max-effort
Pass
Audited by Gen Agent Trust Hub on Aug 18, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill instructs the agent to adopt an 'adversarial' and 'red-team' posture to 'tear down' work. While this is a functional role-play for quality assurance, it involves overriding standard helpful behavior with a skeptical persona.
- [INDIRECT_PROMPT_INJECTION]: The skill requires the agent to ingest untrusted data by instructing it to 'read the cited file' and 'open the URL' to check claims. This creates a vulnerability surface where external content could influence the agent's behavior.
- Ingestion points: Processes external URLs and local files provided in the task context.
- Boundary markers: None specified to differentiate external content from instructions.
- Capability inventory: Includes file reading, web browsing, and shell execution (tests/builds).
- Sanitization: No explicit instructions for sanitizing or escaping content from external sources.
- [COMMAND_EXECUTION]: The instructions direct the agent to 'run the tests/build/repro' to verify claims. Executing local scripts or build processes is a high-privilege capability that should be monitored, as it involves running potentially untrusted code locally.
Audit Metadata