diagnosing-bugs
Pass
Audited by Gen Agent Trust Hub on Aug 13, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external artifacts such as network traces, HAR files, and log dumps which are inherently untrusted and could contain malicious instructions. The risk is mitigated by explicit instructions to redact sensitive information and the passive nature of the diagnosis loop.
- Ingestion points: Processed artifacts (HAR files, event logs, core dumps) mentioned in SKILL.md.
- Boundary markers: No explicit prompt delimiters for ingested data are defined.
- Capability inventory: The agent is authorized to perform network operations (curl), file system writes, and subprocess execution (scripts/tests).
- Sanitization: SKILL.md mandates a strict redaction policy for secrets and auth headers.
- [DYNAMIC_EXECUTION]: The skill encourages the creation and execution of reproduction scripts, test harnesses, and CLI commands to verify bug behavior. This dynamic generation is restricted to the context of bug diagnosis and reproduction, and the skill provides a template (scripts/hitl-loop.template.sh) to ensure these loops are structured and human-auditable.
Audit Metadata