diagnose
Pass
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as defect descriptions, logs, and stack traces, which serves as a potential vector for indirect prompt injection.
- Ingestion points: The 'Intake and Causal Evidence' section in SKILL.md describes the ingestion of logs, error reports, and stack traces.
- Boundary markers: The 'Operating Invariants' section in SKILL.md restricts investigation writes to a local '.agents/scratch/' directory and explicitly forbids access to sensitive files such as .env, .key, or .pem files.
- Capability inventory: The skill is capable of reading production source code and executing scripts created within the scratch directory.
- Sanitization: SKILL.md mandates that the agent redact secrets before logging evidence and use sanitized payloads for replaying events.
- [DYNAMIC_EXECUTION]: The skill generates and executes local reproduction scripts based on user-provided templates.
- Evidence: SKILL.md instructs the agent to adapt the 'scripts/hitl-loop.template.sh' file within the scratch directory to establish feedback loops for debugging.
Audit Metadata