code-reviewer

Pass

Audited by Gen Agent Trust Hub on Jul 24, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes git diff and git status commands to gather context on code changes for the review process.
  • [PROMPT_INJECTION]: Indirect prompt injection surface identified. The skill ingests untrusted data from code diffs and external project rules without utilizing explicit boundary markers or sanitization techniques. This data is then used to generate verdicts and reports which are written to the filesystem or posted to external platforms like GitLab.
  • Ingestion points: Reads from git diff output, project rules in {harness_root}/rules/*.md, and GitLab issue context.
  • Boundary markers: None defined to isolate the content being reviewed from the agent's own instruction context.
  • Capability inventory: Can write files to {product_root}/docs/versions/ and post network-based comments via the mcp-gitlab-usage tool.
  • Sanitization: No sanitization or filtering of instructions embedded in code comments or strings within the diff is specified.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 24, 2026, 05:00 PM
Security Audit — agent-trust-hub — code-reviewer