self-review
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze current and historical conversation data, which are untrusted sources that could contain malicious instructions from previous sessions.\n
- Ingestion points: historical chat sessions and current conversation context as described in SKILL.md.\n
- Boundary markers: The instructions do not specify any delimiters or warnings to ignore embedded instructions within the mined data.\n
- Capability inventory: The skill identifies and proposes modifications to powerful control surfaces including CLAUDE.md, global instructions, hooks, and other skills.\n
- Sanitization: There is no explicit sanitization or filtering logic defined to prevent instructions within the history from influencing the agent's analysis.\n- [NO_CODE]: The skill body references a 'mining script' for historical session filtering but does not provide the script code, specified dependencies, or the exact execution environment, preventing a full evaluation of its behavior.
Audit Metadata