interface-review

Pass

Audited by Gen Agent Trust Hub on Aug 21, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill employs Git and the GitHub CLI (gh) to resolve repository states and fetch pull request information. Commands like git fetch, git diff, and gh pr view are used to gather context for the review. These are standard operations for development tools and are restricted to read-only or metadata-retrieval purposes within the skill's workflow.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, including Git diffs, commit messages, and pull request titles/bodies.
  • Ingestion points: Data is ingested via git diff, git show, and gh pr view (referenced in SKILL.md and scope-resolution.md).
  • Boundary markers: The instructions mandate a structured analysis by resolving scope and classifying findings, which helps the agent distinguish between existing code and new changes.
  • Capability inventory: The agent's capabilities are limited to read-only repository operations and the creation of isolated worktrees (git worktree add) for specific verification tasks. There are no instructions for arbitrary file writing or network exfiltration to unknown domains.
  • Sanitization: The skill relies on structured analysis and tool-based data retrieval rather than explicit string sanitization.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 21, 2026, 01:39 PM
Security Audit — agent-trust-hub — interface-review