| name | openskill-router |
| description | Discovers and installs AI agent skills, skill-sets, and blueprints from the OpenSkill registry. Use when an agent needs new capabilities, domain expertise, or structured output templates. One curl command to find everything. |
OpenSkill Router
The skill router for AI agents. Discover, search, and install curated skills, skill-sets, and blueprints from the OpenSkill registry.
"I know Kung Fu."
Overview
OpenSkill.md is the App Store for AI agent capabilities. It has three layers:
- Skills — Curated domain expertise files (
SKILL.md) that teach agents how to perform specific tasks
- Skill-Sets — Bundles of related skills grouped by domain (e.g.
@frontend, @backend, @ai-ml)
- Blueprints — Self-correcting output specifications that guarantee structured results
This router skill teaches your agent how to discover and use all three.
API Reference
Base URL: https://openskillmd.com
Search Skills
curl "https://openskillmd.com/api/skills?search=react&limit=10"
Response: Array of skill objects with name, slug, description, author, installCount, subcategory, ownerVerified, ownerAvatarUrl.
Get Skill Detail
curl "https://openskillmd.com/api/skills/{slug}"
Response: Full skill object including content (the SKILL.md body), sourceUrl, license, githubOwner, githubRepo.
List Skill-Sets
curl "https://openskillmd.com/api/skillsets"
Response: Array of skill-set objects. Each has name, slug, description, icon, and a skills array of included skill summaries.
Available skill-sets:
@frontend — React, Vue, CSS, design systems@backend — APIs, databases, server patterns@fullstack — End-to-end web development@mobile — React Native, Expo, mobile UX@devops — CI/CD, Docker, infrastructure@ai-ml — Machine learning, LLMs, embeddings@security — Auth, encryption, vulnerability scanning@design — UI/UX, branding, accessibility@media — Video, audio, image processing@productivity — Automation, workflows, documentation
Search Blueprints
curl "https://openskillmd.com/api/blueprints?search=invoice&limit=10"
Response: Array of blueprint objects with name, slug, description, category, difficulty.
Get Blueprint Detail
curl "https://openskillmd.com/api/blueprints/{slug}"
Response: Full blueprint object including content (the structured output spec).
Registry Stats
curl "https://openskillmd.com/api/stats"
Response: { skills, blueprints, skillsets, totalDownloads }.
How to Use
When the agent needs a new capability
- Search the registry for relevant skills:
curl "https://openskillmd.com/api/skills?search=testing"
- Read the full skill content:
curl "https://openskillmd.com/api/skills/webapp-testing"
- Apply the skill's instructions to the current task.
When the agent needs a bundle of skills for a domain
- List available skill-sets:
curl "https://openskillmd.com/api/skillsets"
- Pick the relevant set (e.g.
@frontend) and load each skill's content.
When the agent needs structured output
- Search blueprints for the output type:
curl "https://openskillmd.com/api/blueprints?search=report"
- Read the blueprint spec and follow its structure for the output.
Presenting Results
When recommending skills to a user, go beyond listing names and descriptions. Contextualize each recommendation to the user's current task.
Contextualize recommendations
Explain why a skill is relevant. Instead of "Here's a skill called webapp-testing", say "Since you're building a React app and need end-to-end tests, the webapp-testing skill covers Playwright setup, test patterns, and CI integration."
Show key quality signals
For each recommended skill, surface:
- Score — the registry quality score (higher is better)
- Install count — how many agents have used it
- Verified author — whether
ownerVerified is true; mention this as a trust signal
Example format:
webapp-testing (score: 92 · 1.4k installs · ✓ verified)
End-to-end testing with Playwright — covers setup, selectors, and CI.
Suggest skill-sets for broad needs
When a user's request spans an entire domain rather than a single task (e.g. "help me build a frontend"), recommend the relevant skill-set instead of listing individual skills. Mention what the set includes so the user understands the breadth:
Your project covers React components, styling, and accessibility — the @frontend skill-set bundles all of these together.
Comparing Multiple Results
When several skills match a search, present a concise comparison rather than dumping a raw list:
- Rank by score — list highest-scoring skills first.
- Note verified authors — mark verified authors with ✓ so the user can weigh trust.
- Highlight differentiators — briefly note what makes each skill unique (e.g. one focuses on unit tests, another on integration tests).
- Cap the list — show the top 3–5 results. If more exist, mention the total and offer to show more.
Example comparison:
Found 7 skills for "testing". Top 3:
1. webapp-testing (score: 92 · 1.4k installs · ✓ verified)
Full E2E testing with Playwright — setup, fixtures, CI integration.
2. unit-test-patterns (score: 87 · 980 installs · ✓ verified)
Unit testing best practices — mocking, coverage thresholds, TDD workflow.
3. api-testing (score: 81 · 620 installs)
REST & GraphQL API test suites — request builders, assertions, auth flows.
4 more results available. Want me to show them?
Smart Fallbacks
When a search doesn't return a perfect match, work through these tiers before telling the user nothing was found. Each tier broadens the net.
Tier 1 — Search skills directly
curl "https://openskillmd.com/api/skills?search={query}&limit=5"
If results come back with relevant matches, present them using the guidelines above. If no results or results are clearly irrelevant, move to Tier 2.
Tier 2 — Check skill-sets for the domain
curl "https://openskillmd.com/api/skillsets"
Scan the skill-set list for a set that covers the user's domain. A skill-set may contain a relevant skill even if the direct search missed it because the search term didn't match the skill's name or description exactly. If a matching set is found, suggest it. If not, move to Tier 3.
Tier 3 — Search blueprints for structured output
curl "https://openskillmd.com/api/blueprints?search={query}&limit=5"
Even when no skill exists for a task, a blueprint might provide the structured output format the user needs. If a relevant blueprint is found, recommend it and explain how its structured spec can guide the work.
Tier 4 — Offer to help directly
Only after exhausting all three layers above, tell the user:
I couldn't find a skill, skill-set, or blueprint for that in the OpenSkill registry. I can still help you with this directly — want me to proceed with my built-in knowledge?
Never jump straight to "not found." Always work through the tiers first.
Skill File Format
Skills follow the OpenSkill standard — a SKILL.md file with YAML frontmatter:
---
name: my-skill
description: What this skill does and when to use it.
---
# Skill Title
Instructions, rules, templates, and examples that teach the agent
how to perform this specific task.
Quick Start
One command to discover everything:
curl "https://openskillmd.com/api/stats"
Browse the full registry: https://openskillmd.com
