diff-review

name	diff-review
description	Multi-perspective code review on the current diff or specified files. Examines correctness, security, performance, maintainability, and scope creep. Activate when #dr appears anywhere in the user message.
argument-hint	[scope-or-empty]
disable-model-invocation	false
allowed-tools	Read, Glob, Grep, Bash, Task

Review the current changes: $ARGUMENTS

Perform a multi-perspective code review on the current diff.

Step 1 — Gather the diff:

• If git is available, run git diff and git diff --cached to see all changes.
• If scope is specified, limit review to those files.
• If no git, review recently modified files.

Step 2 — Review from 5 perspectives (ALWAYS run in parallel — dispatch as simultaneous agents):

1. Correctness — Does the code do what it claims? Logic errors, off-by-ones, null handling, edge cases, race conditions.
2. Security — OWASP top 10 risks: injection, XSS, auth bypass, secrets in code, unsafe deserialization, missing input validation.
3. Performance — N+1 queries, unnecessary re-renders, missing indexes, unbounded loops, memory leaks, large payload risks.
4. Maintainability — Naming clarity, dead code, stale comments, code duplication, overly complex logic, missing error handling.
5. Scope — Does the diff contain accidental scope expansion, unrelated changes, debug leftovers, placeholder text, or TODO items?

Dispatch specialists in parallel for deeper analysis:

• security-auditor for perspective 2 (always, not just for security-heavy diffs).
• deepsearch to trace how changed code is used elsewhere.
• test-commander to identify what tests should cover the changes.
• oracle for complex architectural concerns.

Do not do shallow inline analysis when a specialist can go deeper. The goal is thoroughness, not token savings.

Step 3 — Compile findings:

• Categorize each issue as BLOCKING (must fix before merge), WARNING (should fix), or NOTE (consider for future).
• Provide specific file paths and line numbers for each issue.
• Suggest concrete fixes, not vague recommendations.

End with:

• Summary (1-2 sentences overall assessment)
• Blocking issues (must fix)
• Warnings (should fix)
• Notes (optional improvements)
• Missing tests (what the diff should have tested)