diagnosing-bugs
Pass
Audited by Gen Agent Trust Hub on Jul 12, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions in
SKILL.mdand the provided scriptscripts/hitl-loop.template.shinvolve executing shell commands, tests, and CLI tools as part of the debugging workflow. - [PROMPT_INJECTION]: The skill processes untrusted external data, creating an attack surface for indirect prompt injection.
- Ingestion points: Application logs, error messages, and network traces ingested in
SKILL.mdduring Phase 1. - Boundary markers: None defined for the ingested diagnostic data.
- Capability inventory: Shell command execution, script generation, and network requests via curl.
- Sanitization: No filtering or validation of diagnostic input is specified.
Audit Metadata