Merges FN-3534 to surface task provenance in CLI extension outputs (task list/show) with corresponding tests and a changeset for the published `@runfusion/fusion` package. Also includes FN-3297 test coverage for incomplete runtime distribution trees in the test-artifacts script.
Fusion-Task-Id: FN-3534
- Clarify task memory scope behavior across core types, project memory logic, engine tool prompts, and related docs
- Add regression coverage for memory scope and runtime plugin alias handling in core/dashboard/engine tests
- Add agent avatar API routes and dashboard UI support for avatar display and storage documentation
- Add line-number gutter toggle support in FileEditor and Files modal with accompanying component tests
- Include changeset for @runfusion/fusion documenting memory scope guidance update
Fusion-Task-Id: FN-3485
pi-ai's openai-codex-responses provider surfaces ChatGPT-plan WebSocket
drops as bare "WebSocket error" / "WebSocket closed <code>". The
underlying ErrorEvent.error is dropped by extractWebSocketError (it only
inspects event.message), so the cause is gone by the time we see it.
The engine then propagated these as fatal failures instead of retrying
a transient network blip.
- Add WebSocket patterns to TRANSIENT_ERROR_PATTERNS so the executor's
existing transient retry path picks them up.
- Tag the model id onto the thrown error and emit a structured warn in
promptSessionAndCheck so future drops can be triaged by transport.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Moonshot/Kimi returns a non-standard finish_reason of "repeat" when its
server-side repetition detector trips. pi-ai surfaces this as a fatal
session state error, killing the agent heartbeat. Filter that case in
promptSessionAndCheck so the truncated turn is logged and the heartbeat
continues on the next tick.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The merge adds an auto-reload setting (FN-3334) with UI controls in the settings modal, documentation, and a new version-check module, while also fixing a bug (FN-3338) where extension providers incorrectly resolved the project root when invoked from git worktrees — moving the project-root resolutio
Fusion-Task-Id: FN-3338
Merges FN-3335 (worktree project resolution) and FN-3333 (spurious version reloads). The engine's `createFnAgent` now resolves project root from the worktree's cwd rather than the parent process, with `resolveProjectRoot` added to skill-resolver for consistency. The dashboard's `versionCheck` was up
Fusion-Task-Id: FN-3335
Merges FN-3008 to add a "fallback-used" notification system: the engine now emits events when AI model fallbacks are triggered, dispatches notifications via ntfy/webhook providers, surfaces a session banner in the dashboard, and exposes a settings toggle to enable or disable these alerts.
Fusion-Task-Id: FN-3008
This merge adds readonly custom tool preservation (FN-3140) with new plugin SDK types and documentation, fixes PluginManager responsive overflow (FN-3093), and integrates the fn-3065 branch with enhanced plugin authoring capabilities. The core plugin-types module was significantly expanded with 230+
Fusion-Task-Id: FN-3140
The merge introduces a new droid CLI extension with path reconciliation across daemon, dashboard, and serve commands, along with supporting core and engine utilities. It also removes stale lint suppression comments (FN-2983). Tests cover the new CLI extension and path reconciliation logic.
Fusion-Task-Id: FN-2984
The title summarizer ran with `tools: "readonly"` but host extensions
(`@runfusion/fusion`) were still injected, exposing `fn_task_create` and
the rest of the `fn_*` mutation surface. A summarizer model called
`fn_task_create` mid-summary, spawning an unintended sibling task and
leaving its chat-style reply ("Created **FN-xxxx** with the full spec…")
sliced as the original task's title.
- pi.ts: in `tools: "readonly"` mode, skip host extension paths and drop
caller-supplied customTools so the session truly only has read/grep/
find/ls.
- ai-summarize.ts: harden all four system prompts (title, merge summary,
commit body, commit subject) with explicit no-tool / treat-input-as-
content framing; wrap the title prompt's user content in a
`<description>` delimiter; route the AI response through new
`sanitizeTitle` that strips chatty preambles, markdown emphasis,
surrounding quotes, and trailing punctuation before truncation.
