Documentation

What is Loomcraft?

Loomcraft is an open-source preset manager for AI coding agents. It lets you scaffold agents, skills, and complete workflow pipelines into any project with a single command.

Think of it as a package manager for AI workflows: you install a preset, and your project gets a team of specialized agents, a curated set of skills.sh skills, and a defined pipeline that orchestrates them.

Works with Claude Code, Cursor, and any tool that reads CLAUDE.md or .cursorrules.

Getting started

Install a preset

The fastest way to get started. Pick a preset and scaffold it into your project:

npx @loomcraft/cli add loomcrafthq/presets/saas

This creates the following structure in your project:

.claude/
  agents/
    database/AGENT.md
    backend/AGENT.md
    frontend/AGENT.md
    ux-ui/AGENT.md
    tester/AGENT.md
    review-qa/AGENT.md
    security/AGENT.md
  skills/
    brainstorming.md
    writing-plans.md
    ...
CLAUDE.md
loomcraft.config.json

Add a single agent

You can also add individual agents without a full preset:

npx @loomcraft/cli add loomcrafthq/agents/frontend

The agent is fetched from GitHub, written to .claude/agents/, and automatically referenced in your CLAUDE.md.

Discover agents and presets

Search the registry from your terminal:

npx @loomcraft/cli find database

Or browse the leaderboard on this site.

Search presets and agents

npx @loomcraft/cli find

Agents

An agent is a markdown file (AGENT.md) that gives an AI coding assistant a specialized role with domain-specific knowledge, guidelines, and best practices.

Built-in agents

Loomcraft ships 7 universal, stack-agnostic agents:

Agent file format

An AGENT.md is a markdown file with YAML frontmatter:

---
name: frontend
description: Builds UI components, manages client-side state, enforces accessibility.
tools: Read, Edit, Write, Bash, Grep, Glob
---

# Frontend Agent

## Bootstrap
Read `CLAUDE.md` to learn about the project stack and conventions.

## Component Architecture
- Use composition over inheritance
- Prefer server components by default
...

The name and description fields are used for display. The rest of the markdown is the agent's knowledge — instructions, guidelines, checklists that the AI will follow.

Presets

A preset is a complete workflow configuration: a team of agents, a curated set of skills, and a pipeline that defines the order of operations.

Available presets

Preset file format

A preset.yaml defines the full workflow:

name: SaaS
description: Full-stack SaaS with auth, billing, and TDD pipeline.

agents:
  - loomcrafthq/agents/database
  - loomcrafthq/agents/backend
  - loomcrafthq/agents/frontend

skills:
  - obra/superpowers/brainstorming
  - github/awesome-copilot/conventional-commit
  - anthropics/skills/frontend-design

pipeline:
  - agent: database
    when: "if migrations needed"
  - agent: tester
    mode: tdd
    scope: "database + backend"
  - agent: backend
  - agent: frontend

verification:
  - review-qa
  - security

conventions:
  commits: conventional
  branches: "feat/<ticket-id>-<description>"

Pipeline

The pipeline section defines the execution order. Each step can have:

• agent — which agent handles this step
• when — optional condition (e.g. "if migrations needed")
• mode — tdd or test-after
• scope — what part of the codebase this step covers

Verification

The verification section lists agents that always run after the pipeline completes — quality gates like review, testing, and security audit.

Create your own

Create an agent

1. Create a GitHub repo (e.g. my-org/agents)
2. Add a folder with an AGENT.md file:

   my-org/agents/
     my-agent/
       AGENT.md

3. Add frontmatter with name, description, and tools
4. Write your agent's instructions in markdown
5. Anyone can install it:

   npx @loomcraft/cli add my-org/agents/my-agent

Create a preset

1. Create a GitHub repo (e.g. my-org/presets)
2. Add a folder with a preset.yaml file:

   my-org/presets/
     my-preset/
       preset.yaml

3. Define agents, skills, pipeline, and verification
4. Agents can reference any public GitHub repo. Skills reference skills.sh packages.
5. Anyone can install it:

   npx @loomcraft/cli add my-org/presets/my-preset

How ranking works

Rankings on the leaderboard are based on anonymous telemetry. Every time someone runs loomcraft add or loomcraft find, the CLI pings our API with the ref and type. That's it.

We collect zero personal data — no IP addresses, no usage patterns, no identifiers. Just a counter: "this ref was installed N times."

The source code for both the CLI and the telemetry endpoint is open-source. You can verify exactly what is sent.

Ref format

All agents and presets are identified by a ref — a GitHub path in the format org/repo/name:

loomcrafthq/agents/frontend    → github.com/loomcrafthq/agents/tree/main/frontend
loomcrafthq/presets/saas       → github.com/loomcrafthq/presets/tree/main/saas
my-org/my-agents/custom-agent  → github.com/my-org/my-agents/tree/main/custom-agent

The CLI fetches the AGENT.md or preset.yaml directly from the main branch via GitHub raw content.

Refs starting with loomcrafthq/ are marked as Verified. All others are Community contributed.