the-honoured-one
Pass
Audited by Gen Agent Trust Hub on Oct 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill mandates a protocol requiring the agent to read multiple files from the local environment to build context (Ingestion points: local codebase files). This increases exposure to external data that may contain malicious instructions. Evidence chain: 1. Ingestion via file reads (e.g., Phase 2); 2. Boundary markers are absent; 3. Capability involves read operations; 4. Sanitization is not specified.
- [SAFE]: The skill uses thematic role-play and strict instructional language to enforce a context-loading workflow. These instructions are focused on improving the quality of the AI's output by preventing assumptions and are not designed to bypass safety or security constraints.
Audit Metadata