Firm Acp Bridge

SuspiciousAudited by ClawScan on May 10, 2026.

Overview

This skill needs review because it directs agents to store session keys, copy provider API keys into spawned agents, and schedule host-level cron commands outside the sandbox.

Install only if you trust the ACP/MCP tool implementation and need host-level autonomous session management. Before use, verify the external packages, use scoped disposable API keys, require manual approval for cron jobs, confirm where session keys are stored, and ensure there is a clear way to audit and remove scheduled tasks and persisted sessions.

Findings (5)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

What this means

A scheduled command could run on the user's host outside the sandbox, affecting files or processes beyond the intended isolated agent session.

Why it was flagged

The skill explicitly works around sandbox-denied cron tooling by scheduling commands on the main host-access session.

Skill content
Cron tools sur la denylist sandbox ... Solution : Planifier sur la session `main` (accès hôte) via `fleet_cron_schedule`.
Recommendation

Require explicit user approval for every host cron entry, prefer sandboxed execution, restrict commands to a reviewed allowlist, and provide clear list/remove/audit controls.

What this means

Spawned agents could gain access to paid model-provider accounts or other provider privileges if keys are copied broadly or into the wrong session.

Why it was flagged

The skill instructs users to propagate provider API keys into spawned or cron-driven non-main sessions, while the registry declares no env vars or credentials.

Skill content
`fleet_session_inject_env` — injecte provider env vars dans sessions non-main ... `ANTHROPIC_API_KEY`: `<your_key>`
Recommendation

Use least-privilege, per-session, revocable tokens; declare required credentials in metadata; inject only with explicit user approval and only into named trusted sessions.

What this means

If the persisted session store is readable by unintended users or agents, they may be able to resume or interfere with ACP sessions.

Why it was flagged

Gateway session keys are authentication/session material; persisting them to disk is high-impact, and the artifact does not specify file path, permissions, encryption, or access controls.

Skill content
`acp_session_persist` — persiste run_id → gateway_session_key sur disque
Recommendation

Document and enforce a restricted storage path, file permissions, retention limit, encryption or OS keychain use, and safe deletion of stale session keys.

ConcernHigh Confidence
ASI10: Rogue Agents
What this means

Automated tasks may continue running after the original request, potentially using credentials or modifying the host environment later.

Why it was flagged

The documented workflow creates recurring autonomous activity on the main session, but the artifact does not describe lifecycle controls such as listing, disabling, or cleanup.

Skill content
`fleet_cron_schedule`, `schedule`: `0 9 * * 1-5`, `session`: `main`
Recommendation

Add explicit lifecycle management: show created jobs, require confirmation, set expiration by default, and document how to disable or remove scheduled tasks.

What this means

Users cannot verify from these artifacts how the credential injection, session persistence, locking, or cron scheduling tools are implemented.

Why it was flagged

The high-impact tools are referenced as external dependencies, but the submitted package has no install spec, no code files, no capability tags, and no reviewed implementation for those tools.

Skill content
requires: - mcp-openclaw-extensions >= 2.0.0 - "@agentclientprotocol/sdk >= 0.14.0" ... Tools activés
Recommendation

Provide reviewed implementation files or pinned install specifications, declare capabilities and credentials in registry metadata, and pin trusted dependency versions or checksums.