diagnosing-bugs
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process potentially untrusted external content.
- Ingestion points: User bug reports, HAR files, logs, and event traces (SKILL.md Phase 1).
- Boundary markers: The skill lacks explicit instructions for delimiting or ignoring instructions within these inputs, though it does mandate secret redaction (SKILL.md).
- Capability inventory: Shell command execution, file system access, and network operations (SKILL.md, hitl-loop.template.sh).
- Sanitization: Instructions require redacting credentials and sensitive headers before displaying captured artifacts (SKILL.md).
- [COMMAND_EXECUTION]: The skill performs shell command execution to run debugging scripts, tools, and bisection loops (SKILL.md).
- [DYNAMIC_EXECUTION]: The agent is instructed to generate and execute dynamic code harnesses and test loops to reproduce bugs (SKILL.md).
Audit Metadata