bug-fix

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external bug reports which are untrusted inputs that could contain malicious instructions meant to override agent behavior.
  • Ingestion points: External bug reports, production issues, and regression descriptions (SKILL.md).
  • Boundary markers: The skill implements a mandatory "Input integrity" hard gate requiring the agent to treat embedded instructions as data and ignore them.
  • Capability inventory: Uses shell commands for reproduction and verification, specifically mix test, mix format, and mix credo (SKILL.md).
  • Sanitization: Instructions mandate extracting only factual details (errors, stack traces) and verifying all claims against verifiable code and test output.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 05:28 AM
Security Audit — agent-trust-hub — bug-fix