resolve-bug
Pass
Audited by Gen Agent Trust Hub on Aug 7, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted user-reported bug descriptions and complaints, which could potentially contain malicious instructions. The risk is managed by a structured pipeline and human-in-the-loop oversight.
- Ingestion points: Verbatim user reports and expectations are gathered in
references/phase-1-intake.md. - Boundary markers: The skill uses structured markdown templates and specific headers (e.g.,
Reported:,Established:) to segregate user input from the agent's logic. - Capability inventory: While the skill can modify code (Phase 5) and draft communications (Phase 6), it explicitly requires human approval at 'gates' in Phase 2 and Phase 4 before proceeding with these actions.
- Sanitization: No specific input sanitization is mentioned; however, the requirement for human verification of findings and verdicts serves as a primary control.
Audit Metadata