Files

gsxdsm 189f237a07 fix(security): make the approval and permission gates actually enforce (#2457 )

## 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>

2026-07-30 21:50:37 -07:00

src

fix(security): make the approval and permission gates actually enforce (#2457 )

2026-07-30 21:50:37 -07:00

AGENTS.md

FN-6457: add ACP ask runner and bundled Claude bridge setup

2026-06-15 02:30:41 -07:00

CHANGELOG.md

chore(release): v0.74.0-beta.5

2026-07-26 18:11:47 -07:00

manifest.json

feat(acp): scaffold fusion-plugin-acp-runtime (U1)

2026-06-03 08:35:51 -07:00

package.json

chore(deps-dev): bump vitest from 4.1.8 to 4.1.10 (#2446 )

2026-07-27 19:40:08 -07:00

README.md

FN-6457: add ACP ask runner and bundled Claude bridge setup

2026-06-15 02:30:41 -07:00

tsconfig.json

feat(acp): scaffold fusion-plugin-acp-runtime (U1)

2026-06-03 08:35:51 -07:00

vitest.config.ts

FN-6042: raise dependency security floors

2026-06-08 15:19:27 -07:00

README.md

@fusion-plugin-examples/acp-runtime

A Fusion runtime plugin that drives any external Agent Client Protocol (ACP) agent over JSON-RPC/stdio. One integration unlocks every ACP-compatible agent (Gemini CLI, the Claude Code ACP adapter, and any future agent that speaks the protocol) through the standard protocol instead of a bespoke per-CLI integration.

Selected via runtimeId: "acp". Installed on demand (experimental) — see the Fusion plugin catalog (fn plugin install fusion-plugin-acp-runtime).

Security posture

The ACP agent is an untrusted subprocess that calls back into Fusion for permissions and filesystem access. This plugin enforces a defense-in-depth floor:

Per-category permission gating. Each session/request_permission is classified by tool kind into a Fusion action category and checked against the live permission policy — never a preset shortcut. allow_once only (never a persisted blanket grant). Unmappable kinds and missing policy default-deny.
Unrestricted-risk acknowledgement (acpAllowUnrestricted). Because the shipped default policy is unrestricted (allow-all), a blanket allow on a sensitive category is escalated to approval unless the user explicitly sets acpAllowUnrestricted: true. Prefer running the ACP runtime under an approval-required policy.
Filesystem jail. fs/read_text_file / fs/write_text_file are opt-in (acpFsRead / acpFsWrite, writes default OFF), confined to the session cwd by a real symlink-resolving jail (realpath + O_NOFOLLOW), with a deny-list for secrets (.env, *.pem, …) and git internals (.git/**). Writes are gated through the file_write_delete permission category.
Untrusted-input bounds. Streamed output is sanitized (ANSI/control strip) and bounded (per-turn + per-chunk caps; bounded tool-call correlation map).
Subprocess isolation. The agent env is built from an allow-list (acpEnvAllowList) — inherited process.env is not forwarded.

Not sandboxed in v1: the agent's own process/network syscalls run with Fusion's user privileges (OS-level sandboxing is recommended future work).

Settings

Key	Default	Meaning
`acpBinaryPath`	`acp-agent`	Agent binary to spawn
`acpArgs`	`[]`	Args that launch the agent in ACP/stdio mode (e.g. `["--acp"]`)
`acpModel`	—	Optional model identifier reported via `describeModel`
`acpFsRead`	`false`	Advertise/register `fs/read_text_file`
`acpFsWrite`	`false`	Advertise/register `fs/write_text_file` (gated)
`acpEnvAllowList`	`[]`	Env var names forwarded to the agent subprocess
`acpAllowUnrestricted`	`false`	Acknowledge the untrusted-agent risk under an allow-all policy

Upstream / third-party integration evidence

Per AGENTS.md (External-integration evidence):

Protocol homepage / docs: https://agentclientprotocol.com
Upstream protocol repo: https://github.com/agentclientprotocol/agent-client-protocol
TypeScript SDK repo: https://github.com/agentclientprotocol/typescript-sdk
Dependency (npm): @agentclientprotocol/sdk — https://www.npmjs.com/package/@agentclientprotocol/sdk
Pinned release: 0.24.0 (Apache-2.0)
Tarball: https://registry.npmjs.org/@agentclientprotocol/sdk/-/sdk-0.24.0.tgz
Integrity (sha512): sha512-vvu9appvGvfYstBj19C6NCepV6SvUhY5VRv60KUZ4XzhTah/olOYul5Zo4C+x2enyshMSvgB2mm/OEmrsHaSmA==
Agent binaries driven: user-supplied ACP agents (e.g. gemini --acp) and the bundled Claude bridge below. User-configured agents remain upstream-pending-verification per agent.

Bundled Claude ACP bridge evidence

Canonical upstream repo URL: https://github.com/moabualruz/claude-code-cli-acp
Docs / homepage URL: https://github.com/moabualruz/claude-code-cli-acp#readme
Release / download URL: npm package claude-code-cli-acp (version 0.1.1) — https://www.npmjs.com/package/claude-code-cli-acp
Binary / CLI name: claude-code-cli-acp
Checksum: sha512-qpfRGOXkOs9mqI7oumsGistWisyXcCC0r7ng7wdLvGMIORdzHjmUUa+94Jftgr/NYAVnAUe6N7kimD8PaO3D5g== (from pnpm-lock.yaml for claude-code-cli-acp@0.1.1)
Pinned-commit spot-review: tag v0.1.1 points to commit c93f4f4ca449f451d9f3b7db536caf4060883da9 (annotated tag ca33404fc1128d6a88a55b248f042f70b4bc9f9a, unsigned). License: Apache-2.0. Behavior reviewed for this integration: runs claude through a PTY, reads transcript JSONL, exposes an ACP server over stdio, and requires @anthropic-ai/claude-code installed + authenticated.

See docs/acp-contract.md for the launch/readiness contract and failure taxonomy.