ahead-of-origin.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Address PR #1363 review findings:
- core: pausedReason was written in-memory and read by SELECT but never
persisted by the task upsert (missing column/value) nor mapped back in
rowToTask — so it was lost on every reload. Add it to both. This is the
root cause behind the workflow CLI-approval / await-input pause cycle and
also fixes token-budget / worktrunk pause reasons silently vanishing.
- dashboard: approve-cli now derives the approved command exclusively from
the task's pausedReason; a caller-supplied body.command is ignored, closing
a trust-on-first-use bypass.
- engine: await-input nodes resume only when THIS node paused the task (its
marker on pausedReason), not on any pre-existing steering comment.
- engine: write-capable custom nodes (coding/script/CLI) are refused until a
task worktree exists, so they never mutate the shared repo root before the
execute seam.
- engine: document cliSkipApproval as an intentional workflow-author-only
escape hatch; scriptName is now const (ESLint).
- tests: pausedReason round-trip coverage in store-persistence; approve-cli
body-command-ignored + no-pending-command coverage; built-in-aware list
assertion in workflow-routes.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Users can now watch everything the agent does while a CE stage works, steer
it mid-stage, and read the whole conversation as a proper chat surface.
Live output:
- New host capability: CreateInteractiveAiSessionOptions.onProgress — the
engine adapter streams thinking/text deltas + tool start/end markers from
the pi agent hooks (any plugin can use this).
- Orchestrator buffers per-session live activity (merged deltas, discrete
tool lines, capped), emits throttled progress events over SSE, and
GET /sessions/:id attaches it as liveActivity for the polling fallback.
- Routes detach turn execution: start/answer/resume return immediately
(status active) and clients converge via push/poll — the turn is watchable
instead of hidden inside a blocking POST.
- Turn timeout is now INACTIVITY-based: an actively-working long turn is
never killed; a quiet one interrupts with its working trace preserved.
- On settle the trace persists into history as a condensed record.
Steering:
- Stage protocol: responses may be a direct answer, {value, comment}
(answer + guidance), or {feedback} (guidance without answering); the
system prompt instructs agents to treat steering as first-class input.
- CeFlow: guidance textarea alongside selectable questions — attach to the
clicked answer, or "Send guidance" on its own.
Q&A UI:
- Transcript no longer hides control records: past questions/answers render
as chat bubbles (option ids → labels), steering turns marked, working
traces as collapsible "Agent work" blocks, completion marker.
- Live working pane (pulse + streaming thinking/tool lines) while a turn runs.
Tests: 130 plugin tests green (14 new: live buffer/flush ordering, inactivity
watchdog survives active work, detached convergence, steering payload shapes,
transcript rendering, live pane). Engine seam tests green; plugin/core/
engine/dashboard tsc clean. Core full suite OOMs locally (known orchestrator-
shell issue) — covered by CI shards.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A CLI node can set cliSkipApproval to bypass the trust-on-first-use pause and
run its command immediately. Exposed as a checkbox in the node inspector.
- runFeatureValidation now lazy-ensures a linked assertion (FN-5902) instead
of the removed zero-assertion auto-pass, for both task-completion and the
stranded-feature recovery path
- CONCEPTS.md: union of main's Merge-lifecycle cluster and this branch's
Missions clusters; Contract Assertion entry updated for FN-5902 semantics
- AGENTS.md: take main's docs/solutions + CONCEPTS.md pointer wording
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
CLI prompt nodes now accept a raw cliCommand (any command + args), not just
named scripts. A raw command must be explicitly approved by the user before it
runs: an unapproved command pauses the task (status awaiting-cli-approval) with
the command shown; the user approves via POST /tasks/:id/workflow/approve-cli,
which records the exact command string in settings.approvedWorkflowCliCommands
and resumes. Named scripts (settings.scripts) still never require approval.
Adds POST /tasks/:id/workflow/input to answer await-input nodes (records a
steering comment + resumes).
Prompt nodes now support an executor kind: model (provider/model override),
agent (adopts a named agent's model + custom instructions), skill (prompt
becomes a skill invocation), and cli (named project script with the prompt in
FUSION_NODE_PROMPT — raw commands still never accepted). Per-node maxRetries
overrides the executor-wide default (capped at 10). Await-input nodes pause the
task with status 'awaiting-user-input' and the question as pausedReason; on
unpause the newest steering comment is consumed as the answer and exposed in
graph context.
