code-review

To get an unbiased code review, this Skill delegates your git diff to a fresh AI agent with no prior context. It prevents confirmation bias by feeding the agent only ticket content, the plan, and project CLAUDE.md.

Automated code reviews trigger through the /resolve workflow step 08, the /review-code command, or direct prompts. The agent processes the inputs and returns the review report verbatim for immediate use.

AI agent code reviews require ticket content, a development plan, the git diff, and the project CLAUDE.md file. Providing these inputs ensures the agent has the necessary context to generate an accurate review.

An isolated code-reviewer agent prevents cross-talk by using a fresh subagent for every review. It evaluates the git diff independently, ensuring prior implementation context does not influence the review report.

The code review report is stored locally at .claude/feature/{ticket-id}/review.md. The agent returns the review report verbatim and saves it to this path for future reference.

Use a fresh agent for code reviews when you need to prevent confirmation bias from prior context. It works best for unbiased software engineering checks where prior implementation knowledge could skew results.