nit-pick-supreme

A unified code review system for Claude Code. Replaces separate code-level and architectural review tools with a single flag-driven invocation.

What It Does

Runs a multi-stage pipeline that combines independent analysis engines into one review:

• Static sweep -- deterministic checks (dead imports, truncated files, hardcoded secrets, lint)
• Code agents -- LLM-powered code-level analysis (frontend, backend, safety) via Claude Haiku
• Architecture agents -- structural and design analysis via Claude Opus
• Browser E2E -- Playwright test execution against a running dev server
• AI explorer -- LLM-driven exploratory browser testing via Playwright MCP
• Log sweep -- scan application logs for errors and anomalies
• Fix pipeline -- parallel fix execution in isolated git worktrees

Engines emit findings in a unified schema. The orchestrator deduplicates across engines, plans fix groups, executes fixes in parallel worktrees, and generates a report.

Prerequisites

• Claude Code CLI
• Git (with worktree support for the fix pipeline)
• Python 3.10+
• Playwright (optional, for browser engines)

Installation

Clone the repo and run the deploy script:

git clone https://github.com/Senkichi/nit-pick-supreme.git
cd nit-pick-supreme
./deploy.sh

This creates symlinks from the repo into ~/.claude/:

• ~/.claude/skills/nit-pick-supreme/ -- the orchestrator skill
• ~/.claude/agents/nps-*.md -- analysis and fix agents

On Windows, the script creates junctions instead of symlinks (works without admin privileges).

Usage

Run from any project root inside Claude Code:

/nit-pick-supreme --standard

Presets

Preset	Engines	Use Case
--quick	Static sweep only	Fast, no LLM cost
--standard	Static + code agents + E2E + fix	Everyday review
--deep	All engines + fix	Pre-merge comprehensive review

Individual Flags

Mix and match engines:

/nit-pick-supreme --static --code --fix
/nit-pick-supreme --arch --no-fix
/nit-pick-supreme --e2e --explore

Flag	Description
--static	Static sweep engine
--code	Code-level agents (Haiku)
--arch	Architecture agents (Opus)
--e2e	Browser E2E tests
--explore	AI exploratory browser testing
--fix	Enable parallel fix pipeline
--logs	Log file sweep
--no-fix	Strip --fix from any preset
--scope=diff	Only review changed files
--diff-base=REF	Git ref for diff scope
--auto-approve	Skip fix approval prompt

Output

All artifacts are written to .nit-supreme/ in the target project:

• REVIEW.md -- human-readable report with findings, fixes, and engine breakdown
• findings.json -- deduplicated findings in unified schema
• fix-plan.json -- fix groups and execution waves
• fix-results.json -- per-group outcomes
• context.json -- detected project context and baseline test state

Architecture

/nit-pick-supreme [flags]
        |
   SKILL.md (orchestrator)
        |
   +----+----+----+----+----+----+
   |    |    |    |    |    |    |
static code  arch  e2e  AI  logs
sweep agents agents     explorer
   |    |    |    |    |    |    |
   +----+----+----+----+----+----+
        |
   Synthesis & Deduplication
        |
   Fix Planning (nps-fix-planner)
        |
   Fix Execution (nps-fix-engine x N, parallel worktrees)
        |
   REVIEW.md

Engines are independent -- each is a Claude Code agent definition (agents/nps-*.md) or a Python script (skill/scripts/). The orchestrator dispatches them in parallel, collects results in a unified schema, and coordinates the fix pipeline.

Configuration

Create .nit-supreme/config.yaml in your project root to customize behavior:

default_preset: standard
test_cmd: "uv run pytest"
dev_server_cmd: "npm run dev"
exclude_paths:
  - "vendor/"
  - "generated/"

License

MIT