Agent Archive

An open-source library of practical AI agents, with prompts, runbooks, evaluations, and install kits for Codex, Claude Code, and Cursor.

Live site · Features · Quick start · Install kits · Contributing · 한국어

Why Agent Archive?

Most prompt collections stop at "copy this text". In a real project you also need to know:

• When an agent is the right tool, and when it isn't
• What context to give it, and what a weak input looks like
• How to judge the output before you trust it
• How to install it where you work: AGENTS.md, CLAUDE.md, or a Cursor rule

Agent Archive treats every agent as a reusable work recipe. Each one documents its inputs, outputs, limits, and review criteria, and the project-ready agents can be dropped straight into a repository.

Features

• 100 agents across 9 roles and 11 categories, from PRD review to incident postmortems
• 20 project-ready agents with runbooks, quality evaluations, sample runs, and one-click install kits
• Install kits for Codex (AGENTS.md), Claude Code (CLAUDE.md), and Cursor (.mdc rule), plus agent.json, RUNBOOK.md, and EVALUATION.md
• Workflow packs that chain several agents into end-to-end playbooks, with a visual handoff at each step
• Starter packs that bundle agents for engineering, product planning, design QA, and operations
• Search and filters by role, category, tool, difficulty, and automation level, with a project-ready-only toggle
• Portable exports: copy the full agent card or download it as Markdown or JSON
• Bookmarks stored locally in your browser; no account required
• English and Korean UI

Agent detail: expected effect, runbook, evaluation	Install: starter packs and project-ready agents

Quick start

Requirements: Node.js 20.9 or newer, and npm.

git clone https://github.com/yohan-work/agentive.git
cd agentive
npm install
npm run dev

Open http://localhost:3000. The app redirects to /en; switch to Korean with /ko.

Install kits

Open any project-ready agent and choose Download Kit in the Use this agent panel. You get:

pr-review-agent-AGENTS.md         # Codex / generic AGENTS.md instructions
pr-review-agent-CLAUDE.md         # Claude Code project instructions
pr-review-agent-cursor-rule.mdc   # Cursor project rule
pr-review-agent-agent.json        # machine-readable manifest
pr-review-agent-README.md         # setup notes
pr-review-agent-RUNBOOK.md        # context to prepare, good/bad inputs, output checklist
pr-review-agent-EVALUATION.md     # quality score, known weaknesses, sample runs

Or fetch the same kit from a terminal. Every file has a stable URL at /kits/<slug>/<file>, and the agent page shows the exact command:

mkdir -p agent-kits/pr-review-agent && (cd agent-kits/pr-review-agent && curl -fsSL --remote-name-all \
  "https://yohan-work.github.io/agentive/kits/pr-review-agent/{AGENTS.md,CLAUDE.md,cursor-rule.mdc,agent.json,README.md,RUNBOOK.md,EVALUATION.md}")

This saves the files into agent-kits/pr-review-agent/ without their slug prefix and never touches your existing project files.

Then place the files where your tool expects them:

Tool	Where to put it
Codex and other AGENTS.md-aware agents	Append the contents to AGENTS.md at your repo root
Claude Code	Append the contents to CLAUDE.md at your repo root
Cursor	Save as .cursor/rules/<name>.mdc

How it works

Agent Archive is a static-data-first Next.js app with no database or backend, exported to static HTML and deployed to GitHub Pages on every push to main. Each agent is a single YAML file in content/agents/, validated against a JSON Schema. Workflows and taxonomy are typed data in src/data.

content/
└── agents/           # One YAML file per agent (the source of truth); _template.yaml to start
schema/
└── agent.schema.json # JSON Schema generated from src/types/agent.ts
src/
├── app/[locale]/     # Next.js App Router pages (/en, /ko), statically exported
├── components/       # UI: agents, workflows, layout, common primitives
├── data/             # Workflows, starter packs, taxonomy, and the agent loader
├── i18n/             # Locale config and UI dictionaries
├── lib/              # Search, export, and install-kit generation
└── types/            # Agent, workflow, and taxonomy types
scripts/
├── build-content.mjs # Bundles content/agents into the app
├── generate-schema.mjs
└── check-data.mjs    # Schema and integrity checks (slugs, references, required metadata)

Command	What it does
npm run dev	Start the dev server
npm run build	Static export to out/ (set NEXT_PUBLIC_BASE_PATH for a sub-path)
npm start	Preview out/ locally the way GitHub Pages serves it
npm run lint	ESLint
npm run typecheck	TypeScript type check
npm test	Unit tests for install kits, search, and locale helpers
npm run check:data	Validate agents against the schema, plus workflow and taxonomy references
npm run schema	Regenerate schema/agent.schema.json

Contributing

Contributions are welcome, especially new agents, better prompts, and honest evaluations. Read the contributing guide for what we accept and how agent verification levels work.

• Suggest an agent: open a Suggest an agent issue, or use the Submit page on the site.
• Add or improve an agent: copy content/agents/_template.yaml (or edit an existing file in content/agents/), run npm run check:data, and open a pull request.
• Report a bug: open a bug report.
• Security issues: please report them privately (see SECURITY.md).

This project follows the Contributor Covenant. AI coding agents should read AGENTS.md.

Roadmap

• One file per agent, with schema validation
• Hosted site with a public URL (yohan-work.github.io/agentive)
• Download install kits straight from a URL, no browser needed
• More hand-verified agents and sample runs

License

• Code: MIT
• Agent content in content/ and src/data (prompts, runbooks, evaluations, workflows): CC BY 4.0

You can use exported prompts and install kits in your own projects, including commercial ones. Attribution is appreciated.