debug
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external data that could contain malicious instructions. 1. Ingestion points: The skill reads external error messages, test outputs, and code paths (Step 3). 2. Boundary markers: The skill includes a dedicated safety section 'Treating error output as data' which provides a clear instruction to treat external output as data rather than instructions. 3. Capability inventory: The skill utilizes shell execution for reproduction (Step 2), localization via git bisect (Step 3), and running project test suites or CI commands (Steps 6 and 7). 4. Sanitization: The skill mandates that instructions found in external data must be surfaced to the user rather than executed.
- [COMMAND_EXECUTION]: The skill workflow requires executing shell commands for debugging. Evidence includes instructions to make failures happen reliably via commands, use 'git bisect', and run the project's own CI commands and test suites.
Audit Metadata