claude-intelligence

Plug-and-play Claude Code setup: hooks, agents, and settings templates that make Claude Code smarter in any repo.

What this installs

| Component | Location | Purpose | |-----------|----------|---------| | CLAUDE.md | .claude/CLAUDE.md | Global brain — coding conventions, response style, workflow rules | | settings.json | .claude/settings.json | Permissions, model config, hook wiring | | .claudeignore | .claudeignore | Keeps Claude from reading node_modules, build artifacts, secrets | | pre-commit.sh | .claude/hooks/pre-commit.sh | Blocks commits if TS check / lint / tests fail | | lint-on-save.sh | .claude/hooks/lint-on-save.sh | Auto-formats files after every Edit/Write | | claude-md-enforcer.sh | .claude/hooks/claude-md-enforcer.sh | Warns when CLAUDE.md exceeds 100 lines | | vantage-track.js | .claude/hooks/vantage-track.js | Posts Claude Code token costs to VantageAI (optional) | | agents/code-reviewer.md | .claude/agents/code-reviewer.md | Reviews code for bugs, security, performance | | agents/test-writer.md | .claude/agents/test-writer.md | Writes comprehensive test suites | | agents/debugger.md | .claude/agents/debugger.md | Systematic root-cause bug diagnosis |

Prerequisites

• Claude Code CLI installed (~/.claude/ must exist)
• Node.js 18+ (for vantage-track.js)
• bash

One-command install

bash <(curl -s https://raw.githubusercontent.com/Amanjain98/VantageAI/main/claude-intelligence/install.sh)

Run from your project root. The script is interactive — it will ask before overwriting existing files.

Manual install

1. Copy templates/CLAUDE.md → .claude/CLAUDE.md
2. Copy templates/.claudeignore → .claudeignore
3. Copy templates/settings.json → .claude/settings.json
4. Copy agents/*.md → .claude/agents/
5. Copy hooks/* → .claude/hooks/ and chmod +x .claude/hooks/*.sh
6. Edit .claude/CLAUDE.md — replace {{YOUR_NAME}} and fill in ## Project Context

Components

Hooks

pre-commit.sh — Runs before every git commit. Blocks if:

• TypeScript type check fails (npx tsc --noEmit)
• ESLint fails on staged .ts/.tsx/.js/.jsx files
• npm test fails
• ruff or pytest fails (Python projects)

lint-on-save.sh — Runs after every Edit or Write. Auto-formats:

• JS/TS: Prettier + ESLint --fix
• Python: ruff format + ruff check --fix (or black)
• Go: gofmt
• Rust: rustfmt

claude-md-enforcer.sh — Monitors CLAUDE.md length. Emits a system message if it exceeds 100 lines, prompting Claude to compress it.

vantage-track.js — Reads Claude Code session .jsonl files from ~/.claude/projects/ and posts token usage + cost data to VantageAI. Requires a free API key. Silent on errors — never breaks Claude Code.

Agents

code-reviewer — Invoke with /agent code-reviewer. Reviews the last commit for security issues, performance problems, type safety, and code quality. Reports as CRITICAL / WARNING / SUGGESTION.

test-writer — Invoke with /agent test-writer. Reads source code, identifies all code paths, and writes tests following the project's framework and patterns.

debugger — Invoke with /agent debugger. Traces bugs systematically: reproduce → isolate → root cause → fix → verify.

Settings

settings.json configures:

• Model: claude-sonnet-4-6
• Permissions: broad allow list for common dev tools, deny list for destructive commands
• Hooks: wires pre-commit, lint-on-save, and vantage-track to the right events

CLAUDE.md

A generalized "global brain" with:

• Stack preferences (Next.js, Tailwind, PostgreSQL, etc.)
• Code conventions (TypeScript strict, functional components, etc.)
• Workflow rules (branch → PR → CI → merge)
• Response style rules (concise, no filler)

Edit the ## Project Context and ## Who I Am sections to match your project.

Optional: Vantage API key

Get a free key at vantageaiops.com to track Claude Code token costs across all your projects.

The install script will prompt for it. Or set it in your shell profile:

export VANTAGE_API_KEY=vnt_your_key_here

License

MIT