scout
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill interacts with untrusted external data such as bug reports, application logs, and video screen recordings, presenting a surface for indirect prompt injection.
- Ingestion points: External bug reports (SKILL.md), server logs (reference/observability-debugging.md), and video recordings (reference/video-bug-analysis.md).
- Boundary markers: Use of structured YAML handoffs and canonical markdown report templates (reference/output-format.md) to delimit data from instructions.
- Capability inventory: Extensive use of command-line tools for profiling, debugging, and sub-agent execution (codex, agy, claude).
- Sanitization: Automated PII masking for video content and instructions to verify all evidence independently before reaching conclusions.
- [SAFE]: The skill adheres to the principle of least privilege by focusing solely on investigation and reproduction, explicitly forbidding code modifications or fix implementations.
- [SAFE]: Network operations are restricted to internal debugging endpoints (localhost) or interaction with trusted/well-known services (e.g., GitHub, Google SRE resources).
Audit Metadata