diagnose-ci-failures

Pass

Audited by Gen Agent Trust Hub on Nov 19, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources.\n
  • Ingestion points: GitHub Actions logs are fetched via gh run view <run-id> --log-failed (SKILL.md).\n
  • Boundary markers: No specific delimiters or 'ignore embedded instructions' warnings are provided to separate log content from instructions.\n
  • Capability inventory: The skill uses gh for read operations. Although the skill itself is read-only, the agent environment may include tools for code modification or command execution that could be targeted.\n
  • Sanitization: No log sanitization or filtering is performed before processing.\n- [COMMAND_EXECUTION]: The skill utilizes the gh (GitHub CLI) tool to perform its primary function.\n
  • Evidence: Commands include gh run view, gh run list, and gh pr view to fetch workflow and pull request information.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 19, 2026, 01:13 AM