189f237a07
## Why
An AI agent autonomously deleted a live task (FN-8609) via
`fn_task_delete`. Investigating that led to two audits, which found the
gate machinery is well-built and its *enforcement* well-tested — but the
*decision boundary* was unguarded and the tools that matter never
reached the gate at all.
The headline defect: all ~76 tools in `packages/cli/src/extension.ts`
are delivered through `hostExtensionPaths` → `DefaultResourceLoader`,
which never passes through `wrapToolsWithActionGate`. `fn_task_delete`
*was* classified as gated in `gating-classifications.ts` — that
classification was dead code. Roughly 15 destructive tools were ungated
even for an operator who had locked their policy all the way down.
Second: `POST /api/approvals/:id/decision` took the approver identity
from the request body, validated for shape only, with no check that the
decider was human or differed from the requester. Combined with a bash
tool whose command string was never inspected (only `params.cwd`), an
agent could read the daemon token from `~/.fusion/settings.json`, find
its own pending request, and approve it.
## Scope
This does **not** tighten defaults. The `unrestricted` preset is
byte-for-byte unchanged — `git diff` on `agent-permission-policy.ts` is
empty — and regression tests assert that an out-of-the-box install
behaves exactly as before. The bug was never "the default is too
permissive"; it was "strict policy doesn't enforce." This makes turning
security up actually work.
The one deliberate exception: the containment that stops an agent
escalating its *own* privileges (reading the daemon token / credentials,
calling the approvals API to self-approve) applies at every preset
including `unrestricted`. That is a privilege-escalation boundary rather
than a permission preference — if it only engaged under strict policy it
would not have prevented the incident that prompted this.
## What changed
8 bisectable commits:
- **Approval lifecycle** — self-approval blocked via server-derived
deciders; same-verdict replay 409s; decide re-reads and re-validates
inside the transaction; expiry TTLs; `markCompleted` ownership check;
session identity registry in core.
- **Engine gates enforce for real** — unclassified tools resolve to a
policy-governed category instead of hardcoded `allow`; missing-policy
fail-open closed; bash containment floor + exact-command approval
binding.
- **Dashboard decision routes** — stop trusting client-supplied actors
(decision, bypass-review, worktrunk → 403 on forged actors).
- **`fn serve` authenticated by default** — auto-mints a token following
the existing `fn dashboard` precedent; `--no-auth` opts out.
- **Sibling entry points closed** — user-sourced hard-cancel moves, ACP
execute-once approvals, plugin task-store gating.
- **pi-extension principal resolution** — the extension resolves the
acting principal and can withhold or policy-gate the previously ungated
destructive tools.
- **Root-cause bonus fix** — `findLatestByDedupeKey` was broken in
PostgreSQL backend mode (already-parsed jsonb fed through a string-only
parser), so approved-grant redemption **never matched in production**,
minting duplicate requests. This explains the live DB state of 17
approved / 0 completed. *(Also cherry-picked to `main` as `a9b30013bb`,
since it is an active production defect on its own.)*
- **Review follow-ups** (`627f1b1fa8`) — operator-configured
provisioning privilege and a configurable grant TTL; see below.
## Review follow-ups
**Provisioning privilege is operator-configured, not role-derived.**
`isCallerPrivileged` had gone from `caller.reportsTo == null` (every
top-level agent privileged — permanent escalation by creating a
manager-less agent) to `caller.role === "ceo"`, which swapped an
implicit rule for a magic string: any agent config can claim that role,
while an operator who genuinely wants a privileged agent had no
supported way to say so. Privilege now derives solely from
`agentProvisioning.trustedAgentIds` / `trustedRoles` and fails closed
when settings are unresolvable.
It is also no longer forwarded to `resolveAgentProvisioningPolicy` as
`isPrivileged`, because that flag short-circuits ahead of
`alwaysApproveDelete` — a trusted caller was bypassing delete approval
entirely. The policy applies the same trusted rules itself, in the right
order. The function now governs only the org-chart escape hatch (acting
outside your own direct reports).
**Grant TTL defaults to 1 hour and is configurable.** Approval →
redemption is not instantaneous: an operator approving from their phone,
an engine restart, a queued lane, or a task waiting on a worktree all
routinely exceeded 15 minutes, after which the grant expired and the
agent silently re-requested. One hour remains far short of the
"redeemable forever" hazard the TTL exists to bound. Override via
`FUSION_APPROVAL_GRANT_TTL_MS` or `configureApprovalRequestTtls()`;
invalid overrides are ignored rather than widening the window to
infinity or collapsing it to zero.
## Behavior changes requiring operator review before rollout
1. `fn serve` requires a bearer token by default (`--no-auth` opts out);
unauthenticated clients get 401.