- liveness: graph-routed tasks count as executing in getExecutingTaskIds/
isTaskActive and are skipped by recoverCompletedTask + the completed-task
resume fast-path — recovery can no longer drive a parallel lifecycle (P0)
- handleGraphFailure: sets status 'failed' (self-healing revival exemption,
prevents FN-5704-style re-run loop) and leaves paused tasks untouched
- execute seam distinguishes pause/abort from implementation failure
- merge seam: 30-minute timeout so a wedged queue cannot strand the run
- awaitAbortInFlightTaskWork: defensive interceptor/routing cleanup
- deleteWorkflowDefinition: cascades to selections, materialized steps,
affected tasks' enabledWorkflowSteps, and the project default
- Add behavior-level tests for the shared merge-enqueue funnel
(enqueueEligibleInReviewTasks) with a Surface Enumeration of all
in-review entry surfaces, per review
- Seed real stale in-review fixtures in the FN-5147 no-mutation
regression block so sweeps enumerate candidates and the assertions
are non-vacuous
- Keep per-task auto-merge gating uniform across reclaim/contamination
candidate columns: the suggested in-review-only scoping broke the
FN-5704 regression contract (reclaim short-circuits when autoMerge
is off); documented the tension in code comments and the learning doc
- Drop hardcoded commit hash from the learning doc
Review residuals #3/#4/#6/#10: per-group in-process promotion lock (concurrent
route+auto promotion could double-create PRs), finalized-but-PR-less groups can
be repaired by re-promotion without re-merging, auto-promotion failures emit
merge:branch-group-promotion-failed instead of silent swallow, exported
reconcileBranchGroupPr for out-of-band merged reconciliation, and the merger
sync block drops its (store as any) casts (TaskStore already carries the
methods).
Code review (Tier 2) found two P1s: (1) the early no-op fast-path persisted
mergeConfirmed/mergeTargetBranch without mergeTargetSource, so a shared-group
member landing via it could never satisfy the strict completion predicate —
promotion permanently blocked; thread mergeTarget.source through like the
standard landing sites. (2) createGroupPrCallback's findPrForBranch used
state:'all' and could reuse a closed/merged PR from a prior group, persisting
a terminal prState onto a fresh promotion; create path now matches open PRs
only.
Real engine seams: execute delegates to the legacy implementation phase via a
completion interceptor that stops execute() at the implementation-complete
boundary (no double review/merge); review performs the in-review handoff; merge
resolves through ProjectEngine.onMerge over the same serialized merge queue
(wired via a late-bound setMergeRequester, mirroring setMergeEnqueuer). Custom
graph nodes run on the proven WorkflowStep machinery (readonly tool policy,
verdict parsing). Adds a 'planning' seam to the vocabulary (no-op for
pre-specified tasks; custom planning is a prompt node today).
Entry point: execute() routes graph-selected tasks through the runner when
experimentalFeatures.workflowGraphExecutor is on, with process-wide routing
claims (FN-4811 posture), duplicate-dispatch dropping, pre-run errors falling
back to legacy, and mid-run errors parking the task in review (never re-running
the implementation, never stranding the task).
Flag off by default: all 587 executor tests pass unchanged.
Engine half: real-git E2E covering planning- and mission-sourced groups —
members land on the group branch (never main/sibling), completion-gated single
PR via injected callback, re-promote idempotency, sync on later landing,
abandon→closed, and a self-healing finalize mid-flow staying group-anchored.
Core half: real triageFeature stamps the BG- id, member enumeration, and the
canonical completion gate flipping on landing.
Loads a task's selected workflow, runs the graph with injected legacy seams
(execute/review/merge) and a custom-node runner, and maps the terminal outcome
to completed/failed/fell-back. Any interpreter-level error falls back so the
caller can run the legacy pipeline — a task is never stranded. Covered with
fake seams: lifecycle ordering, failure routing, gate blocking, fallback
reasons, diagnostics isolation. Includes the interpreter-cutover plan doc.
Non-seam prompt/script nodes now dispatch to an injected WorkflowCustomNodeRunner
instead of throwing; gate nodes support an executable (prompt/script-backed) form
alongside the original context-gate contract. WorkflowGraphExecutor accepts the
runner via deps.
Push the single group PR's body (member checklist, x/N landed) on each member
landing via an injected SyncGroupPrFn — new updatePr/closePr GitHubClient
helpers (gh CLI + API parity); refreshPrInBackground is task-scoped/wrong
direction and intentionally not reused. Sync failures are non-fatal+retryable;
out-of-band closed/merged PRs reconcile prState instead of erroring. New
POST /branch-groups/:id/abandon closes the PR best-effort and marks the group
abandoned. Also fixes the U5-introduced stub-context regression in the U4
dashboard bridge test (missing options).
