qa

Pass

Audited by Gen Agent Trust Hub on Jul 23, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill invokes the GitHub CLI via gh issue create to automate the transition from conversation to a recorded bug report.
  • [DATA_EXFILTRATION]: The skill identifies user behavior issues by reading local files and reporting them to GitHub. It employs specific constraints to prevent the leakage of sensitive internal details, explicitly prohibiting the inclusion of file paths and line numbers in the output.
  • [PROMPT_INJECTION]: The skill exposes an indirect prompt injection surface by ingesting untrusted data from both user dialogue and codebase files into the context used for generating GitHub issues. Evidence chain:
  • Ingestion points: User dialogue and codebase files (via background exploration).
  • Boundary markers: None explicitly defined beyond template adherence.
  • Capability inventory: gh issue create (write access to external repository).
  • Sanitization: Instructions to filter out file paths and code-level implementation details help minimize the impact of malicious content embedded in the codebase.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 23, 2026, 01:52 AM
Security Audit — agent-trust-hub — qa