claudex (binxai-claudex)

Set up Claude Code for any project in one command

Python CLI that analyzes your project and installs a complete .claude/ configuration: project-specific CLAUDE.md, enforcement hooks, slash commands, and a full multi-agent layer.

Installation

pip install binxai-claudex

Windows note: If the claudex command isn't found after install, run python -m claudex or add your Python Scripts directory to PATH.

Quick Start

# Standard setup (auto-detects stack)
claudex init /path/to/project --yes

# With multi-agent layer (orchestrator + 12 specialist agents)
claudex init /path/to/project --yes --multi-agent

# See what would be detected without making changes
claudex info /path/to/project

# Update templates without regenerating CLAUDE.md
claudex update /path/to/project

# Health check an existing .claude/ setup
claudex validate /path/to/project

Why Multi-Agent?

Each context handoff between agents carries a ~5% degradation risk in quality and context fidelity. Across a 10-agent pipeline: 0.95^10 = 60% effective throughput.

ClaudeX's --multi-agent layer addresses this with a structured Context Handoff Protocol — a machine-readable format that passes task state, decisions, file ownership, and done conditions between agents. Every handoff is explicit, auditable, and zero-ambiguity.

Features

Smart Detection

Analyzes your project to detect:

• Language (Python, TypeScript, JavaScript)
• Framework (FastAPI, Django, Flask, Next.js, React, Vue)
• Package manager (uv, poetry, pip, npm, pnpm, yarn)
• Database (PostgreSQL, MySQL, MongoDB, SQLite)
• Infrastructure (Docker, CI, Git)

Architecture Enforcement (Stack-Aware)

claudex init writes real layer enforcement rules into pre-tool-use.py based on your detected stack.

Python / FastAPI:

LAYER_CONFIG = {
    "src/core/": ["sqlalchemy", "fastapi", "redis", "httpx"],  # no framework imports
    "src/db/":   [],                                            # can import core
    "src/api/":  [],                                            # can import core + db
}
SIBLING_BLOCKS = {
    "src/core/":   ["from src.db", "from src.api", "from src.worker"],
    "src/worker/": ["from src.api"],
}

Next.js:

LAYER_CONFIG = {
    "src/lib/":        ["react"],           # lib must not import components
    "src/components/": ["next/server"],     # components must not use server APIs
}

Rules are enforced on every Write/Edit tool call — violations are blocked before they land.

Generated CLAUDE.md

Uses actual project data (not generic templates):

• Real directory tree from your project
• Quick-start commands for your package manager
• Framework-specific testing strategies
• Layer rules tailored to your architecture

Multi-Agent Layer (--multi-agent)

Installs a complete agent coordination system:

.claude/agents/
  orchestrator.md          — routes tasks to the right specialist automatically
  implementers.yml         — role registry (source of truth)
  verifiers.yml            — verifier registry
  api-engineer.md          — compiled from registry
  database-engineer.md
  ui-designer.md
  testing-engineer.md
  devops-engineer.md
  security-engineer.md
  data-engineer.md
  docs-engineer.md
  architecture-verifier.md
  security-verifier.md
  quality-verifier.md
  test-verifier.md

.claude/commands/
  /orchestrate             — route any task to correct specialist, no manual selection
  /new-spec                — create a spec from a task description
  /create-spec             — turn a rough idea into a structured spec
  /implement-spec          — 5-phase: plan → handoff → implement → verify → close
  /context-handoff         — structured agent-to-agent context transfer

Hooks (Installed Automatically)

Auto-Preset Selection

Commands Reference

claudex init [DIR]

Initialize .claude/ for a project.

Options:

• --preset <name> — Override auto-detection
• --yes — Skip confirmation
• --force — Overwrite existing .claude/
• --global — Also install ~/.claude/ global config
• --multi-agent — Install orchestrator + 12 specialist agent files
• --dry-run — Preview without writing

What it creates:

.claude/
  hooks/        — 7 Python hooks (enforcement + lifecycle)
  commands/     — 17+ slash commands (/dev, /audit, /parallel, ...)
  rules/        — Development guidelines (workflow, naming, testing)
  session/      — Task persistence (CURRENT_TASK.md, TASK_PROGRESS.md, TOKEN_LOG.md, ...)
  agents/       — [--multi-agent] orchestrator + 12 agent files + registries
  feedback/     — Violation tracking
  knowledge/    — 100x patterns reference
CLAUDE.md       — Generated from your actual project structure
.mcp.json       — MCP server config template

claudex update [DIR]

Refresh templates without overwriting user files.

Preserved on update:

• session/CURRENT_TASK.md, session/TASK_PROGRESS.md, session/TOKEN_LOG.md
• session/BACKGROUND_QUEUE.md, session/PARALLEL_SESSIONS.md
• feedback/* — violations, lessons, corrections
• knowledge/* — user-added knowledge files
• docs/*

claudex validate [DIR]

Health check the .claude/ setup.

✓ PASS: .claude/ directory exists
✓ PASS: All required directories present
✓ PASS: settings.json exists and hooks registered
✓ PASS: CLAUDE.md exists
✗ FAIL: .gitignore does not include .claude/

claudex info [DIR]

Show detection results without making changes.

claudex presets

List all available presets.

100x Developer Workflow

The scaffold implements the 100x Developer Framework for maximum throughput.

Background Agents (Night Queue)

/background-queue add "Add unit tests for user service"
/background-queue add "Fix lint warnings in api/ directory"
/night-kick              # generates headless launch commands
# copy-paste, agents run overnight
/background-queue review # check results next morning

Parallel Sessions

/parallel plan "Sprint 5: Add real-time features"
# Creates PARALLEL_SESSIONS.md with file ownership and merge order

# Launch sessions in separate terminals
cd ../project-session-a && /dev start ...
cd ../project-session-b && /dev start ...

/parallel status         # check progress
/parallel merge-order    # get correct merge sequence

Multi-Agent Spec Implementation

# Route any task automatically
/orchestrate "add a POST /users endpoint with email validation"

# Full spec lifecycle
/new-spec "user authentication with JWT"
/implement-spec docs/specs/user-auth.md
# → plans agent assignments
# → generates context handoff docs
# → delegates to implementers in order
# → runs verifiers (architecture, security, quality, tests)
# → produces final-verification.md

Slash Commands

Development

# Install with dev dependencies
pip install -e ".[dev]"

# Run tests
pytest tests/ -v

# Lint + format
ruff check claudex/ tests/
ruff format claudex/ tests/

Requirements

• Python 3.11+
• pyyaml>=6.0 (installed automatically)
• Git (for parallel session worktree support)

Customization

Add New Presets

Create claudex/presets/your-preset.yaml:

name: your-stack
description: Your custom stack description
architecture_tree: |
  project/
    src/
    tests/
layer_description: |
  - **src/**: Application code
layer_rules:
  - Core layer: pure business logic
quick_start: |
  npm install && npm run dev

Modify Templates

Edit files in claudex/templates/project/:

• hooks/ — customize Python hooks
• commands/ — add slash commands
• rules/ — modify development guidelines
• agents/ — edit agent role definitions

Roadmap

• Publish to PyPI (pip install binxai-claudex)
• Multi-agent layer (--multi-agent flag)
• Agent file compilation (YAML registries → individual agent .md files)
• Stack-aware LAYER_CONFIG (architecture enforcement on init)
• Add more presets (Flask, Express, Vue, Svelte)
• Preset inheritance (extends: in YAML)
• preview command (show what CLAUDE.md would look like)
• Monorepo support (detect sub-projects)

Made for Claude Code — Anthropic's official CLI for Claude.