Project Template — AI-First, Multi-Assistant

This repository is a drop-in starter for any project that wants to use AI assistants (Claude, OpenAI / ChatGPT, Gemini, Local models) in a consistent, safe and reproducible way. It ships with:

Shared documentation (/docs) to give assistants and humans a single source of truth
Per-assistant configuration and prompts (/ai/assistants/*)
Reusable prompts, playbooks, tools and evals (/ai/*)
Example CI automations for headless assistant tasks (/.github/workflows)
Minimal dev tooling (/scripts, /docker) and language-agnostic defaults

You can safely delete language sections you don't need (e.g., pyproject.toml, package.json) after scaffolding.

Quick Start

Clone the template

git clone <your-fork-url> my-project
cd my-project

Set environment variables
Copy .env.example to .env and fill secrets:
```
cp .env.example .env
```
Pick your language/tooling
- Node/TypeScript: edit package.json, run npm i (or pnpm i), then npm run dev
- Python: edit pyproject.toml, create venv and run pip install -e .[dev]
Wire up assistants
- Update ai/CLAUDE.md (repo hints) and .claude/settings.json
- Adjust ai/assistants/*/model.config.json and system.md
- If using MCP, edit .mcp.json

Run the basic sanity checks

./ai/tools/format.sh
./ai/tools/typecheck.sh
./ai/tools/run-tests.sh

Enable CI
In GitHub, ensure Actions are enabled. The preconfigured workflows will run tests and (optionally) headless assistant triage.

Repository Layout

docs/ — Project and technical documentation intended for humans and AI
ai/ — Assistant configurations, prompts, tools, evals and policies
.claude/ — Claude Code settings and project slash-commands
.mcp.json — Model Context Protocol servers/clients shared across the repo
.github/workflows/ — CI pipelines with examples for AI-assisted tasks
docker/ — Optional local dev container and services
scripts/ — Human-oriented scripts (bootstrap, pre-commit hooks, repo map)

Using This Template with AI Assistants

Claude Code automatically loads ai/CLAUDE.md, docs/* and repo maps to build context.
Slash commands in .claude/commands and ai/assistants/claude/commands let you run repeatable workflows (e.g., /project:review-pr).
Headless automations (see ai/assistants/claude/headless/ and .github/workflows/) can triage issues or perform subjective linting on PRs.
LLM-agnostic prompts in ai/PROMPTS unify behavior across providers.

Conventions

Docs are canonical. If code contradicts docs/, open an issue.
Small PRs, with tests. Every change should include tests when possible.
Security-first. Never commit secrets; follow docs/06-security-and-privacy.md.
Conventional Commits and trunk-based branching are encouraged.

Cleanup Checklist (after cloning)