recipe-build

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 content from external files and user arguments to drive an autonomous execution pipeline. \n
  • Ingestion points: Untrusted data enters the agent context through the $ARGUMENTS variable and by reading markdown files from the docs/plans/, docs/plans/tasks/, and docs/design/ directories. \n
  • Boundary markers: The instructions do not define explicit delimiters or 'ignore embedded instructions' blocks when interpolating external file content into prompts for sub-agents like task-decomposer or task-executor. \n
  • Capability inventory: The skill possesses significant capabilities, including writing to the file system, deleting files in the docs/plans/tasks/ directory, and performing git commit operations. \n
  • Sanitization: There is no evidence of sanitization, escaping, or strict schema validation for the external documents before their contents are processed by the orchestrator.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 23, 2026, 10:09 AM
Security Audit — agent-trust-hub — recipe-build