skill-safety-reviewer

Pass

Audited by Gen Agent Trust Hub on Aug 24, 2026

Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill runs a local script 'scripts/review_action.py' to perform safety evaluations. This script is limited to the skill's bundle and uses only standard libraries for its analysis.
  • [PROMPT_INJECTION]: The skill manages a surface for indirect prompt injection by processing external action requests. 1. Ingestion points: 'scripts/review_action.py' reads 'assets/example-request.json' and 'assets/sandbox-policy.json'. 2. Boundary markers: The 'SKILL.md' file contains clear instructions to treat external claims as untrusted and strictly prohibits execution of reviewed commands. 3. Capability inventory: The auditing script is restricted to local file reading and lacks the ability to write files, make network requests, or execute the commands it inspects. 4. Sanitization: The script implements rigorous validation including path resolution to prevent traversal, canonical origin checks for network URLs, and regex scanning for potential secrets.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 24, 2026, 04:55 PM
Security Audit — agent-trust-hub — skill-safety-reviewer