agent-hub

✨ The idea

You don't want a different assistant per project, or a different setup per tool. You want one great team that knows which project it's in — and works in whatever harness you use.

Open your backend repo → the team thinks in your backend stack. Open your web repo → the same team thinks frontend. Same roster. Different brain. Zero manual switching.

Three things make it a system, not a pile of prompts:

• 🧠 Project-aware — every agent reads a tiny per-project profile.yml before acting.
• 📝 Memory — decisions are recorded as ADRs the whole team reads and honors.
• ⚡ Auto-load — the profile loads the moment you open the project.

🆚 Why agent-hub

🚀 Quick start

One line — any harness:

curl -fsSL https://raw.githubusercontent.com/berkcangumusisik/agent-hub/main/install.sh | sh

This scaffolds the project profile and installs the team for every supported harness. Want just one? Pass it: … | sh -s -- cursor (or claude, codex, gemini, opencode, windsurf).

Windows (PowerShell):

irm https://raw.githubusercontent.com/berkcangumusisik/agent-hub/main/install.ps1 | iex

Claude Code (richest experience — real subagents):

/plugin marketplace add berkcangumusisik/agent-hub
/plugin install super-team@agent-hub

Then open your repo with your agent and say “onboard” — it detects your stack and fills the profile. From there, just describe what you want; the tech-lead routes the work.

🧰 Supported harnesses

One source of truth in adapters/ generates each harness's file.

🎬 See it

🧠 How it works

Two layers: a universal team (this repo) and a per-project profile. Every agent runs a profile-first protocol — read the profile and decisions, then act in that project's idioms. Read the full walkthrough in the Getting Started guide.

👥 The roster

(On non-Claude harnesses the single agent wears these hats; on Claude Code they're real, parallel subagents.)

⌨️ Commands

onboard · team · review · decide: <X> · handoff — available as /commands in Claude Code, or as plain words in any harness reading AGENTS.md.

📄 The profile

.claude/agent-hub/profile.yml is the single source of truth the team reads first:

project: { name: my-service, type: backend }
stack: { language: TypeScript, framework: NestJS, database: PostgreSQL }
commands: { install: pnpm install, test: pnpm test }
team:
  active: [tech-lead, architect, backend-engineer, code-reviewer, security-reviewer]
conventions:
  - Use the repository pattern for data access
doNot:
  - Never expose entities directly in API responses

Starter profiles for common stacks live in templates/profiles/.

🗂 Repo layout

agent-hub/
├── install.sh · install.ps1            # one-line installers
├── adapters/                           # one source → AGENTS.md, CLAUDE.md, GEMINI.md, cursor, windsurf
├── plugins/super-team/                 # Claude Code plugin (12 subagents, skills, hooks, commands)
├── templates/                          # per-project profile + starter stacks
├── docs/                               # guide, diagrams, website
├── site/                               # marketing site (Vite + React)
└── scripts/validate.mjs                # zero-dep CI validation

🤝 Contributing & community

PRs welcome — a sharper agent prompt, a new harness adapter, a starter profile. See CONTRIBUTING.md. Questions and show-and-tell go in Discussions.

⭐ Like it?

If agent-hub saves you time, star the repo — it helps others find it.

📝 License

MIT © Berkcan Gümüşışık