- Tests: add a regression covering the exact incident shape plus
unit coverage for `sanitizeTitle` edge cases.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
FN-2910 surfaced concurrent reviewer + merger activity on the same task.
Root cause: asymmetric in-flight guards let an unpause-resume kick off a
fresh executor session while a recovery path was already running, and the
auto-merge handoff fired before the executor's finally block finished
cleanup. This sweeps the surrounding lifecycle paths for similar races and
tightens the reviewer pause gate against TOCTOU through runtime setup.
- Symmetric in-flight tracking across `executing`, `recoveringCompleted`,
and `resumingUnpaused`; `recoverCompletedTask` bails when any are set.
- Atomic claim of the recovery slot in the completed-task watchdog before
any awaited work.
- Workflow-rerun bounce returns "bounced" | "skipped-pending" so the
watchdog can no longer log a false-success retry when the original
bounce is still mid-flight.
- Self-healing's completed-task scan re-checks executing IDs inside the
loop instead of trusting a pre-await snapshot.
- 300ms grace period before auto-merge enqueue, giving the executor's
finally block (session disposal, child cleanup) time to drain and
eliminating the residual log-overlap symptom from FN-2910. Test uses
fake timers, no real sleep added.
- New AgentSemaphore.runNested for synchronously nested helper agents
(reviewers): bumps activeCount for honest observability while bypassing
the wait queue, preserving forward-progress fairness for the parent at
low maxConcurrent. Both createReviewStepTool and triage's
createReviewSpecTool now use it.
- New beforeSpawnSession hook on AgentRuntimeOptions/AgentOptions fired
inside createFnAgent immediately before createAgentSession, past every
awaited setup step. Reviewer wires a pause re-check that throws a
sentinel error converted to UNAVAILABLE, closing the TOCTOU window
where pause flipped during runtime resolution or resource loading.
All 2887 engine tests pass; engine + core + cli + dashboard + plugin-sdk
+ pi-claude-cli + desktop typecheck clean.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Triage with claude-sonnet-4-6 via pi-claude-cli kept looping on
fn_review_spec calls that were rejected by pi's validator with
"root: must be object". Parameterless MCP tools (schema
{type:"object", properties:{}}) emit zero input_json_delta events,
so partialJson stayed "" and the catch fell through to
finalArgs = "" — a string, which TypeBox's Type.Object({}) rightly
refuses. Default empty partialJson to {} so the call lands.
Also:
- Add a 2-step reminder loop in triage before swapping to the
fallback planning model — primary models that wrote PROMPT.md
but forgot fn_review_spec recover from a nudge, no need to pay
the cold-start tax of a new triage on a different model.
- Inject @runfusion/fusion's own pi extension into dashboard/
daemon/serve sessions and propagate the path to createFnAgent
via setHostExtensionPaths so fn_* tools register globally
without requiring `pi install npm:@runfusion/fusion`.
- Drop the "historical" qualifier from replayed tool labels —
Claude was reading "TOOL RESULT (historical Read):" as
"previous session, ignore" and looping on verification.
- Remove subprocess-lifecycle stderr debug logs that landed for
hang diagnosis — root cause is fixed, the noise can go.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
pi-coding-agent 0.70's createCodingTools preset only includes read/bash/edit/write,
but pi-claude-cli's tool-mapping translates Claude's Glob→find and Grep→grep.
Triage sessions running through Claude CLI hit "Tool find not found" the moment
the model called Glob and looped on the error.
Compose the built-in set explicitly so every tool referenced by tool-mapping.ts
is registered. Read-only sessions also gain ls (was silently dropped earlier
when readonly was redefined upstream).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
When users have an external pi-claude-cli (e.g. a global `npm install -g
pi-claude-cli`, or `npm:pi-claude-cli` in ~/.pi/agent/settings.json packages),
pi's extension discovery loaded the upstream copy and shadowed our fork. The
upstream has a once-and-lock MCP-config bug that throws "Extension runtime not
initialized" during early streamSimple calls and never recovers.
