Project Manager Agent

Monitors active sub-agents, detects stalls, provides status updates, and steers stuck agents via sessions_send.

This skill appears to do what it says: it reads the local OpenClaw sessions.json (default ~/.openclaw/...) to detect stalled sub-agents, prints a summary/JSON, and asks the agent to call sessions_send to steer them. Before installing, confirm you are comfortable with the skill reading session files (which may contain transcripts or sensitive state) and writing a steer_state file under OPENCLAW_HOME/logs. If you plan to run it as a cron job, ensure the cadence and max_steer_per_run are appropriate so it doesn't spam sub-agents. Also verify that the agent invoking this skill has only the sessions_list/sessions_send privileges you intend. If you want to audit behavior, run the script locally once (with --json and inspect output) in an isolated environment to confirm it behaves as expected.

Type: OpenClaw Skill Name: project-manager-agent Version: 0.1.0 The OpenClaw AgentSkills skill bundle is designed to monitor and steer sub-agents. The `SKILL.md` provides clear instructions for the agent to run the `project_manager.py` script, parse its JSON output, and use the `sessions_send` tool to interact with stalled sub-agents. The `scripts/project_manager.py` script reads OpenClaw's internal `sessions.json` to determine sub-agent status and manages its own state in `project-manager-agent.steer_state.json` within the OpenClaw logs directory. There is no evidence of data exfiltration, unauthorized command execution, persistence mechanisms, or malicious prompt injection attempts. All actions are consistent with the stated purpose of a project management agent.