Project Loom — Multi-Repo AI Workspace Generator

Project Loom unifies multi-repository development by generating AI-ready workspace configurations. Define your repos, dependencies, and impact zones once—Loom generates everything: symlinks, orchestration scripts, and AI tool configs (Claude Code, Codex, Cursor, Aider).

Why Project Loom?

When you're working across 3+ repositories:

• Scattered configs — Claude Code, Codex, Cursor, Aider each need different markdown formats
• Manual orchestration — Running tests/installs across repos is shell-script spaghetti
• Lost context — Which files are risky to change? Which repos will break?
• Repeated setup — Every team member re-discovers the same patterns

Loom solves this:

# Define your workspace once
loom init

# Apply to your local machine
loom apply

# Everything works together
loom install          # Install all repos
loom test-all         # Run all tests
loom status           # Git status across workspace
loom sync-commit "bugfix: auth token expires correctly"  # Atomic multi-repo commit

Quick Start

Installation

pip install project-loom

Or install from source:

git clone https://github.com/ypollak2/project-loom
cd project-loom
pip install -e .

Usage

1. Create a configuration (interactive)

loom init

This launches an interactive wizard that asks about your repos, dependencies, and AI tools. Outputs loom.yaml.

2. Apply the configuration

loom apply loom.yaml

This:

• Clones repositories (if not already cloned)
• Creates symlinks in meta-workspace/services/
• Generates AI tool configs (.claudecode.md, .cursorrules, etc.)
• Generates scripts/loom orchestration CLI

3. Use the workspace

# Run tests in all repos
loom test-all

# Git status across all repos
loom status

# Health check all repos
loom doctor

# Synchronized commit across dirty repos
loom sync-commit "message"

Core Commands

Command	Purpose
loom init	Interactive wizard to create loom.yaml
loom apply [yaml]	Clone repos, create symlinks, generate AI configs
loom validate [yaml]	Validate YAML against schema
loom install	Run install commands in all repos
loom test-all	Run tests in all repos
loom doctor	Run health checks in all repos
loom status	Git status across all repos
loom diff	Git diffs across all repos
loom pull	Git pull --ff-only across all repos
loom logs [--count N]	Recent commits across all repos
loom sync-commit <msg>	Commit to all dirty repos with shared trailer

Configuration (loom.yaml)

Complete Schema

# Workspace metadata
name: my-workspace              # Workspace name (required)
description: "Your description" # Workspace description (required)

# Repositories
repos:
  - name: api                   # Unique repo name
    url: https://github.com/user/api
    role: backend               # Role in ecosystem
    language: python            # python | typescript | golang | rust
    local_path: ~/Projects/api  # Optional; defaults to ~/Projects/<name>
    install_command: pip install -e .
    test_command: pytest
    doctor_command: python -m app doctor

# Cross-repo dependencies
dependencies:
  - from: client
    to: api
    description: REST API calls

# Cross-repo impact zones
impact_zones:
  - id: IZ-001
    name: Auth Flow Changes
    risk: HIGH                  # HIGH | MEDIUM | LOW
    source:
      repo: api
      file: src/auth.py
      function: verify_token
    target:
      repo: client
    trigger: When auth middleware changes
    shared_env:                 # Shared environment variables
      - AUTH_SECRET
      - JWT_EXPIRY

# AI tool configuration flags
ai_tools:
  claude_code: true             # Generate .claudecode.md
  codex: false                  # Generate .codex-plugin/
  cursor: false                 # Generate .cursorrules
  aider: false                  # Generate AGENTS.md

Real-World Example

See examples/llm-router-chronicle/loom.yaml for a complete implementation with:

• 2 repos (Go + Python services)
• 1 dependency (REST API)
• 7 impact zones tracking auth, caching, and protocol changes

Generated Artifacts

After running loom apply, you get:

meta-workspace/
├── .claudecode.md         # Claude Code AI context
├── .cursorrules           # Cursor IDE rules
├── .codex-plugin/         # Codex configuration
│   └── loom-context.md
├── AGENTS.md              # Aider task agents
├── services/
│   ├── repo-a → ~/Projects/repo-a  (symlink)
│   └── repo-b → ~/Projects/repo-b  (symlink)
├── configs/
│   └── dependency-graph.json
└── scripts/
    └── loom               # Bash orchestration CLI

Commands

loom init [--output PATH]

Interactive wizard to create loom.yaml.

loom apply YAML [--workspace PATH]

Apply configuration: clone repos, create symlinks, generate AI configs.

loom status [--config PATH]

Git status of all repos.

loom test-all [--config PATH]

Run tests in all repos.

loom install [--config PATH]

Install all repos (run install_command for each).

loom doctor [--config PATH]

Health check all repos (run doctor_command for each).

loom diff [--config PATH]

Git diff summary all repos.

loom logs [--config PATH] [--count N]

Show last N commits per repo.

Design Principles

1. loom.yaml is the single source of truth — commit it to version control so your whole team can run loom apply and get identical workspaces.

2. Generators are pure functions — they take LoomConfig and return strings. Easy to test, easy to extend.

3. apply is idempotent — safe to re-run. Won't duplicate content or break existing symlinks.

4. No mutation of existing code — loom only creates net-new files under meta-workspace/. Your repos are untouched.

5. Impact zones are explicit — document cross-repo dependencies upfront so teams understand consequences before coding.

Extending Loom

Adding a New AI Tool

1. Create loom/generators/mytool.py with a generate_mytool(config: LoomConfig) -> str function
2. Add to loom/generators/__init__.py
3. Add to loom/workspace.py's _generate_ai_configs()
4. Add to AITools model in loom/config.py

Customizing Generators

Edit loom/generators/*.py to change output format for any AI tool. For example, to add custom rules to .claudecode.md, modify loom/generators/claude_code.py.

Development

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

# Run tests
pytest

# Format code
black loom

# Lint
ruff check loom

License

MIT

See Also

• Chronicle — Architectural decision memory layer
• LLM Router — LLM dispatch and budget management