recipe-review

Pass

Audited by Gen Agent Trust Hub on Sep 23, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted local files (implementation code and design documentation) which are then used as context for sub-agent tasks.
  • Ingestion points: Source code, tests, and configuration files identified via git diffs in SKILL.md Step 1.
  • Boundary markers: Uses specific serialization rules ('mechanical extraction') and structured JSON objects to separate external data from agent instructions.
  • Capability inventory: The skill has the capability to modify the repository via the task-executor sub-agent (Step 6).
  • Sanitization: Requires explicit human approval through a decision menu (Step 4) and uses a diff-mapping process to prevent unauthorized or accidental code changes.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 23, 2026, 10:09 AM
Security Audit — agent-trust-hub — recipe-review