diagnosing-bugs
Pass
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions require the agent to execute shell commands to build and run a "feedback loop." This includes running unit/integration tests, CLI tools, and the provided
scripts/hitl-loop.template.shscript to facilitate human-in-the-loop debugging. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process potentially untrusted external data during the diagnostic process.
- Ingestion points: The agent is instructed to read application logs, HAR files, core dumps, and user-provided bug reports (Phase 1 and Phase 6).
- Boundary markers: The instructions do not define specific delimiters or warnings to ignore instructions embedded within these external artifacts.
- Capability inventory: The agent has the ability to execute shell commands, read/write to the filesystem, and perform network operations (curl) as part of the debugging workflow.
- Sanitization: There is no explicit requirement for the agent to sanitize or filter content from logs or traces before processing them.
Audit Metadata