# Last Mile Labs starter kit

Rules, skills and review agents that help AI coding tools build like a product team. For vibe coders who are serious about quality and getting to market. Works with Claude Code, OpenAI Codex, GitHub Copilot in VS Code, Cursor, and builders such as Lovable, Floot, Riff, Bolt and v0.

From Last Mile Labs (lastmilelabs.io), the playbook for getting AI-built products across the last mile. Questions or help: contact@lastmilelabs.io

## What is inside

| Folder | What it is | Who reads it |
|---|---|---|
| `AGENTS.md` | Project rules for people and AI agents: never, always, when unsure, commands | Codex, Copilot, Cursor, Lovable, most agents |
| `CLAUDE.md` | Points Claude Code at AGENTS.md | Claude Code, Lovable |
| `skills/` | Ten skills in the open Agent Skills format (one SKILL.md each) | Claude Code, Codex, Copilot, Cursor, Lovable |
| `agents/` | Four review agents: security reviewer, release gatekeeper, test writer, readiness coach | Claude Code, Copilot custom agents |
| `builders/` | Setup and prompts for Lovable, Floot, Riff and any other builder | You, pasted into the builder |
| `ci/` | GitHub Actions quality gate, pull request checklist, `.env.example` | GitHub |

### The skills

| Skill | Use it to |
|---|---|
| `stage-assessment` | Find where the project stands and what to do now, next and never |
| `security-review` | Catch secrets, open tables, UI-only checks, bad input, risky packages |
| `access-rules` | Write and test row-level security with two accounts |
| `secrets-hygiene` | Find leaked keys, move them out of the code, rotate them |
| `quality-gate` | Set up CI that blocks broken changes |
| `critical-path-tests` | Test sign-up, the paid action and every data write |
| `release-check` | Confirm a deploy is safe and can be undone |
| `backup-drill` | Prove you can restore yesterday's data |
| `last-mile-audit` | Find the 20 percent a demo skips |
| `handover-runbook` | Write the runbook a new developer needs |

## Install in a code editor or coding agent

Requires Node 18 or newer. From the kit folder:

```
node install.mjs --tool claude --target ../my-app
node install.mjs --tool codex --target ../my-app
node install.mjs --tool copilot --target ../my-app
node install.mjs --tool cursor --target ../my-app
node install.mjs --tool all --target ../my-app --with-ci
```

Existing files are never overwritten unless you add `--force`. Use `--dry-run` to see what would be written.

Where things go:

| Tool | Skills | Agents |
|---|---|---|
| Claude Code | `.claude/skills/` | `.claude/agents/` |
| OpenAI Codex | `.agents/skills/` | uses AGENTS.md |
| GitHub Copilot (VS Code, Visual Studio) | `.github/skills/` | `.github/agents/*.agent.md` |
| Cursor | `.cursor/skills/` | uses AGENTS.md |

To install for yourself across all projects instead, copy the skills folders into your user folder for the tool, for example `~/.claude/skills/` or `~/.agents/skills/`.

Then open AGENTS.md, fill in the Stack and Commands sections, and ask your agent: "Run the stage-assessment skill and tell me what to fix first."

## Use with a builder

- **Lovable**: see `builders/lovable.md`. AGENTS.md in the repository root, a block for Project Knowledge, and skills imported from GitHub or a zip.
- **Floot**: see `builders/floot.md`. Prompts to reuse, since Floot has no standing rules.
- **Riff**: see `builders/riff.md`. Custom instructions and the knowledge library.
- **Bolt, v0, Replit, Base44 and others**: see `builders/any-builder.md`.

## Check the kit itself

```
node --test tests/*.test.mjs
```

## Licence

MIT. Use it, change it, share it. Prices and tool behaviour change; check the vendor documentation before relying on a detail.