Tasks with autoMerge explicitly enabled never auto-merged when the
project-level setting was disabled: the merge enqueue gate
(allowInReviewMergeProcessing) and all 19 in-review self-healing sweeps
checked only settings.autoMerge, and the board stall-signal hydration
passed the raw global into the diagnostic gates.
Introduce allowsAutoMergeProcessing(task, settings) in core — additive
relative to the global setting so configs with global auto-merge ON are
unchanged (explicit autoMerge:false tasks still flow to the merger's
manual-required parking) — and use it at the enqueue gate, every
self-healing sweep, and the store's stall/stalled signal contexts.
Group promotion in PR mode previously flipped prState to 'open' without ever
calling GitHub — prNumber/prUrl were never populated. Add an injected
CreateGroupPrFn (mirrors the processPullRequestMerge seam, no engine→dashboard
import): coordinator creates-or-reuses exactly one PR per group, persists
prNumber/prUrl/prState, and leaves state untouched on GitHub failure so
re-promotion retries. Idempotent via persisted prNumber +
getBranchGroupByBranchName. Wired at all three CLI engine-construction sites
(daemon/dashboard/serve).
The dashboard promote route called engine.promoteBranchGroup(groupId) as a
method that never existed — only a standalone coordinator function did — so
the route was dead, masked by a vi.fn mock in the test. Add the real method on
ProjectEngine delegating to the coordinator (resolving store/cwd/settings like
attemptBranchGroupPromotion), and de-mock the test so it now fails if the
method goes missing. No PR-creation behavior yet (U5).
Audit of all shared-member merge + self-healing finalize paths: routing,
merger finalize-success, and the 6 self-healing recovery paths were already
group-branch-safe (FN-5846). Found a residual of the 2026-05-23 lost-work
incident bug #2: already-merged-detector's ancestry strategy used bare
git log --grep first-hit, and the ownership regex made the conventional scope
optional (bare 'feat:' matched). Anchor attribution on trailers or task-scoped
subject; scan candidates instead of accepting the first grep hit. Adds real-git
characterization tests.
Route and coordinator disagreed on landed/complete: the route required
mergeConfirmed + matching mergeTargetBranch, the coordinator accepted bare
column===done/in-review and never checked the branch. Extract canonical
isBranchGroupMemberLanded/isBranchGroupComplete in @fusion/core (stricter
route semantics win — load-bearing for merge-target safety) and consume from
both sides. Tightens promotion gating to fire only when all members are
merge-confirmed onto the group branch.
A mission feature could be left status="done" while its loopState never
advanced past "implementing" and it had no linked board task, so it was
never validated. The slice-completion gate (computeSliceStatus) correctly
refuses to count an assertion-linked done feature until its validator
passes, but nothing re-drove a task-less feature — so the slice, milestone,
and whole mission could never auto-progress.
Active-mission recovery now detects these stranded done features and re-runs
assertion validation directly (read-only judge, no board task): on pass the
feature becomes legitimately complete, on fail the normal fix-feature flow
takes over. Extracted the feature-validation path into a shared
runFeatureValidation helper used by both task-completion and recovery.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The FN-5930 squash merge combined two wait-for-exit strategies,
leaving a shadowed 'const exited' and double await. Collapse to
a single register-before-kill pattern.
Fusion-Task-Id: FN-5930
Reduce flakiness in the real-git verification spawn supervision test.\n\n- write the child PID to stdout with an awaited newline flush before the parent exits\n- collapse the scenario branching so the SIGTERM path is mutually exclusive with crash handling\n- await parent process exit during cleanup and document coverage across normal, signal, and crash teardown paths\n\nFiles changed:\n .../verification-spawn-supervision.real-git.test.ts | 9 ++++++---\n 1 file changed, 6 insertions(+), 3 deletions(-)
Fusion-Task-Id: FN-5930
Fusion-Task-Lineage: 20c3fe73-a640-4c34-8a07-edf5bdda4b25
Closes the U2/U5 skill-discovery carry-forward so the plugin's interactive ce-*
sessions actually load the stage's bundled skill in a live agent (not just in
scripted-fake tests).
Root cause: createFnAgent built its DefaultResourceLoader without forwarding any
skill-discovery path, and the interactive seam options couldn't carry one. The
loader's skillsOverride only *filters* skills already discovered from cwd's
standard roots, so the plugin-local .fusion-ce-skills/<id>/SKILL.md was never
discoverable.