2. Agents can no longer run withheld destructive tools
(`fn_task_delete`, `fn_task_bypass_review`,
mission/milestone/slice/feature/workflow deletes, `experiment_finalize`,
`skills_install`). Operators keep them via CLI/dashboard. **This is the
incident fix.**
3. Agents get provisioning privilege only when the operator lists them
in `agentProvisioning.trustedAgentIds` / `trustedRoles`; the
provisioning gate is now live in production. Previously-implicit
privilege (top-level position, or a `ceo` role) no longer grants
anything on its own.
4. Decision replay 409s (was 200); pending approvals expire after 24h,
approved grants after 1h (configurable); bash approvals bind per exact
command.
5. Forged/body actors on decision, bypass-review, worktrunk routes →
403; `archive-all-done` requires `{confirm:true}` (external scripts
affected).
6. `fn_secret_get` approvals grant exactly one reveal (previously
granted nothing and looped forever); ACP approvals are execute-once
(previously infinite reuse).
7. Bash containment denies token/credential/approvals-API commands in
all agent sessions at every preset.
## Verification
Independently re-run against the branch, not just self-reported:
- 5 typechecks (core, engine, cli, dashboard `tsconfig.json` +
`tsconfig.app.json`) — clean
- `pnpm lint` — clean
- `pnpm test:gate` — 379 passed
- `pnpm build --force` — green (a plain `pnpm build` skips packages as
unchanged and does **not** compile the branch)
- `pnpm check:changesets` — clean
- ~650 file-scoped tests including new negative-path suites for the
decision boundary, which previously had **zero** test coverage
`packages/engine/src/__tests__/plugin-runner.test.ts` fails 56/80 —
**verified pre-existing**, reproducing identically at base commit
`93a403af67` on `main`. Not in the merge gate.
### A mutation check that failed to fail
Worth recording, because it nearly shipped an untested security fix. The
first mutation check on the provisioning change reintroduced the `ceo`
hardcode and **all 17 tests still passed** — the tests asserted through
the policy path, which can no longer observe `isCallerPrivileged` at
all, precisely because `isPrivileged` is no longer forwarded there.
Org-chart cases that do exercise the function were added; the hardcode
now fails exactly 1 of 19, and restoring is green. A green mutation run
is only meaningful if the test can actually see the code under test.
## Known limitations (stated, not papered over)
- The bash containment floor is string-matching: a cost-raiser, not a
sandbox. Quoting, encoding, `$HOME`, symlinks, or an interpreter
one-liner can evade it. The durable protection is the decision route
refusing agent-originated deciders — the filter is the belt, not the
braces.
- Approval expiry is lazy (evaluated at decide/complete/redeem), not
swept, so an expired pending row stays visible in lists until touched.
- The extension's require-approval path returns a pending message but
cannot suspend a pi session mid-turn; engine-side pause hooks cover
engine lanes only.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **Security**
* Hardened approval and permission gating with server-side decider
attribution, self-approval blocking, ownership checks, replay/race
protection, and status/TTL enforcement.
* Added fail-closed behavior for sensitive/unclassified tools and
sandbox provisioning approvals.
* Blocked credential/approval access via bash containment; plugin
destructive task operations now require explicit permission.
* **New Features**
* `fn serve` now defaults to bearer-token auth, with `--no-auth` as the
explicit opt-out.
* **Bug Fixes**
* Improved task move-source attribution (`moveSource: "user"`) and
tightened dashboard archive/bypass confirmation and operator attribution
behavior.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
@runfusion/fusion
From rough idea to production code — automatically.
Multi-node agent orchestrator — tasks, agents, missions, git, files, and worktrees, with any model, local or cloud.
runfusion.ai → · GitHub · Docs
Install
Zero install, straight from npm:
npx runfusion.ai
Boots the dashboard. Subcommands forward through (npx runfusion.ai task list, etc). Long form: npx @runfusion/fusion dashboard.
One-line installer (macOS & Linux — auto-picks Homebrew, falls back to npm):
curl -fsSL https://runfusion.ai/install.sh | sh
Homebrew (macOS & Linux):
brew install runfusion/fusion/fusion
Fully-qualified install auto-taps and, on Homebrew 6.0+, trusts only this formula. If short-name install fails with “untrusted tap”, run brew trust --formula runfusion/fusion/fusion then brew install fusion.
npm global:
npm install -g @runfusion/fusion
fn dashboard # or: fusion dashboard
Launch the dashboard
From a shell:
fn dashboard # or: fusion dashboard / npx @runfusion/fusion dashboard
fn dashboard --paused # start with automation paused
fn dashboard --dev # development-mode dashboard + AI engine
fn dashboard --no-engine # web UI only, no AI engine
The dashboard gives you:
- A live kanban board — tasks move through columns automatically as AI works on them
- Task detail view — generated spec, step-by-step progress, reviewer verdicts, full execution log
- Dependency-aware scheduling — declare task dependencies or let the engine infer them
- Auto-merge — on by default; reviewed work squash-merges without you lifting a finger
- Parallel execution — independent tasks run simultaneously in isolated git worktrees
- Self-sustaining board — agents may spawn follow-up tasks; the board feeds itself
Your entire dev environment. On a single pane of glass.
