Security audit
Security checks for vulnerabilities and agentic risk
This skill is a coherent task router that uses file paths and optional LLM classification to choose specialist prompts, with no evidence of persistence, exfiltration, destructive behavior, or hidden installation behavior.
Before installing, confirm that your OpenClaw LLM endpoint and model routing are acceptable for the sensitivity of your task text and file paths, especially for proprietary code or client work. The skill itself appears narrowly scoped to routing and prompt selection.
The skill sends raw task text to an external LLM during stage 2 classification, but the description does not disclose that user input may leave the local boundary. In a routing skill, tasks can contain proprietary code requests, credentials, incident details, or other sensitive data, so undisclosed third-party transmission creates a real privacy and compliance risk.
Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.
'.xcworkspace', '.m', '.h'}
if any(any(f.endswith(ext) for ext in ios_extensions) for f in files):
return {"domain": "ios", "hard_gate_triggered": True, "confidence": "high"}
if 'info.plist' in [Path(f).name.lower() for f in files]:
return {"domain": "ios", "hard_gate_triggered": True, "confidence": "high"}
# Strong iOS keywords
The skill returns the full specialist system prompt as an output field, which exposes internal prompt engineering and control instructions to downstream callers. While not inherently a direct exploit, this unnecessarily leaks implementation details that can help attackers tailor prompt-injection attempts or evade routing and policy behavior.
No suspicious patterns detected.