Moses Coordinator
ReviewAudited by ClawScan on May 10, 2026.
Overview
This is a disclosed local monitoring daemon, but it overstates what it enforces and runs an external audit helper that is not included for review.
Only install this if you want a local background monitor for OpenClaw session sequencing. Do not assume it enforces all governance claims in the documentation. Before running it, review or install the referenced moses-governance audit_stub.py helper, install the websockets dependency intentionally, and know how to stop any launchd/systemd persistence you configure.
Findings (4)
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.
A user could rely on this as a governance enforcement layer when the reviewed code only performs a narrower sequence check.
The included coordinator.py only parses session_update events, checks the agent name against the Primary → Secondary → Observer sequence, and logs/prints sequence violations. It does not implement the documented mode, constitutional-drift, audit-before-response, or blocking checks.
detects when agents respond out of order, modes are violated, or constitutional drift occurs ... Did the response comply with the active governance mode? ... Was the audit log appended before the response?
Treat it as a sequence-monitoring logger only unless the missing enforcement checks are implemented and reviewed; update the documentation to match the actual behavior.
When a violation is observed, the coordinator will run local code that was not included in this review, so behavior depends on whatever file exists at that external path.
The script executes a helper from a separate moses-governance skill path on violations. That helper is not present in the provided file manifest, while SKILL.md says its source is included and reviewable.
AUDIT_SCRIPT = os.path.expanduser("~/.openclaw/workspace/skills/moses-governance/scripts/audit_stub.py") ... subprocess.run(["python3", AUDIT_SCRIPT, ...])Include the audit helper in the skill package or declare a pinned, reviewable dependency on the moses-governance bundle; users should inspect that file before running the daemon.
If installed persistently, the coordinator may continue running after the immediate task and keep monitoring local session events.
The artifacts explicitly describe a long-running or persistent background daemon. This is aligned with the monitoring purpose and is not hidden, but it affects ongoing local agent behavior.
Requires background process ... python3 scripts/coordinator.py & ... Or via launchd/systemd for persistent operation ... <key>KeepAlive</key><true/>
Run it manually first, and only configure launchd/systemd persistence if you understand how to unload or stop it.
The coordinator can observe local session metadata and react to messages from the local gateway.
The daemon subscribes to local gateway session events over WebSocket. This is disclosed and purpose-aligned, but the artifacts do not show authentication or event-boundary controls.
GATEWAY_WS = "ws://127.0.0.1:18789" ... await ws.send(json.dumps({"type": "subscribe", "events": ["session_update"]}))Use it only with a trusted local OpenClaw Gateway and avoid exposing the gateway port beyond localhost.