RHL AI / open source / MIT
Agents follow instructions.Write them once.
RHL AI packages engineering judgment as SKILL.md files — and ships a CLI that installs them into ~/.claude/skills, where Claude Code picks them up as slash commands like /code-reviewer.
npx @rhl-ai/cli skill install code-reviewer --target ~/.claude/skills: Installing skill 'code-reviewer'... Skill 'code-reviewer' successfully installed to '~/.claude/skills/code-reviewer'.
ls ~/.claude/skills/code-reviewer: SKILL.md examples/ references/ scripts/
- Agent skillscode-reviewer, frontend-performance-engineer, pr-description-writer
- 3
- CLI commandsrhl skill …, rhl create
- 6
- Starter templatesnextjs, node, react, whatsapp-bot
- 4
- Workspace packages@rhl-ai/cli, @rhl-ai/scaffold, @rhl-ai/skills, @rhl-ai/utils
- 4
01 — Anatomy
A skill is a folder, not a framework.
No runtime, no SDK, nothing in your bundle. Four plain parts that any agent able to read a file can follow.
02 — The loop
Four steps from catalog to code review.
- Discover: See what is on the shelf. The CLI scans the catalog and prints every skill with its version, author and bundled files.
- Validate: Check the contract first. Frontmatter, instructions and folder structure are verified before anything touches your machine.
- Install: Copy it where agents look. The folder lands in ~/.claude/skills (or any agent’s skills folder) as plain files. No daemon, no lockfile, nothing added to your app.
- Apply: The agent reviews the way you would. Type /code-reviewer in Claude Code (or just ask for a review). Every finding arrives in the same shape: file, severity, problem, suggested fix.
03 — What ships
Three tools. One way of working.
The rhl command discovers, validates and installs skills, and scaffolds new projects. Strict TypeScript, published on npm.
- ❯rhl skill listList every skill in the catalog
- ❯rhl skill create <name>Scaffold a new skill folder
- ❯rhl skill validate <name>Check structure and SKILL.md metadata
- ❯rhl skill install <name> -t ~/.claude/skillsInstall into Claude Code; call it with /<name>
- ❯rhl skill update [name]Refresh installed skills
- ❯rhl create <template> <name>Start a project from a template
Behavior packs written in Markdown. Also published to npm as @rhl-ai/skill-*.
- code-reviewerv1.0.0Expert AI code reviewer for strict TypeScript, architecture, and performance.
- frontend-performance-engineerEvidence-led frontend performance engineering skill.
- pr-description-writerv1.0.0Write rigorous, reviewer-friendly pull request descriptions and actually do the verification and visual evidence they claim.
rhl create copies a starter and gets out of the way.
- nextjsModern full-stack web application built with Next.js App Router, React, Tailwind CSS, and TypeScript.
- nodeMinimal, modern Node.js application configured with TypeScript and ESM.
- reactLightweight single-page application built with React, Vite, and TypeScript.
- whatsapp-botReceive WhatsApp alerts for GitHub pull request and CI events.
04 — Ground rules
Built the way it tells agents to build.
- 01
Strict TypeScript
no any, anywhere
Every package and script is compiled under strict settings, and the linter rejects unsafe assignments, calls and returns.
- 02
Composable tooling
independent packages
Use one tool on its own or all of them through the CLI. Each module has clear boundaries and is versioned independently.
- 03
Agent-native
written for LLMs to follow
Structured instructions, YAML frontmatter and explicit decision frameworks that coding agents apply the same way every time.
- 04
Open source
MIT licensed
Built in the open. Read the source, contribute your own skills, and self-host the automations.
05 — Start