matt-diagnosing-bugs
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions require the agent to execute various system commands, including test suites, curl requests, and project-specific CLI tools (e.g.,
tw brief,tw read) to diagnose software defects. - [INDIRECT_PROMPT_INJECTION]: The skill acts as a surface for indirect injection as it processes untrusted external data.
- Ingestion points: The agent ingests bug reports, terminal outputs, network traces, and user-provided inputs through an interactive bash script.
- Boundary markers: The instructions explicitly command the agent to redact secrets with
<REDACTED>markers, providing a conceptual boundary for sensitive data. - Capability inventory: The agent possesses the capability to execute shell commands, generate scripts, and perform file system operations.
- Sanitization: The skill mandates the redaction of secrets from all outputs and artifacts before they are processed or displayed.
- [DYNAMIC_EXECUTION]: The agent is instructed to dynamically generate and run specialized "feedback loops" at runtime, which may include temporary test harnesses, Playwright scripts, and automation for bisection processes.
- [SAFE]: The skill incorporates strong security-conscious instructions, specifically requiring the redaction of authentication headers, credentials, and secrets from any artifacts or command outputs captured during the debugging process.
Audit Metadata