AgentsKit Registry

AgentsKit Registry

Profile: top-level-repository

Copy a production-minded AI agent into your project and own the source. AgentsKit Registry is a shadcn-style catalog of 346 validated agents built with AgentsKit. It adds no Registry runtime and no lock-in: the CLI copies readable TypeScript that your team can review and edit.

It is intended for teams that want a ready agent starting point without adopting a proprietary registry runtime.

Tags: agentskit · agent-registry · typescript · ai-agents · shadcn-style

Topics: ai-agents · registry · developer-experience

Browse the Registry · Start in five minutes · Contribute an agent · Security · Governance · Code of Conduct

Verified proof

• Catalog counts come from public/r/index.json and ecosystem-claims.json.
• Clean-fixture install/run contract: npm run test:quickstart (scripts/quickstart.test.mjs).
• Structural validation and Doc Bridge gates: npm run validate, npm run docs:bridge:gate.
• Deterministic discovery build: npm run discovery:build.
• Credential-free catalog proof: node examples/verify-readme.mjs.

Validate agents in GitHub Actions

Use the Registry contract as a CI gate for copy-owned agent folders in another repository. The action checks required source, test, metadata, and README files; metadata IDs and referenced files; and the Registry content policy. It does not send source code to an API or require credentials.

name: Validate agents

on:
  pull_request:
  push:
    branches: [main]

jobs:
  agentskit:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: AgentsKit-io/agentskit-registry@v1
        with:
          path: agents

The path input may point to one agent folder or to a directory whose immediate children are agent folders.

Registry is beta. All 346 entries pass structural and repository tests; 333 also publish sanitized independent-review evidence. Templates remain code you must review against your provider, tools, data policy, and risk profile.

From catalog to code

flowchart LR
  F["Find by ID, category, or capability"] --> A["npx agentskit add <id>"]
  A --> C["Readable source copied locally"]
  C --> R["Review and adapt"]
  R --> S["Ship with your adapter and policy"]

npx agentskit add research

import { openai } from '@agentskit/adapters'
import { createResearchAgent } from './agents/research/agent'

const agent = createResearchAgent({
  adapter: openai({
    apiKey: process.env.OPENAI_API_KEY!,
    model: 'gpt-4o',
  }),
})

const result = await agent.run('What changed in the EU AI Act?')
console.log(result.content)

Swap the adapter, tools, memory, model, and policy through supported AgentsKit contracts. The copied factory stays ordinary application code.

What is actually here

• 346 validated, installable agents across 23 categories, from coding and security to legal, clinical, marketing, operations, support, and research.
• 333 independent-review summaries with scores, confidence, strengths, and reviewer notes; private raw reviewer output is never published.
• Typed source + focused tests + usage guide for every installable entry.
• Eval replay fixtures for repeatable behavioral checks without live keys.
• JSON, llms.txt, MCP, and deterministic discovery artifacts for people, tools, and LLMs to navigate the same canonical catalog.

Claims are derived from public/r/index.json and the generated ecosystem-claims.json, not hand-maintained marketing copy.

Discovery without unnecessary backend calls

Exact agent IDs, titles, install commands, categories, capabilities, contribution links, and ecosystem navigation resolve from a local artifact pinned by content hash. Semantic requests such as “which agent fits my incident workflow?” escalate through AgentsKit Chat to the trusted cited backend only after a validated local miss.

See architecture and ownership for the full boundary.

Repository ownership

This repository owns catalog data and generated machine artifacts. The live Fumadocs application and its AgentsChat presentation live in AgentsKit-io/agentskit/apps/registry. AgentsKit owns runtime primitives and the CLI; AgentsKit Chat owns the shared answer protocol and renderer behavior. There is no second framework here.

Contributing

New agents and improvements are welcome. Follow CONTRIBUTING.md, use a neighboring validated agent as a reference, and run:

npm ci
npm run check:no-legacy-chat-imports
npm run docs:bridge:config
npm run validate
npm run lint
npm test
npm run test:quickstart
npm run eval:run -- --ecosystem-doc-bridge
npm run build
npm run docs:bridge:index
npm run docs:bridge:gate
npm run docs:bridge:conformance

Generated indexes, claims, deterministic artifacts, and Doc Bridge handoffs must remain fresh in the pull request.

Maturity

Registry is beta. All 346 entries pass structural and repository tests; 333 also publish sanitized independent-review evidence. Templates remain code you must review against your provider, tools, data policy, and risk profile.

AgentsKit ecosystem

• AgentsKit — build the agent runtime.
• Registry — start from owned, ready-made source.
• AgentsKit Chat — deliver one agent experience across Web, native, and terminal interfaces.
• Doc Bridge — turn documentation into executable agent handoffs. Optional companion for repositories that install agents: how it fits.
• Code Review — run deep, low-noise review with the model already in use.
• Harness — the keep-pushing delivery loop for your SDLC.

Compatibility and license

Registry source targets the package ranges declared by each meta.json; the repository CI currently runs on Node.js 22. Review an agent's generated bundle before upgrading its dependencies. Licensed under MIT.