Describe a task in plain language. A triage agent reads your project, understands context, and writes a full PROMPT.md spec — steps, file scope, acceptance criteria. Then Fusion plans, reviews, executes, and reviews again, in an isolated git worktree, with a human approval gate wherever you want one.
One board. Controlled from anywhere. Laptop, Mac mini, Linux server, cloud VM, phone — all connected.
Run an agent company
Import a team. Run it autonomously for weeks. 440+ agents across 16 companies, wired for missions, mailboxes, and inter-agent delegation.
npx companies.sh add paperclipai/companies/gstack
How it works
You create a task with a rough description. A pipeline of specialized agents takes over.
Specification. A triage agent reads your codebase — file structure, existing patterns, related code — and turns your rough idea into a detailed spec. It breaks the work into discrete steps, identifies which files are in scope, writes acceptance criteria, and assigns a complexity rating that determines how aggressively the work gets reviewed.
Scheduling. Tasks declare dependencies on each other. The scheduler builds a dependency graph and starts work only when upstream tasks are done. Independent tasks run in parallel — each in its own isolated git worktree, so there are no conflicts during execution.
Execution & review. An executor agent works through the spec step by step in the worktree. At each step boundary, a separate reviewer agent, with read-only access, independently evaluates the work. The reviewer can approve (continue), request revisions (fix specific issues), or force a rethink (change the approach entirely). Review depth scales with the task's complexity rating: trivial tasks get light checks, complex tasks get thorough multi-pass review.
Merge. When execution finishes and the reviewer signs off, the task moves to In Review:
- Direct merge (default) — automatically squash-merges the completed task branch into your current branch with a clean commit.
- Pull request — automatically creates or links a GitHub PR, waits for reviews/checks, then merges once policy conditions are satisfied.
autoMerge controls whether Fusion performs completion automatically. If disabled, tasks stay in In Review until you finish the merge yourself. For PR-first mode, authenticate GitHub with gh auth login.
Tasks flow through: Triage → Todo → In Progress → In Review → Done.
This execution model is heavily based on Taskplane.
What makes it different
| 🧠 AI specification | Rough idea in, detailed PROMPT.md out — steps, file scope, acceptance criteria. |
| 🔁 Workflow gates | Plan → Review → Execute → Review on every step. Block or pass automatically. |
| 🌳 Worktree isolation | Each task runs in its own branch and worktree. Parallel tasks. Zero conflicts. |
| ⚡ Smart merge | Passing every gate? Fusion squash-merges and moves on. |
| 🛰️ Multi-node mesh | Laptop, server, cloud, phone — all synced. Desktop, mobile, web. |
| 🧩 Any model | Anthropic, OpenAI, Ollama, and more. |
| 🏢 Agent companies | Import pre-built teams — 440+ agents across 16 companies. |
| 📬 Inter-agent messaging | Built-in mailbox between agents. Delegate, clarify, coordinate. |
| 🗺️ Missions | Hierarchical planning with autopilot and validation contracts. |
| 🔓 Open source. MIT. | No vendor lock-in. Run it on your own hardware. |
Working from chat
Manage tasks without leaving the conversation:
"Every ten minutes, analyze the server code for logic the client hasn't implemented yet and create tasks. Tasks may spawn additional tasks, so just add enough to keep the board saturated."
"Create a Fusion task to fix the login redirect bug"
"Add a task for dark mode support, it depends on FN-003"
"What's the status of FN-042"
"Attach screenshot.png to FN-007"
"Pause FN-012 — I want to add more context first"
The Fusion extension exposes tools to create tasks, check progress, attach files, and pause or resume automation.
Standalone CLI
See STANDALONE.md for additional installation and usage options.
Optional provider: Factory AI via Droid CLI
@runfusion/fusion now ships a vendored @fusion/droid-cli extension in the published CLI bundle.
To use it:
- Install the
droidbinary and ensure it is on yourPATH - Authenticate with Droid CLI (
droid auth login) - In Fusion dashboard, go to Settings → Authentication and enable Factory AI — via Droid CLI
- Restart Fusion when prompted so the extension is loaded into the runtime
Once enabled, droid-cli models appear in Fusion model selection.
Maintainer note: workspace plugins in published CLI bundles
When CLI or dashboard runtime code imports workspace plugin packages (for example @fusion-plugin-examples/roadmap), those imports must stay statically analyzable and covered by packages/cli/tsup.config.ts noExternal rules so plugin runtime code is inlined into dist/bin.js.
Do not introduce dynamic or variable module specifiers for workspace plugin runtime paths in the published execution path. If a workspace plugin is needed for bundled auto-install, stage a bundled plugin entry (dist/plugins/<id>/bundled.js) rather than copying raw TypeScript source into dist/.
Full documentation
Architecture details, development setup, and contributor info live in the project README.
License
MIT — see LICENSE.