Using ryl with AI agents¶

ryl is built for programmatic use (stable exit codes, machine-readable output, --fix/--diff, stdin), so AI coding agents can drive it directly. This page points to the agent-facing resources; the rest of these docs (start with Quick start) cover the CLI itself.

The ryl Agent Skill¶

Install the cross-tool Agent Skill so an agent learns the correct ryl invocation and config patterns:

gh skill install owenlamont/ryl ryl --agent claude-code --scope user

Swap --agent for cursor, codex, antigravity, github-copilot, and others (run gh skill install --help for the full list); or run gh skill search ryl to discover it. The skill is the canonical, tightened agent reference; this page only links to it.

Prompting in AGENTS.md¶

If a repo has an AGENTS.md (or CLAUDE.md), a one-line nudge helps agents reach for ryl and avoid its main gotcha:

Lint YAML with ryl. It has no default-on rules, so make sure a config (ryl.toml or a yamllint-style .yamllint) enables rules first, or it exits 2.

Feeding the docs to an LLM¶

The docs site publishes llmstxt.org files for agents that fetch documentation live (Cursor, Copilot, Claude Code, Windsurf):

https://ryl-docs.pages.dev/llms.txt        # index: links + one-line summaries
https://ryl-docs.pages.dev/llms-full.txt   # the full docs concatenated in one file

No MCP server¶

ryl is a CLI an agent can shell out to, so it intentionally ships no MCP server (as with ruff, Biome, Prettier, and yamllint). The skill plus llms.txt cover the agent use case.