code-review

Warn

Audited by Gen Agent Trust Hub on Jul 27, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill dynamically extracts and runs commands defined in the target repository's configuration files (e.g., manifest scripts) and documentation. A malicious repository could specify dangerous shell commands that the agent would execute during the review process.
  • [REMOTE_CODE_EXECUTION]: The skill instructions mandate running the project's own validation gates, such as tests and builds, on the PR branch. This results in the execution of untrusted code provided within the pull request.
  • [PROMPT_INJECTION]: The skill processes untrusted documentation and PR content to determine execution steps. Ingestion points: README.md, CLAUDE.md, docs/ files, and PR diffs. Boundary markers: Absent; instructions emphasize following rules found in project documentation as authoritative. Capability inventory: Git worktree operations, GitHub CLI for PR and project management, and execution of various language toolchains. Sanitization: No validation or filtering is applied to the commands or data extracted from the untrusted repository files before they are used in operations.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 27, 2026, 12:23 AM