find-skill: Discover & Install 50+ Agent Skills with Orthogonal 2026
Let AI agents autonomously search and install skills from Orthogonal's library. Works with Claude Code, Cursor, OpenClaw. One command. No key management.
find-skill lets agents discover and install capabilities from Orthogonal's 50+ skill library in one command. Works with Claude Code, Cursor, OpenClaw, Codex. Pay-per-use pricing, no subscriptions. Built for teams scaling agentic systems autonomously.
find-skill is a capability within Orthogonal's unified API platform that lets AI agents autonomously search and install skills from a library of 50+ pre-built integrations. It works with Claude Code, Cursor, OpenClaw, and Codex via CLI commands like `orth skills search` and `orth skills add`. Pricing is pay-per-use; agents only pay for the APIs their installed skills call, with no subscriptions. Built by Y Combinator W26 startup Orthogonal.
Maker: Orthogonal · Protocol: CUSTOM · Auth: api key
Compatible agents: Claude (Claude Code, Desktop), Cursor, OpenClaw, Codex, Gemini CLI, Windsurf
Required runtime: Node >= 18, npm or Deno, Orthogonal CLI or MCP client
About find-skill
find-skill is a capability within the Orthogonal platform that lets agents autonomously search, discover, and install skills from a growing library of 50+ pre-built skills designed for AI agents. Built by Orthogonal (Y Combinator W26), this skill enables Claude Code, Cursor, OpenClaw, Codex, and any markdown-aware agent to explore the Orthogonal skills ecosystem and add new capabilities on demand, without manual intervention or documentation hunting. When an agent needs to perform a task outside its current capabilities, find-skill acts as a marketplace navigator, letting the agent search by keyword (e.g. 'web scraping', 'email validation', 'person enrichment') and install the matching skill with a single command like `orth skills add orthogonal/scrape`. Behind the scenes, find-skill exposes the Orthogonal Skills API and registry, providing structured metadata about each skill including runtime requirements, supported agents, pricing, and setup instructions. The best use cases for find-skill are agents that need to work autonomously across multiple domains—an agentic workflow might start without web scraping, then discover and install the ScrapeGraph skill when the user asks for competitor research. It works especially well for teams building agentic systems that must adapt capabilities at runtime, avoiding hardcoded tool lists and enabling agents to learn what's available in the ecosystem. find-skill is open source and available via npm (`npm install @orthogonal-sh/skills`), as a CLI (`orth skills search`), and as a native MCP server for Claude, Cursor, and Windsurf. Pricing follows Orthogonal's pay-per-use model—agents only pay for the APIs they actually call through the skills they install, with no subscriptions or minimums. Recent updates (June 2026) added support for GitHub and Aviato integrations, bringing the library to 50+ skills with expanded compatibility for data enrichment, web navigation, and automation workflows.
Key Features
- Keyword-based skill search: Agents search by intent (e.g. 'web scraping', 'email verify') and get ranked results from 50+ skills with one-line metadata.
- One-command installation: Install any skill with `orth skills add orthogonal/skill-name`; no manual config, auth, or key management required.
- Autonomous discovery: Agents can call find-skill at runtime to discover new capabilities on demand without hardcoded tool lists.
- Multi-agent compatibility: Works with Claude Code, Cursor, OpenClaw, Codex, and any agent that reads markdown or supports MCP.
- Unified authentication: Single Orthogonal API key handles all installed skills; no per-tool API key juggling.
- Runtime metadata: Each skill includes setup time, required runtime, supported agents, and pricing so agents can make informed choices.
How to Invoke
CLI command (orth skills search / orth skills add) or native MCP tools exposed to the agent
Pricing
Free to discover and install skills. Charges apply only when installed skills call third-party APIs—Orthogonal's pay-per-use model. No monthly fees or subscriptions.
Strengths
- Dramatically reduces time to add new capabilities—agents discover and install skills in seconds instead of hunting docs and managing keys.
- Works out-of-the-box with 6+ major agents (Claude, Cursor, OpenClaw, Codex) without agent-specific skill packaging.
- Unified authentication removes the friction of managing 20+ API keys across different providers.
- Pay-per-use pricing (no subscriptions) means teams only pay for skills their agents actually use.
Weaknesses
- Library is still growing (50+ skills as of June 2026); less comprehensive than mature tooling ecosystems like npm.
- Skill quality and maintenance varies since creators submit to the registry; some skills may be unmaintained or have sparse documentation.
- Dependency on Orthogonal's infrastructure; if the platform goes down, agents cannot discover new skills (though installed skills still work if the underlying APIs are live).
- Lock-in risk if many of your custom skills are built on Orthogonal's proprietary skill.md format.
