diagnosing-bugs
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill prioritizes data security by instructing the agent to redact all secrets, environment variables, and authentication headers before displaying command outputs or captured artifacts.\n- [COMMAND_EXECUTION]: The skill relies on executing local shell commands, test runners, and browser automation tools to establish reproduction loops. These activities are central to the debugging process and do not involve unauthorized access or external exfiltration.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data sources.\n
- Ingestion points: Captured artifacts such as log dumps, HAR files, and core dumps referenced in SKILL.md.\n
- Boundary markers: Instructions to quote only specific lines carrying the signal and ignore irrelevant parts of the artifacts.\n
- Capability inventory: Shell access for running tests, curl for API debugging, and browser automation via Playwright/Puppeteer (SKILL.md).\n
- Sanitization: Proactive instructions to redact all credentials and use environment variables for sensitive data.\n- [DYNAMIC_EXECUTION]: The skill generates and executes local reproduction scripts, such as hitl-loop.template.sh, to provide a structured interface for human-in-the-loop debugging. These scripts are simple utility wrappers and do not execute untrusted remote code.
Audit Metadata