he-code-review

Pass

Audited by Gen Agent Trust Hub on Jun 18, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: The skill instructions (Procedure #3) and risk profile (contract.yaml) explicitly identify that reviewer comments and PR bodies are untrusted. It directs the agent to re-verify findings and refuse unsafe instructions (e.g., 'ignore previous instructions'), effectively mitigating indirect prompt injection risks.
  • [DATA_EXFILTRATION]: The skill includes constraints to redact secrets and verify redaction status for session traces before approval. No network exfiltration or unauthorized file access patterns were identified.
  • [COMMAND_EXECUTION]: The execution boundaries restrict mutation authority (autofix/PR update) to explicitly authorized modes. The evaluation suite (evals.yaml) contains deterministic checks that forbid dangerous utilities like curl, wget, and nc.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 18, 2026, 08:18 PM
Security Audit — agent-trust-hub — he-code-review