Adds reconcileClaudeCliPaths in @fusion/core, used by both the daemon's
extension assembly and the engine's per-session registerExtensionProviders, to
drop any path with a `pi-claude-cli` segment that isn't our vendored fork and
prepend the vendored path. Engine resolves the fork via require.resolve and
gracefully no-ops when it isn't reachable (e.g. embedded standalone usage).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Agents working on a task that depends on other tasks (e.g. documentation
alignment tasks needing the sibling tasks' specs) were repeatedly
rejected by the worktree boundary when reading .fusion/tasks/FN-NNNN/PROMPT.md,
which also contributed to the malformed-tool-result crash we just fixed.
Add a read-only exception to isWorktreeAllowedPath: the read/glob/grep
tools may access .fusion/tasks/*/PROMPT.md and .fusion/tasks/*/task.json
at the project root. Writes and bash cwd remain restricted.
Update the system-prompt boundary docs (executor.ts) so agents know the
exception exists and stop burning turns re-trying rejected reads.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
wrapToolsWithBoundary returned a bare {ok:false,error} object when an
agent tried to read/write/bash outside the worktree. pi-coding-agent
wraps tool returns into a toolResult message whose content field it
expects to be an array of content blocks; a bare object leaves content
undefined, which later crashes downstream with "Cannot read properties
of undefined (reading 'filter')" — the failure we've been chasing on
FN-2479 and similar.
Return { content:[{type:"text",text:...}], isError:true, ok:false, error:... }
so pi records a valid toolResult block while existing callers that
inspect .ok / .error still work.
Diagnostic evidence: transcript tail for the failing task showed three
consecutive `read` toolResults with content=array(len=0) (normalized
from undefined by our earlier guard) immediately before the assistant
message with stopReason="error".
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
pi-coding-agent swallows its own exceptions onto session.state.errorMessage
without preserving a stack, so fusion sees only "Cannot read properties
of undefined (reading 'filter')" with no indication which message is
malformed. When promptSessionAndCheck rethrows an error that matches the
generic TypeError shape, dump the last few state.messages (role, content
type, toolName, stopReason) so the offending message can be identified
next run.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
pi-coding-agent's _getUserMessageText calls content.filter(...) on user
messages; if a message lands in state.messages with content === undefined
(string or array expected), the library throws
"Cannot read properties of undefined (reading 'filter')", which gets
caught and stored on session.state.errorMessage and rethrown without a
stack. Fusion's existing message-content guard already normalized
assistant/toolResult messages — extend it to user messages as well, and
sweep state.messages once when the guard is installed so content loaded
from a session file is repaired before the first event fires.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Resolve merge markers across plugin loader, agents view, dashboard server tests, and engine pi session setup
- Keep current cache-busting and hook-guard implementations while removing stale branch-side fragments
- Preserve rate-limit test imports and useAgents-driven loading flow in dashboard agent UI
- Ensure workspace passes required verification gates with pnpm test and pnpm build
- Update Fusion skill docs, prompts, and capability references to use public fn_* tool names consistently
- Align engine system prompts and tool schemas for messaging/task actions with fn_send_message, fn_read_messages, fn_task_* naming
- Refresh related tests across CLI, engine, dashboard, and core to match normalized tool naming and behavior
- Add a patch changeset for @runfusion/fusion describing the skill-tool namespace normalization
Three API shifts converge here:
- DefaultResourceLoaderOptions.agentDir is required as of 0.68 (the
old process.cwd() fallback was removed). Pass getFusionAgentDir()
explicitly in createFnAgent.
- createAgentSession({ tools }) is now a string[] allowlist of tool
names, not a Tool[] array (0.68). Our boundary-wrapping via
wrapToolsWithBoundary produces Tool instances, so we can no longer
pass them through \`tools\`. Move them into \`customTools\` and
suppress the built-in defaults with \`noTools: "builtin"\`. The
wrapped tools keep the same names (read, bash, ...) as the built-ins
they replace, so no call-site or prompt changes are needed.
