recipe-lesson-cleanup
Pass
Audited by Gen Agent Trust Hub on Jul 11, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it processes untrusted data from external lesson documents to derive plan deliverables.
- Ingestion points: The skill reads lesson-{lesson_id}.md in Step 1 to resolve the lesson body and actionable directives.
- Boundary markers: No explicit delimiters or 'ignore' instructions are documented when the agent interpolates extracted lesson content into the solution outline deliverables.
- Capability inventory: The skill can read and write files, manage plan documents, and log decisions using the plan-marshall toolset via the execute-script.py wrapper.
- Sanitization: No explicit content sanitization is mentioned; however, the skill incorporates 'Premise Verification' and 'Proposed-Fix Verification' steps that validate directives against the live code tree, which effectively mitigates the execution of malicious or hallucinated directives.
Audit Metadata