review-fix-address-bots
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses Git and Node.js to manage repository state, integrate code, and handle telemetry logging.
- Executing
gitfor status, diff, merge, commit, and push operations. - Running
node scripts/review-run-log.mjsfor canonical logging and session verification. - Running repository-native commands such as
npm testorlintto validate fixes. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data (code changes in PRs) which is then analyzed by sub-agent 'reviewers'.
- Ingestion points: Processes branch diffs, committed code, and untracked files in the workspace (SKILL.md, step 1.4).
- Boundary markers: The skill mandates including an explicit boundary in every reviewer prompt: 'Operate in read-only mode. You are advisory only. Never modify the workspace or Git/PR state...'
- Capability inventory: Possesses capabilities to write to the filesystem (to fix findings), execute shell commands (Git), and run repository scripts (tests/lints).
- Sanitization: Relies on instructional constraints (prompt-based boundaries) to limit the behavior of sub-agents processing untrusted code.
Audit Metadata