AI and agents
AI and agents
How an AI agent reads this documentation and works on the tui4s repository.
curl -s https://<this site>/llms.txt # the index of the pages
curl -s https://<this site>/raw/getting-started/first-app.md # one page as Markdown
Read the documentation
| What | Where |
|---|---|
| An index of the pages | /llms.txt |
| All pages in one Markdown file | /llms-full.txt |
| One page as Markdown | /raw/<page-path>.md |
An MCP server with the tools list-pages and get-page | /mcp |
Each page also has a Copy page button, which copies the page as Markdown.
Each Scala block of these pages is the text of a region of a source file of the repository. A test (MarkdownRegionsSuite) checks each block in CI, so a block cannot drift from the code that compiles.
Work on the repository
AGENTS.md at the root of the repository is the entry for agents. It imports two files. CONTRIBUTING.md is the rulebook, for humans and for agents: the commands, the code rules, the purity rules and the release process. ARCHITECTURE.md tells how the runner works. AGENTS.md adds the rules for agents only, for example the marks of a change. Some rules that an agent must know:
- Use JDK 25. Run one sbt process at a time in one checkout.
- Never start a terminal UI with
sbt run, and never in the shell of the agent: only the PTY kit runs one. - Tests do not need a real terminal: use
Headless. - A release publishes a public artifact. An agent can start a dry run at any time. It starts a real release only when a human asks for it, and the maintainer approves the deployment in GitHub (ADR 0020).