Ai Team Skills

🤖 AI Team Skills

A complete, autonomous AI engineering team for your codebase — across every major AI IDE.

[](https://github.com/ByRohat/ai-team-skills) [](LICENSE) [](#-the-team) [](#-platform-support) [](#-contributing)

18 specialized agents · 4 core protocols · cross-session memory · fully autonomous activation

[Quick Start](#-quick-start) • [Platforms](#-platform-support) • [The Team](#-the-team) • [How It Works](#-how-it-works) • [Quality Gates](#-quality-gates) • [Brain](#-brain-storage)

---

Overview

AI Team Skills turns any AI coding assistant into a coordinated, enterprise-grade engineering team. Instead of one generalist model, you get 18 specialized agents — from Product Owner to SRE — that share a durable memory, enforce quality gates, and adapt to your project.

It activates itself. Open your project in a supported IDE and the team reads its brain, understands where the project stands, and gets to work — you never type "activate."

> Why it's different: most AI assistants forget everything between chats and give generic, textbook advice. AI Team Skills gives the model a persistent brain and four operating protocols so it behaves like a senior team with perfect long-term memory of your codebase.

---

✨ Highlights

• ⚡ Fully Autonomous Activation — Each IDE auto-loads its native rules file and the team activates on the first message. It then operates on its own, pausing only for costly/irreversible decisions and never bypassing the safety gates.
• 🧠 Smart by Default — Four cross-cutting protocols (Continuity, Adaptation, Self-Evolution, Clarification) make every agent project-aware, self-adapting, and continuous across sessions.
• 🧬 Project DNA — project-profile.json captures the project's stack, domain, conventions, glossary, and key decisions, so any session instantly understands what it's building.
• 🛡️ Quality Gates with Teeth — Security holds VETO power; SRE holds DEPLOYMENT HOLD power. Nothing ships until coverage, performance, observability, and docs all pass.
• 🌍 Truly Cross-Platform — Claude Code, Cursor, Windsurf, Antigravity/Gemini, Trae, OpenAI Codex CLI, GitHub Copilot, and any generic assistant — wired to each tool's real auto-load mechanism.
• 💾 Cross-Session Memory — Versioned brain files (JSON/TOML) preserve context, decisions, learnings, and an audit trail.
• 🪶 Token-Optimized — Compact JSON/TOML state and progressive-disclosure skill files keep context usage lean.

---

🚀 Quick Start

One-line install

# macOS / Linux
curl -sL https://raw.githubusercontent.com/ByRohat/ai-team-skills/main/install.sh | bash

# Windows PowerShell
irm https://raw.githubusercontent.com/ByRohat/ai-team-skills/main/install.ps1 | iex

Manual install

git clone https://github.com/ByRohat/ai-team-skills.git
cd ai-team-skills

# Auto-detect your platform, or pass one explicitly
./install.sh /path/to/your/project            # auto-detect
./install.sh /path/to/your/project all        # install for every platform

# Windows
.\install.ps1 C:\path\to\your\project

That's it. Open your project in your AI IDE — the team activates itself and greets you with a status report.

---

📦 Platform Support

Each platform is wired to the exact file it auto-loads, so the team starts with no manual trigger.

| Platform | Auto-loaded file(s) | Auto-Start | |----------|---------------------|:----------:| | Claude Code | CLAUDE.md | ✅ | | Cursor | .cursor/rules/ai-team.mdc (alwaysApply) + .cursorrules | ✅ | | Windsurf | .windsurf/rules/ai-team.md (always_on) + .windsurfrules | ✅ | | Antigravity / Gemini | GEMINI.md + AGENTS.md + .agents/rules/ai-team.md | ✅ | | Trae IDE | .trae/rules/ai-team.md | ✅ | | OpenAI Codex CLI | AGENTS.md | ✅ | | GitHub Copilot / VS Code | .github/copilot-instructions.md (+ .vscode/ settings) | ✅ | | Generic / Other | AI-TEAM.md | ✅ |

---

🧠 How It Works

Every agent runs four cross-cutting Core Protocols ([_core-protocols.md](.claude/skills/_core-protocols.md)) on top of its domain expertise. Together they make the team behave like one engineer with perfect long-term memory of your project.

| Protocol | Guarantee | |----------|-----------| | 🔄 Continuity | Open a new chat → the model reads the brain (project-profile → project-state → per-agent) and reconstructs where the project stands before acting. No lost context. | | 🎯 Adaptation | Every recommendation is tailored to the project's real stack, conventions, and glossary — never generic. Empty profile? It detects the stack and populates it. | | 🌱 Self-Evolution | The team writes project-specific learnings to the brain and proposes skill improvements for your approval — it never edits its own skills silently. | | ❓ Clarification | The model asks when a wrong assumption would be costly or hard to reverse (schema, public API, auth, integrations, architecture). For cheap/reversible choices it proceeds and states its assumption. |

And one meta-protocol that ties it together:

> Protocol 0 — Autonomous Activation: start on the first message without being asked; operate through the task queue on your own; pause only for the Clarification guardrails; never auto-bypass the Security VETO or SRE DEPLOYMENT HOLD.

