Project Manager Agent
Security checks across static analysis, malware telemetry, and agentic risk
Overview
No hidden or malicious behavior is evident; the skill openly monitors OpenClaw sub-agents and can automatically nudge stalled ones when invoked or scheduled.
This skill appears safe for its stated purpose if you want automated supervision of sub-agents. Before installing or scheduling it, review the staleness threshold, max_steer_per_run, cooldown, and steer_message settings, and avoid --check_all_sessions or --no_cooldown unless you specifically want broader or more frequent steering.
Static analysis
No static analysis findings were reported for this release.
VirusTotal
66/66 vendors flagged this skill as clean.
Risk analysis
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.
When run, the skill can nudge active sub-agents to continue working without a separate prompt for each message.
The skill directs the user's agent to message other sub-agent sessions based on locally generated session keys. This is disclosed, capped, and central to the purpose, but it is still an inter-agent control channel.
For each session key in `stalledSessionKeys` (cap at 5 per run to avoid spam), call **sessions_send**(`sessionKey`, `steerMessage` ... `0`) (fire-and-forget; `timeoutSeconds=0`).
Use it only if you want automated sub-agent steering; review the steer message, maximum sends per run, and cooldown before scheduling it.
Session names, IDs, update times, and cooldown history can influence what the skill reports and which sub-agents it recommends steering.
The monitor reads OpenClaw session state and writes persistent steering cooldown state. This is scoped to OpenClaw and supports the purpose, but local session metadata and stored cooldown data affect future behavior.
SESSIONS_JSON = OPENCLAW_HOME / "agents" / "main" / "sessions" / "sessions.json" STEER_STATE_PATH = OPENCLAW_HOME / "logs" / "project-manager-agent.steer_state.json"
Run it only against a trusted OpenClaw home directory and keep the default sub-agent-only scope unless broader session monitoring is intended.
If configured as a cron job, it may repeatedly check and steer sub-agents on a schedule.
The skill recommends recurring autonomous operation. This is an explicit optional integration rather than hidden persistence, but scheduled use would continue operating until the user disables the cron.
This skill is ideal for a cron job, running periodically (e.g., every 5 minutes). The agent runs the script with `--json`, steers stalled subagents via sessions_send, and announces the status.
Only create the cron job if you want continuous monitoring, and set a schedule, staleness threshold, cap, and cooldown that match your workflow.