agent-architecture-audit
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill instructions are focused on technical auditing and debugging. No malicious patterns such as prompt injection, credential theft, or unauthorized persistence were identified.
- [INDIRECT_PROMPT_INJECTION]: The skill defines an attack surface by ingesting external data (SKILL.md: agent repositories, traces, and logs) to perform its audit functions. 1. Ingestion points: External code repositories and execution logs specified in the input contract. 2. Boundary markers: None explicitly defined in the instructions. 3. Capability inventory: The skill reads files and delegates analysis to other diagnostic tools like debug and security-audit. 4. Sanitization: No specific sanitization or escaping mechanisms for external content are described.
Audit Metadata