---

👥 The Team

| Agent | Role | Special Power | |-------|------|:-------------:| | Product Owner | Backlog grooming, user stories, RICE scoring, OKR alignment | — | | Team Lead | Orchestrates all agents, tracks sprint velocity & risk register | — | | Technology Strategist | Technology Decision Protocol — needs-based stack/language/architecture selection, decision matrix, alternatives, draft ADR | — | | Architecture | System design, event-driven patterns, C4 diagrams, ADR registry | — | | UX/UI Designer | Design systems, tokens, wireframes, WCAG 2.2, design-dev handoff | — | | AI Engineer | LLM integrations, RAG pipelines, multi-agent orchestration, LLMOps | — | | Data Engineer | ETL/ELT, dbt modeling, Kafka streaming, data quality, Airflow | — | | Backend | API development, business logic, DB queries, validation schemas | — | | Frontend | UI components, WCAG 2.2 a11y, Core Web Vitals, i18n | — | | Mobile Engineer | iOS/Android/React Native/Flutter, offline-first, app store | — | | DevOps | Containerization, CI/CD, Kubernetes, secrets management | — | | SRE | SLO/SLI/error budgets, incident mgmt, chaos engineering, capacity | 🟦 HOLD | | Performance | Latency budgets, query profiling, k6 load testing, CDN caching | — | | Observability | OpenTelemetry, Prometheus, structured logs, Grafana | — | | Security | OWASP 2024, SAST/DAST, threat modeling | 🟥 VETO | | Privacy | GDPR, CCPA, KVKK compliance, DPIA evaluations | — | | QA | Unit/integration/E2E, contract (Pact), mutation (Stryker) tests | — | | Docs | OpenAPI 3.1 specs, runbooks, post-mortems, changelogs | — |

Agent dependency flow

                  ┌──────────────┐
                  │ Product Owner│
                  └──────┬───────┘
                  ┌──────▼───────┐
                  │  Team Lead   │
                  └──────┬───────┘
        ┌────────────────┼───────────────────┐
        ▼                ▼                    ▼
 ┌──────────────┐ ┌──────────────┐   ┌──────────────┐
 │ Architecture │ │    DevOps    │   │     Docs     │
 └──────┬───────┘ └──────┬───────┘   └──────────────┘
        │                ▼
  ┌─────┼──────┐  ┌──────────────┐
  ▼     ▼      ▼  │Observability │
 UX  AI Eng  Data └──────┬───────┘
  │     │      │         │
  ▼     ▼      ▼         │
 Frontend / Backend / Mobile
        │               │
        └──────► QA ◄────┘
                 │
                 ▼
            Performance
                 │
                 ▼
         Security & Privacy
                 │
                 ▼
                SRE
                 │
                 ▼
            DEPLOYMENT

---

🚫 Quality Gates

Deployment is blocked until every gate passes.

| Gate | Requirement | |------|-------------| | 🟥 Security (VETO) | 0 open Critical/High vulnerabilities · secret scan clean · CSP/HSTS active | | 🟦 Reliability (SRE HOLD) | SLOs defined · error-budget burn alerts (fast + slow) · runbooks & on-call ready | | 🟧 Quality | Unit ≥ 80% · Integration ≥ 70% · contract tests pass · mutation ≥ 60% | | 🟨 Performance | Lighthouse ≥ 90 · LCP -brain.json # per-agent memory (×17)

Every `-brain.json` carries continuity/evolution fields — `last_session_summary`, `learnings`, `conventions_used`, `open_questions`, `proposed_improvements`. The golden rule: *an action not written to the brain didn't happen as far as the next session is concerned.*

---

## 📝 Slash Commands

| Command | Description |
|---------|-------------|
| `/team-status` | Full team report with all component statuses |
| `/team-blockers` | List active blockers |
| `/team-next` | Fetch the next task from the queue |
| `/deploy-check` | Deployment-readiness verification (runs the gates) |
| `/team-init` | Initialize project configuration |
| `/team-stack` | Technology Decision Protocol — needs-based stack/language/architecture selection |
| `/team-sprint` | Sprint planning & velocity |
| `/team-adr` | Generate an Architecture Decision Record |
| `/team-retro` | Generate a sprint retrospective |
| `/team-risk` | Show the risk register |
| `/team-slo` | SRE SLO dashboard & error budgets |
| `/team-data` | Data Engineer pipeline status |

---

## 📏 File Size Rules

**Maximum 1000 lines per file.**

- **0–700 lines** — normal development
- **700–1000 lines** — ⚠️ prepare to split
- **1000+ lines** — 🚫 split immediately at logical boundaries

---

## 🗂️ Project Structure

ai-team-skills/ ├── .claude/skills/ # 18 agent skill files + _core-protocols.md (source of truth) ├── .cursor/ .windsurf/ # skill mirrors + native auto-load rules ├── .trae/ .agents/ .github/ # platform-native auto-load rule files ├── CLAUDE.md AGENTS.md GEMINI.md ANTIGRAVITY.md AI-TEAM.md ├── .cursorrules .windsurfrules ├── install.sh install.ps1 # cross-platform installers ├── skill.json # canonical manifest (v2.4.0) ├── src/ai-team/ │ ├── brain/ # brain templates (profile + state + 18 agents) │ └── scripts/ # brain manager, agent comms, auto-activator (Python) └── cli/ # TypeScript CLI (aiteam)

---

## 🤝 Contributing

Contributions are welcome! A few conventions keep the framework consistent:

- **Source of truth** is `.claude/skills/` — mirror any change to `.cursor/skills/` and `.windsurf/skills/`.
- **Brain templates** live in `src/ai-team/brain/` (the installer copies them).
- **1000-line hard limit** per file (warn at 700).
- **Don't edit skills silently** — propose changes via `.ai-team/brain/proposed-improvements.md`.

Open an issue or PR at [github.com/ByRohat/ai-team-skills](https://github.com/ByRohat/ai-team-skills).

---

## 📄 License

[MIT](LICENSE) © 2026 **ByRohat** — free to use, modify, and distribute.

**Built with ❤️ for autonomous, full-stack AI development.**

## Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [byrohat](https://github.com/byrohat)
- **Source:** [byrohat/ai-team-skills](https://github.com/byrohat/ai-team-skills)
- **License:** MIT

Install and usage instructions live in the source repository linked above.