Docs

Quick start with Claude

The fastest way to get Last Light running is to let Claude Code do it for you. Last Light ships a set of Claude Code skills that know how to install, configure, and operate the whole stack — you describe what you want in plain language and the skill drives the CLI, scaffolds the working directory, and verifies the result.

Prefer to do it by hand? Every step the skills run is documented manually too — start with Prerequisites, Create a GitHub App, and Run it locally.

1. Install the CLI and skills

Install the lastlight CLI, then install the bundled, version-matched skills:

npm i -g lastlight          # or run it via npx
lastlight skills install    # → ~/.claude/skills (user scope)

lastlight skills install uses the claude plugin marketplace when one is present, otherwise it copies the skills into ~/.claude/skills. Add --scope project to install them into ./.claude/skills for a single repo. From a Last Light checkout you can also register the marketplace directly:

claude plugin marketplace add ./
claude plugin install lastlight@lastlight-skills

2. Ask Claude to set it up

Open Claude Code in an empty directory and describe what you want. Each skill is triggered by intent — you don't need to remember its name:

Skill	Ask Claude…	What it does
`lastlight-server`	"set up a Last Light server"	Checks prerequisites, scaffolds the working directory, walks through the GitHub App, model and managed repos, launches the docker-compose stack, and verifies `/health`.
`lastlight-client`	"connect my lastlight CLI to a server"	Installs the CLI, logs in to an existing server, saves the token, and confirms the connection with `lastlight status`.
`lastlight-overlay`	"customize my Last Light workflows"	Scaffolds the private `instance/` overlay and forks built-in workflows, prompts, skills, or the agent persona so you can tune them.
`lastlight-evals`	"scaffold an evals workspace"	Sets up and runs a Last Light Evals workspace — the harness that runs real workflows against a mocked GitHub and grades them deterministically.

The skills run against any model the pi coding agent supports — Claude, GPT, Gemini, Llama, and more (default anthropic/claude-sonnet-4-6).

3. Verify

Once the server skill finishes, confirm the agent is up:

curl http://127.0.0.1:8644/health

Then open the dashboard at /admin, or trigger work from the CLI — see Run it locally for the full command set and the CLI reference for the skills commands.

Next → Introduction