Frequently Asked Questions
What is find-skill and how does it work?
find-skill is a capability from Orthogonal (Y Combinator W26) that lets AI agents autonomously search and install skills from a growing library of 50+ pre-built integrations. An agent calls find-skill with a natural language query like 'I need to scrape websites', and find-skill returns a ranked list of matching skills with metadata about setup time, pricing, and agent compatibility. The agent can then install a skill with one command (`orth skills add orthogonal/skill-name`) without manual configuration or API key management. The underlying protocol is custom; Orthogonal exposes this via their CLI, npm package, and native MCP servers for Claude Code, Cursor, and Windsurf.
How do I install and set up find-skill?
Installation depends on your agent. For Claude Code or Cursor, add the Orthogonal MCP server to your config—no separate installation needed; find-skill is built in. For CLI use, install the Orthogonal CLI via `npm install -g @orthogonal-sh/cli` or `npm install @orthogonal-sh/skills` in your project. Generate or obtain an Orthogonal API key from https://www.orthogonal.com/dashboard. Set it in your environment as `ORTHOGONAL_API_KEY`. Verify setup by running `orth skills search web` to list web-related skills. No additional runtimes are required beyond Node >= 18.
Which agents and LLMs support find-skill?
find-skill works with Claude Code, Cursor, Windsurf, OpenClaw, Codex, Gemini CLI, and any agent that reads markdown or supports MCP (Model Context Protocol). Orthogonal's skill.md format is compatible with 6+ major agents, so skills discovered via find-skill work across these platforms without agent-specific repackaging. The MCP native integration is the fastest path for Claude Code and Cursor users, while CLI commands work in any terminal-based workflow. Agents running inside agentic frameworks (e.g. LangChain, CrewAI) can invoke find-skill via REST API or through framework-specific Orthogonal integrations.
How much does find-skill cost?
find-skill itself is free—discovering and installing skills costs nothing. Pricing applies only when installed skills call third-party APIs. Orthogonal uses a pay-per-use model: if your agent installs the ScrapeGraph skill and calls it 100 times, you pay only for those 100 calls at ScrapeGraph's per-request rate (handled through Orthogonal's unified billing). There are no monthly subscriptions, no seat-based fees, and no minimum commitments. Teams can also access a free tier; check https://www.orthogonal.com/ for current free-tier limits per API partner. This makes find-skill ideal for scaling agent systems cost-effectively.
Is find-skill open source?
Yes. The Orthogonal skills library, CLI, and find-skill capability are open source and hosted on GitHub at https://github.com/orthogonal-sh/skills. The code is licensed under a permissive license and contributions are welcomed. The Orthogonal platform itself (the registry, API service, billing) is proprietary, but agents can fork and host their own skills registry or modify the CLI. The MCP server implementation is also available in the GitHub repository, allowing you to audit how find-skill works and run it in your own environment if needed.
What are the best alternatives to find-skill?
The main alternatives are: (1) Composio — offers 150+ pre-built integrations with built-in auth but charges per-seat; use Composio if you need more integrations and prefer subscription pricing. (2) GitHub MCP Server — free, built-in support for GitHub workflows; use this if your agents primarily work within GitHub. (3) LangChain Tools or CrewAI integrations — self-managed, no centralized discovery, but give you full control. (4) Home-rolled skill registry — build your own discovery system if you need complete customization. find-skill stands out for its low cost (pay-per-use), multi-agent compatibility, and growing library (50+ skills and growing).
Who is find-skill best for?
find-skill is best for AI engineers and teams building autonomous agent systems that need runtime flexibility—teams that can't predict in advance which integrations they'll need or want agents to discover capabilities on demand. It's ideal for startups using Claude Code or Cursor agents for sales, research, or operations workflows, and for enterprise teams deploying agents across multiple departments where different workflows need different skills. It's not ideal for simple single-purpose agents that only need 1-2 fixed integrations, where direct SDK imports are simpler. Teams with strict compliance requirements should audit Orthogonal's security model before relying on it for sensitive data workflows.
How does find-skill compare to Composio in 2026?
find-skill and Composio both help agents discover and use integrations, but differ in scope, cost, and discovery model. Composio offers 150+ integrations out of the box with built-in OAuth and account linking, charging per-user seat or per-request; it's a more complete platform. find-skill offers 50+ integrations (growing) with unified authentication (one Orthogonal API key), charging only per actual API call; it's simpler and cheaper for scaling but requires agents to discover skills autonomously. Composio is better if you need breadth and are comfortable with subscription pricing; find-skill is better if you prefer pay-as-you-go, multi-agent compatibility, and autonomous agent discovery. Both work with Claude Code and Cursor.