- SettingsManager.create's first arg (cwd) became required (was
optional before). Dashboard routes that previously passed
\`undefined\` for a process-global settings view now pass
process.cwd() to match the existing DefaultPackageManager call below.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
pi-coding-agent 0.70 replaced the mutable \`AgentState.error\` field with
a readonly \`AgentState.errorMessage\`. \`session.prompt()\` still does
not throw when retries are exhausted, so we still need to re-raise the
stored error after each prompt.
- checkSessionError (usage-limit-detector): widen parameter to accept
either key; prefer errorMessage so new sessions work, fall back to
error so we can deploy without forcing everyone's caches to rebuild.
- agent-reflection: same widening at the call site.
- pi.ts helpers: read both keys, best-effort clear both (the new field
is readonly, so the write is a no-op on 0.70 sessions but still
matters for mock sessions in tests).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
ModelRegistry's public constructor became private in pi-coding-agent 0.64.
Direct `new ModelRegistry(...)` calls no longer compile. Switch the five
production sites to the factory (`ModelRegistry.create`) and update the
four test modules that mocked the class as a constructor to now mock it
as an object with `create` and `inMemory` static methods.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Rename engine export and call sites to use createFnAgent consistently across runtime flows
- Update core lazy engine loader and dashboard agent-generation/planning/chat paths to reference createFnAgent
- Refresh affected unit and integration tests, including renaming pi-create-kb-agent.test.ts to pi-create-fn-agent.test.ts
- Update AGENTS.md documentation references to match the new createFnAgent name
- Update executor, reviewer, core prompt templates, and path-boundary messaging to reference the .fusion/memory/ directory instead of a single MEMORY.md file
- Broaden worktree boundary checks in pi path validation to allow .fusion/memory/ directory access from task worktrees
- Align memory backend metadata and dashboard backend labels to display file backend storage as .fusion/memory/
- Refresh core/engine tests to assert the new directory-based memory wording and boundary behavior
- Remove legacy .fusion/memory.md fallback references and normalize prompts/docs to .fusion/memory/MEMORY.md
- Stop legacy mirror writes and fallback reads in core memory backend and project memory flows
- Update engine worktree boundary checks and tests for canonical memory file handling
- Align dashboard memory/settings surfaces and route tests with canonical memory behavior
- Add model-favorites persistence test coverage for mission interview and new agent dialogs
- Update core and engine prompt text to reference .fusion/memory/ with MEMORY.md and daily-note guidance
- Allow worktree tool boundary access to .fusion/memory/ files while preserving legacy .fusion/memory.md compatibility
- Revise memory compaction/insight extraction messaging and related type/docs comments to use canonical memory paths
- Refresh dashboard and test expectations across core/engine/dashboard/docs for the new memory path wording
- Detect context-window-limit errors from AI sessions and automatically compact
the session conversation history before retrying (one attempt per session)
- Centralize auto-compaction in promptWithFallback (pi.ts) so executor, merger,
and step-session-executor all benefit from the same mechanism
- Remove scattered context-limit error handling from executor.ts, merger.ts, and
step-session-executor.ts in favor of the centralized approach
- Add comprehensive tests for auto-compaction retry behavior in pi.test.ts
- Remove unused compactSessionContext import from step-session-executor.ts
- Add memory note documenting the centralized auto-compaction design
- Remove unused imports across 25 files in engine package
- Remove unused variable declarations in ipc-worker.ts, child-process-runtime.ts, and mission-autopilot.ts
- Clean up unnecessary imports in agent-instructions.ts, agent-tools.ts, cron-runner.ts, executor.ts, and other modules
- Minor cleanup in notifier.ts, peer-exchange-service.ts, pi.ts, plugin-runner.ts, and other files
- Improves code quality and reduces potential confusion from unused code
- Replace blocking execSync calls with async execAsync in executor.ts for user-configured commands
- Convert self-healing.ts worktree status checks to async to avoid blocking the event loop
- Update step-session-executor.ts to use async worktree operations
- Refactor worktree-pool.ts for fully async worktree creation, cleanup, and listing
- Convert pi.ts agent session handling to async execution
- Update all corresponding tests with async/await patterns
- Add changeset for @gsxdsm/fusion patch release
