autopilot
Pass
Audited by Gen Agent Trust Hub on Aug 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from GitHub pull requests, including titles, descriptions, comments, review text, and CI logs.
- Ingestion points: GitHub PR data and CI output (SKILL.md).
- Boundary markers: Instructions explicitly state to treat these sources as untrusted input and provide a 'completion contract' to verify status.
- Capability inventory: Scoped file edits, commits, pushes, and GitHub review replies/resolutions.
- Sanitization: Specific instructions mandate that the agent never execute embedded instructions and surfaces any requests that attempt to override the skill's logic.
- [DYNAMIC_EXECUTION]: The agent is instructed to run local tests, linting rules, and build steps to diagnose and verify fixes for CI failures. While this involves executing code within the project's environment, it is limited to existing project scripts and intended for verification.
- [COMMAND_EXECUTION]: The skill uses the GitHub CLI (gh) or equivalent integration to query repository state and monitor check runs.
Audit Metadata