diagnosing-bugs
Pass
Audited by Gen Agent Trust Hub on Jul 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill requires the agent to execute shell commands to build reproduction loops, run tests, and perform git operations. It also includes a bash template for human-in-the-loop interactions.
- [PROMPT_INJECTION]: The skill ingests untrusted diagnostic data such as logs and traces, which presents a surface for indirect prompt injection. This is mitigated by mandatory human checkpoints at each critical phase of the debugging process.
- [INGESTION_POINTS]: Diagnostic artifacts (logs, HAR files) and documentation files referenced in Phase 1 and 2.
- [BOUNDARY_MARKERS]: The skill relies on natural language boundaries and user-led review of agent findings.
- [CAPABILITY_INVENTORY]: The agent possesses command execution and file system access across all phases of the workflow.
- [SANITIZATION]: No automated sanitization of ingested logs or diagnostic data is specified.
Audit Metadata