ethos-explore
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill uses tools like Read, Grep, and Glob to ingest data from the local repository (e.g., README.md, design docs) to infer the agent's mission and behavioral rules. This creates an attack surface where malicious content in scanned files could attempt to influence the agent's logic.
- Ingestion points: The skill instructions in SKILL.md (Step 1) direct the agent to scan the codebase and documentation to fill schema fields.
- Boundary markers: The skill incorporates strong procedural boundaries, explicitly stating that implementation inferences are not final and requiring a mandatory user interview with at least three questions to confirm intent.
- Capability inventory: The skill is restricted to standard filesystem tools (Read, Glob, Grep, Bash) for local codebase exploration.
- Sanitization: The skill relies on a human-in-the-loop verification process where the user must review and approve a full draft of the generated ethos.md before completion, mitigating the risk of processing malicious or incorrect data.
Audit Metadata