bug-fix
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted third-party content such as bug reports and issue descriptions, which could contain malicious instructions meant to influence the agent's behavior.
- Ingestion points: Phase 1 and Phase 2 describe processing external bug reports and reproduction steps.
- Boundary markers: The skill includes a 'HARD-GATE: Input Integrity' section explicitly instructing the agent to treat reports as untrusted and disregard embedded instructions.
- Capability inventory: The agent performs file writes (fixing code/writing tests) and shell command execution (
bundle exec rspec). - Sanitization: The instructions require the agent to extract only factual context and verify all claims against actual code and test output.
- [COMMAND_EXECUTION]: The skill performs shell command execution to verify fixes and check for regressions.
- Evidence: Execution of
bundle exec rspecin Phase 4.
Audit Metadata