merge-council

Pass

Audited by Gen Agent Trust Hub on Nov 10, 2026

Risk Level: SAFEREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill performs a dependency installation (e.g., 'frozen-lockfile install') and runs focused checks like tests and linters on the pull request being reviewed. If the repository or pull request contains malicious configuration files (such as npm lifecycle scripts in package.json), this could lead to arbitrary code execution in the agent's environment.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from pull requests, including code, titles, and comments, which could contain malicious instructions designed to manipulate the 'council members' or the 'fixer' agent.
  • Ingestion points: Pull request metadata, code diffs, comments, and internal instruction files (e.g., AGENTS.md, CLAUDE.md) are ingested into the context of the coordinator and member agents.
  • Boundary markers: The skill uses Markdown blocks to separate briefs but lacks explicit delimiters or high-pressure instructions to ignore all embedded commands within the PR data.
  • Capability inventory: The skill has the ability to write files, push code to repositories ('git push'), and merge pull requests ('gh pr merge'), which could be abused if an injection is successful.
  • Sanitization: The skill includes a robust mitigation that forces a 'NEEDS_HUMAN' verdict if any 'instruction files' (like AGENTS.md) are modified in the PR, preventing the PR from steering its own reviewers.
  • [COMMAND_EXECUTION]: The skill utilizes several shell-based tools including 'git' and 'gh' (GitHub CLI) to manage the review process. It uses PR-controlled data such as branch names and repository identifiers in these commands, which presents a surface for potential command injection if the underlying platform does not properly sanitize shell arguments.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 10, 2026, 08:47 PM