diagnose
Pass
Audited by Gen Agent Trust Hub on Jun 24, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill processes user-provided bug reports and error messages captured via Phase 1 feedback loops and the scripts/hitl-loop.template.sh script. This creates a surface for indirect prompt injection, as malicious text within an error report could attempt to manipulate the agent's logic. In the context of a debugging tool, this is a low-risk inherent characteristic.\n
- Ingestion points: bug reports described in SKILL.md and user input captured in scripts/hitl-loop.template.sh.\n
- Boundary markers: No explicit boundary markers or ignore instructions for external data are specified.\n
- Capability inventory: The skill utilizes subprocess calls for running tests, curl for network requests, and bash for interactive loops.\n
- Sanitization: There is no mention of sanitizing or escaping the data ingested from user reports.\n- [COMMAND_EXECUTION]: The skill involves the execution of various shell-based commands and standard development tools such as git bisect and curl. It also executes the provided scripts/hitl-loop.template.sh for interactive feedback. These operations are necessary for the diagnostic tasks described and are used within a appropriate scope.
Audit Metadata