# Lorx for agents

Give coding agents what they need to use Lorx: a skill, plain-text docs, and commands with JSON output.

Agents work on Lorx the way people do: git, bit, and the API. Three things make it easier for them.

## The Lorx skill

A skill teaches an agent (Claude Code, and any agent that reads `SKILL.md` skills) how to use Lorx well:
- sign in with a token;
- create and clone repositories;
- run CI with `bit run` before pushing;
- open, read and merge change requests;
- check how a push's CI went.

Install it for Claude Code:

```sh
mkdir -p ~/.claude/skills/lorx
curl -fsSL https://lorx.space/skills/lorx/SKILL.md -o ~/.claude/skills/lorx/SKILL.md
```

For one project only, put it in that repository's `.claude/skills/lorx/SKILL.md` instead. Read it first at [lorx.space/skills/lorx/SKILL.md](/skills/lorx/SKILL.md).

## Docs as plain text

- [`/llms.txt`](/llms.txt) lists every page of these docs, one line each, with a link to its Markdown.
- Every page is also served as Markdown: add `.md` to its address, such as [`/docs/ci/workflows.md`](/docs/ci/workflows.md).

## Feedback in the loop

`bit run` gives an agent the same CI result a push would get, before it pushes. Steps whose inputs haven't changed are skipped, so after a small edit a rerun costs little. Many commands also print JSON: `bit run --json`, `bit ci check --json`, `bit cr list --json` and `bit cr view --json`.

## Tokens for agents

Give an agent its own personal access token, limited to the repositories and permissions it needs, with a short expiry: see [Access tokens](/docs/account/tokens). A token can never do what needs your password, such as making more tokens or managing runners.

bit keeps credentials in the system keychain. In a container or sandbox with no keychain, give the agent the token as an environment variable and use it with git over HTTPS and the API directly: see [Connect over HTTPS or SSH](/docs/get-started/connect).