Fix (end-to-end):
- AgentOptions.additionalSkillPaths forwarded into DefaultResourceLoader
- CreateInteractiveAiSessionOptions gains requestedSkillNames + additionalSkillPaths
- the interactive engine adapter forwards them to createFnAgent (skills +
additionalSkillPaths)
- the orchestrator runs the session with cwd on the real project root and hands
it [stage.skillId] + the install root
Proven: a real DefaultResourceLoader with additionalSkillPaths discovers ce-plan
and filters out ce-work; the orchestrator passes the right id/path/cwd. Plugin 96,
engine 136, core 99 tests green.
Keep planning summary actions responsive with operation-specific loading indicators.
- pass separate single-task and breakdown loading flags into the planning summary view
- show the Creating spinner only on Create Single Task and the Breaking down spinner only on Break into Tasks while keeping the sibling action disabled
- add regression coverage for both pending-action paths and normalize the restart integration test temp worktree root under /private/tmp
Files changed:
.../dashboard/app/components/PlanningModeModal.tsx | 14 ++-
.../PlanningModeModal.planning-flow.test.tsx | 138 +++++++++++++++++++++
.../src/__tests__/restart.integration.test.ts | 7 +-
3 files changed, 151 insertions(+), 8 deletions(-)
Fusion-Task-Id: FN-5912
Fusion-Task-Lineage: b93da566-0c2b-4ff8-83ea-b6e009dfd650
Add unread mailbox status to the desktop header toggle while keeping pending approvals prioritized.
- show an unread status dot on the desktop header mailbox toggle when unread mail exists without pending approvals
- keep pending-approval indicators taking precedence and hide mailbox indicators while the mailbox view is active
- extend Header coverage for unread-only, pending-only, combined, zero-count, and active-mailbox states
- update restart integration coverage to reuse an existing worktree during orphaned resume concurrency
Files changed:
docs/dashboard-guide.md | 2 +-
packages/dashboard/app/components/Header.tsx | 9 +++++--
.../app/components/__tests__/Header.test.tsx | 26 ++++++++++++++++---
.../src/__tests__/restart.integration.test.ts | 29 ++++++++++++++++++++--
4 files changed, 58 insertions(+), 8 deletions(-)
Fusion-Task-Id: FN-5915
Fusion-Task-Lineage: 3a207e2c-ca58-402f-94cf-3a9e514fe263
Bundle pinned copies of 7 CE pipeline-stage skills (strategy, ideate,
brainstorm, plan, work, code-review, compound) under src/skills/ and declare
them via PluginSkillContribution. Empirical finding: the skills contribution
alone does not make a SKILL.md resolvable in a session -- the engine ingests it
as a name only. So onLoad runs an idempotent, isolation-guarded physical install
into a plugin-local .fusion-ce-skills/ dir (never a global ~/.claude/skills),
which the engine skill-resolver can then discover. Proven against the real
loadSkills + resolveSessionSkills pipeline.
Expose createInteractiveAiSession on route/loader PluginContext alongside the
one-shot createAiSession. The prompt->parse->retry->pause->resume loop is
reimplemented in an engine seam modeled on planning.ts (no engine await-input
primitive exists). Reuses PlanningQuestion/PlanningResponse; stays generic with
no plugin-specific concepts. Injected on route contexts only; tool/runtime
contexts omit it (parity with createAiSession).
Ensure agent-created ntfy notifications fall back to the task description when the title is still empty.
- pass task descriptions through task-created notification payloads
- cover empty-title task creation in notification service tests
- verify ntfy task-created messages render the id+description fallback
- add a patch changeset for the published CLI package
Files changed:
.changeset/fn-5890-task-created-title.md | 5 +++++
.../engine/src/__tests__/notification-service.test.ts | 12 +++++++++++-
packages/engine/src/__tests__/ntfy-provider.test.ts | 16 ++++++++++++++++
packages/engine/src/notification/notification-service.ts | 1 +
4 files changed, 33 insertions(+), 1 deletion(-)
Fusion-Task-Id: FN-5890
Fusion-Task-Lineage: 4a67a4d0-c3a8-4923-a506-7a67a8836127
Retry incomplete executor work with a fresh session after non-continuable session errors.
- add executor recovery handling that clears session state and requeues incomplete tasks to todo while retry budget remains
- preserve terminal failure behavior once the fresh-session retry budget is exhausted
- extend reliability coverage and AGENTS.md backstop notes for the new retry path
Files changed:
AGENTS.md | 1 +
.../post-done-continuation-no-wedge.test.ts | 48 +++++++++++++++++++++-
packages/engine/src/executor.ts | 40 ++++++++++++++++++
3 files changed, 87 insertions(+), 2 deletions(-)
Fusion-Task-Id: FN-5888
Fusion-Task-Lineage: 4ebce260-8923-4599-aced-541533b94543
