review-full

This is a COMPREHENSIVE multi-agent code review. For quick anti-pattern scanning (seconds, not minutes), use review-diff instead.

Step 0: Detect scope and project

• If $ARGUMENTS is empty: review uncommitted changes (staged + unstaged via git diff + git diff --cached)
• If $ARGUMENTS is a file path: review that file only
• If $ARGUMENTS is a branch or range: review diff against that ref
• If $ARGUMENTS is "security": run security-only review (skip to step 3)
• Detect project type from cwd/CLAUDE.md and the project manifest (any language/framework), or fall back to inspecting the file structure

Step 1: Spawn review agents in parallel

Launch up to 4 agents based on what the changes touch:

If changes are small (<50 lines), run code-reviewer only. If security argument, run security-reviewer only. Add architecture-reviewer when the change reshapes structure (new module boundaries, files moved across layers, a refactor spanning several directories) rather than editing within existing files.

Step 2: Code quality review (via code-reviewer agent)

The agent checks:

• Readability and naming conventions (matches project patterns?)
• DRY: duplicated logic that should be extracted
• Error handling: all async paths covered? Consistent error shapes?
• Component/function size: single responsibility?
• Static-typing discipline where the language has it: no unchecked escapes (TypeScript any, Java raw Object, Go interface{}, etc.)