Language: English · Türkçe
The project-aware agent team for every harness.
Not just configs — a system. Install once and every project gets a 12-role specialist
team that adapts to that project's stack, reviews your diffs, and remembers your decisions.
Works across Claude Code, Codex, Cursor, OpenCode, Gemini, Windsurf, GitHub Copilot, Cline,
Zed — and anything that reads AGENTS.md.
🌐 Website · Quick start · How it works · Harnesses · Guide · Contribute
✨ 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.ymlbefore 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
| A pile of agent prompts | agent-hub | |
|---|---|---|
| Per-project fit | You re-explain the stack every time | Reads a profile.yml automatically |
| Memory | Forgets decisions between sessions | Records & honors ADRs |
| Consistency | Different setup per repo & per tool | One team, every repo, every harness |
| Orchestration | You route the work | tech-lead decomposes & delegates |
| Quality gate | Ad-hoc | Built-in code-reviewer + security-reviewer |
| Portability | Locked to one tool | Claude Code, Codex, Cursor, Gemini, Copilot… |
🚀 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
| Harness | What gets installed |
|---|---|
| Claude Code | Plugin with 12 real subagents + CLAUDE.md |
| Codex | AGENTS.md |
| Cursor | .cursor/rules/agent-hub.mdc |
| OpenCode | AGENTS.md |
| Gemini CLI | GEMINI.md |
| Windsurf | .windsurfrules |
| GitHub Copilot | .github/copilot-instructions.md |
| Cline / Roo | .clinerules |
| Zed | .rules |
Anything else that reads AGENTS.md | AGENTS.md |
One source of truth in adapters/ generates each harness's file.
🎬 See it
One line installs the team; onboard detects your stack; tech-lead routes the work.
🧠 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
| Agent | Role | Model |
|---|---|---|
tech-lead | Orchestrator — decomposes work, routes to specialists | opus |
architect | System design, data models, API contracts | opus |
backend-engineer | APIs, business logic, persistence | sonnet |
frontend-engineer | Web UI, state, accessibility, performance | sonnet |
mobile-engineer | Native / cross-platform mobile | sonnet |
devops-engineer | CI/CD, infra, observability | sonnet |
data-engineer | Pipelines, ETL, schema, migrations | sonnet |
ux-designer | User flows, IA, accessibility | opus |
code-reviewer | Correctness, bugs, maintainability | sonnet |
qa-tester | Test plans and test code | haiku |
security-reviewer | Vulnerabilities, authz/authn, secrets | opus |
tech-writer | Docs, READMEs, changelogs | haiku |
(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.