mackit 0.1.1

Tiny agent-ready project initializer and adapter.

mackit

Make a repository ready for multi-agent work with one clone and one command.

This kit gives agents a shared workspace (.agents/), a small set of coordination rules, and a curated skill catalog inside .agents/skills/. Users do not need to design their own handoff protocol or hunt for skills before starting: ask an agent to install this repo, keep the default skill bundles, and start working.

It stays local-first and lightweight. There is no daemon, database, hosted service, lock server, or required dependency.

It is not trying to replace public skill marketplaces. It complements them by making a single project immediately usable by agents: install the repo-local coordination layer, seed a small curated skill set, and add external marketplace skills only after review.

Quick Start

pip install mackit
mackit adapt .

When run in a normal terminal, mackit adapt . asks whether to install all curated skills, choose specific bundles, or skip curated skills. Press Enter to accept the recommended default: install all.

For local development from this repository, use PYTHONPATH=src python -m mackit.

Why Pip Install

mackit supports two distribution paths:

• pip install mackit for normal users who want the fastest path.
• A GitHub checkout for agents and maintainers who need to inspect, adapt, or improve the kit.

Pip matters because the main promise of this project is low-friction adoption. Users should not need to clone this repository, set PYTHONPATH, understand the source layout, or copy files by hand before their project becomes agent-ready.

The installed package includes the CLI, templates, Cursor rule, .agents/ coordination files, and curated skills. After installation, this should work from any repository:

mackit adapt .

GitHub remains the source of truth for review, contribution, and agent-readable implementation details. PyPI is the convenient delivery channel.

Two Use Cases

1. Start a new project

mackit new ./my-project --package-name my_project

This creates a simple Python src/ layout plus the agent-native coordination files.

2. Adapt an existing project

mackit adapt /path/to/existing-project

The adapter is additive and idempotent:

• Existing files are not overwritten.
• Missing files are created.
• AGENTS.md, CLAUDE.md, and .gitignore receive managed sections only once.
• Cursor receives .cursor/rules/multi-agent.mdc.
• Curated skills from this kit's .agents/skills/ catalog are offered during setup.

What Gets Installed

.agents/
├── README.md
├── STATE.md
├── agents/.gitkeep
├── retros/.gitkeep
├── skills/
│   ├── .gitkeep
│   ├── README.md
│   ├── registry.toml
│   └── common/
│       └── improve-this-kit/
│           └── SKILL.md
├── tasks/.gitkeep
├── improve-this-kit/
│   ├── .gitkeep
│   ├── README.md
│   ├── feedback/
│   │   ├── .gitkeep
│   │   └── README.md
│   └── proposals/
│       ├── .gitkeep
│       └── README.md
└── templates/
    ├── agent_state.md
    ├── feedback.md
    ├── proposal.md
    ├── retro.md
    ├── skill.md
    └── task_state.md

.cursor/rules/multi-agent.mdc
AGENTS.md
CLAUDE.md
.gitignore
src/<package_name>/        # only for `new`, or when --package-name is passed

Selected curated skills are copied from this kit's .agents/skills/ into the target .agents/skills/, preserving grouped paths such as design/image-to-editable-slides/.

Design Principles

• AGENTS.md is the cross-tool canonical contract.
• CLAUDE.md is a thin Claude Code shim.
• .cursor/rules/ is a Cursor-native adapter.
• .agents/ stores runtime coordination, not long-term design knowledge.
• .agents/retros/ stores raw post-task learning drafts.
• .agents/skills/ stores reviewed, reusable skills for future agents.
• .agents/improve-this-kit/feedback/ stores raw usage feedback about the kit.
• .agents/improve-this-kit/proposals/ stores review-gated kit improvement proposals and is trackable by default.
• improve-this-kit turns real usage feedback into reviewed proposals and tests.
• .agents/skills/registry.toml describes curated skills and bundles.
• .agents/skills/external/README.md tells agents how to recommend external skills safely.
• Design decisions should move to your real docs, ADRs, or README.
• Work isolation should still prefer branches or worktrees for heavy parallelism.

Self-Improvement Loop

The kit is designed to improve from real usage without letting agents rewrite it from a single anecdote:

1. Agents write raw usage notes in .agents/improve-this-kit/feedback/.
2. Repeatable issues become proposals in .agents/improve-this-kit/proposals/.
3. Proposals require explicit review/approval before implementation.
4. Accepted changes must include or update an idempotency check, fixture, or test.
5. Reviewed, reusable know-how can become a skill in .agents/skills/.

Use .agents/skills/common/improve-this-kit/SKILL.md for the review-gated workflow.

Curated Skill Catalog

The repo ships its maintained catalog directly in .agents/skills/. During manual terminal use, mackit asks whether to install all curated skills, choose specific bundles, or skip them:

mackit adapt /path/to/project

For automation and agents, use explicit flags so the command never waits for input:

mackit adapt /path/to/project --yes
mackit adapt /path/to/project --skill-bundles common,design
mackit adapt /path/to/project --no-curated-skills

Use list-skill-bundles to inspect available bundles.

External skill discovery is optional. Agents may recommend external skills, but must report candidates and wait for user approval before copying them. See .agents/skills/external/README.md.

For Agents

If a user gives you this repository link and asks you to make their project agent-ready, read INSTALL_FOR_AGENTS.md first.

For public sharing, see docs/external_download.md.