resolve-agent-reviews
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill uses package runners (npx, pnpm dlx, yarn dlx, bunx) to fetch and execute the 'agent-reviews' tool directly from the npm registry at runtime.
- [EXTERNAL_DOWNLOADS]: Dependencies are fetched dynamically from the npm registry. The 'agent-reviews' package is a vendor-owned resource matching the skill author's infrastructure.
- [COMMAND_EXECUTION]: The skill is granted permission to run several shell commands including
git config,git add,git commit, andgit push. This allows the agent to modify the local environment's git configuration and push code changes to the remote repository. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from PR bot comments (e.g., Copilot, CodeRabbit) and uses that data to influence code modifications.
- Ingestion points: Fetches bot comments via
npx agent-reviews --bots-only --unanswered --expanded(SKILL.md). - Boundary markers: The skill lacks explicit technical delimiters for bot input but includes high-level instructions to "Verify findings before fixing" and distinguish between true/false positives.
- Capability inventory: The skill can write files to temporary directories, perform network requests via the CLI, and execute
git pushto modify the codebase. - Sanitization: The instructions require the agent to ask the user for confirmation if a finding is "UNCERTAIN" and emphasizes not guessing on architectural or business logic.
Audit Metadata