Create Agent
Security checks across static analysis, malware telemetry, and agentic risk
Overview
The skill appears to do what it says—create persistent Overclaw agents—but it changes core agent configuration and can read local logs when asked.
Install/use this only if you intend to change Overclaw's agent registry, prompts, and routing. Start with --dry-run, review every changed file, keep new agents least-privileged, and avoid log-based analysis unless you are comfortable with the skill inspecting local logs.
Static analysis
No static analysis findings were reported for this release.
VirusTotal
64/64 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.
A mistaken or overly broad new agent could affect how future tasks are routed or executed.
The skill intentionally updates multiple files that affect agent discovery, routing, prompts, and generated context, so a bad configuration could propagate through Overclaw behavior.
Integration points updated: `.overstory/config.yaml`, `.overstory/agent-manifest.json`, `.overstory/agent-defs/<name>.md`, `scripts/overclaw_gateway.py`, `task_router.py`, `generate_agent_context.py`, and regeneration
Run with --dry-run first, review diffs to all changed files, and keep backups or use rollback-on-fail for real changes.
If broad tools such as Bash or canSpawn are enabled, the new agent may have more authority than intended.
The script persists the new agent's tool access, capabilities, spawn permission, and constraints in the agent manifest.
agents[name] = {"file": file_name, "model": model, "tools": tools, "capabilities": capabilities, "canSpawn": can_spawn, "constraints": constraints}Default to read-only constraints, keep canSpawn false unless required, and grant only the specific tools the new agent needs.
Regeneration runs local workspace code, so a compromised or unexpected workspace script would execute during this step.
After updating configuration, the script can execute the workspace's generate_agent_context.py to regenerate gateway context.
subprocess.run([sys.executable, str(GENERATE_CONTEXT_PATH), "--workspace", str(workspace)], cwd=str(workspace), check=True, capture_output=True)
Use this only in a trusted workspace; review generate_agent_context.py or use --no-regenerate if you do not want the regeneration script run.
Local logs may contain private operational details; this script appears to summarize locally rather than transmit them.
Optional analysis mode reads local OpenClaw log files to suggest possible new agents.
OPENCLAW_LOGS = Path.home() / ".openclaw" / "logs" ... for f in sorted(log_dir.glob("*.log"))[:5]: text = _read_text(f, max_chars=50_000)Run log-based analysis only when you are comfortable letting the skill inspect those local logs, and review suggestions before creating agents from them.