diagnosing-bugs
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external artifacts such as log dumps, HAR files, and core dumps (Phase 1) to diagnose bugs. This provides an ingestion point for potentially untrusted data.\n
- Ingestion points: Log files, HAR files, and core dumps provided by the user or captured from the environment.\n
- Boundary markers: While the skill mandates secret redaction, it does not specify the use of delimiters or specific instructions to prevent the agent from following directions embedded in the debug data.\n
- Capability inventory: The skill allows for command execution, network requests via curl, and local file operations.\n
- Sanitization: The instructions focus on redacting secrets rather than sanitizing inputs for potential injection patterns.\n- [DYNAMIC_EXECUTION]: The process involves generating and executing dynamic scripts to reproduce bugs, including HTTP scripts, headless browser automation (Playwright/Puppeteer), and custom human-in-the-loop bash scripts (Phase 1).\n- [COMMAND_EXECUTION]: The skill guides the agent through the execution of multiple system and development tools, such as
curl,git bisect, and various test runners, which is necessary for the stated purpose of debugging.
Audit Metadata