1,500 APIs, Agent-Ready in One Command

LAP Team
registryskills

Compressing API specs is step one. Getting them into your agent’s hands is step two. LAP handles both.

The Flow

pip install lapsh

# Install a pre-built skill from the registry
lapsh skill-install stripe

# Done. Your agent now has Stripe access at 10× fewer tokens.

That’s it. One command. The skill lands in ~/.claude/skills/ ready for your agent to use.

What Gets Installed

A skill isn’t just a compressed spec with a wrapper. It’s an agent-native interface:

# Stripe API

Payment processing platform. 587 endpoints.

## Auth
Set `STRIPE_API_KEY` environment variable.
Auth type: Bearer token in Authorization header.

## Base URL
https://api.stripe.com/v1

## Quick Reference
- Create payment intent: POST /payment_intents
- List customers: GET /customers
- Create subscription: POST /subscriptions

Plus the full compressed spec with every endpoint, parameter, and type constraint. The SKILL.md frontloads what the agent needs to start working — auth, base URL, common operations. The spec provides the complete catalog when the agent needs to go deeper.

The Registry

registry.lap.sh hosts 1,500+ pre-compiled API specs — Stripe, GitHub, Twilio, Slack, Shopify, OpenAI, and hundreds more. Every spec is available in multiple compression tiers.

Browse by name, search by category, or use the CLI:

lapsh skill-install github
lapsh skill-install twilio
lapsh skill-install shopify

Generate Skills from Your Own Specs

The registry covers 1,500+ public APIs. For internal or custom APIs:

# Compile your spec
lapsh compile internal-api.yaml -t lean

# Generate a skill from it
lapsh skill internal-api.yaml

# Publish to the registry (optional)
lapsh publish internal-api.yaml

Namespace verification uses DNS (add a TXT record to prove domain ownership) to prevent spec squatting. Private specs can stay local — you don’t have to publish to use the tools.

Claude Code Marketplace

1,501 skills are also available in the Claude Code marketplace via the lap-skills plugin. Same skills, different distribution channel.

Why Skills > Raw Specs

You could just paste a compressed spec into your agent’s context. Skills are better for three reasons:

1. Auth is pre-configured. The skill tells the agent exactly which environment variable to look for and how to pass it (Bearer header, query param, basic auth). No guessing.

2. Context is minimal. The SKILL.md is typically 1-2K tokens — enough to understand the API and start working. The full spec is only loaded when needed.

3. They’re portable. Same skill format works across Claude Code, OpenClaw, and any framework that reads markdown instructions. Write once, use everywhere.

Inspect and Diff

Two more CLI tools that help manage specs over time:

# Inspect a compiled spec
lapsh inspect stripe.lap

# Diff two versions
lapsh diff stripe-v1.lap stripe-v2.lap

The diff command outputs semantic changes — added/removed endpoints, parameter type changes, new required fields — with semver guidance. Because LAP has a formal grammar, diffing is deterministic. Try doing that with two OpenAPI YAMLs.

What’s Next

The skill layer is where LAP stops being a compression tool and starts being agent infrastructure. We’re working on:

  • Skill versioning — track API changes across spec updates
  • Usage analytics — which skills are agents actually using
  • Community contributions — submit skills for APIs not yet in the registry

Get started:

pip install lapsh
lapsh skill-install stripe

From zero to agent-ready in under a minute. Browse the full catalog at registry.lap.sh.