🚀 TRAE Kit — Multi-Agent System

Unlock the full power of Trae IDE. Bypass SOLO mode. Deploy 20 specialist agents.

The missing layer between you and Trae IDE's full potential.

Quick Install · How it Works · Agents · Skills

⚡ Quick Install

npx --yes trae-kit init

Or manually:

git clone https://github.com/PedroIves/TRAE_Kit-Multi-Agents.git
cp -r TRAE_Kit-Multi-Agents/.trae /your-project/
cp TRAE_Kit-Multi-Agents/AGENTS.md /your-project/

Windows (PowerShell):

git clone https://github.com/PedroIves/TRAE_Kit-Multi-Agents.git
Copy-Item -Recurse TRAE_Kit-Multi-Agents\.trae "C:\your-project\"
Copy-Item TRAE_Kit-Multi-Agents\AGENTS.md "C:\your-project\"

🤖 Create the dev Agent

This is the central router — the most important setup step.

1. In Trae, go to Agents → Create Agent
2. Name: dev
3. Enable all tools: Read ✅ Edit ✅ Terminal ✅ Preview ✅ Web search ✅
4. Paste the entire block below into the Prompt field:

You are the central development agent for this project (TRAE Kit).

## Role
Route every task to the correct specialist. You are the entry point — not the executor.
Read .trae/agents/{specialist}.md before implementing anything.

## Protocol
1. Classify the request (question / survey / simple code / complex / design / command)
2. Select the specialist agent for the domain
3. Announce: 🤖 **@[agent-name] activated**
4. Read the agent file + relevant skills
5. Execute with full specialist depth

## Routing (quick reference)
- UI / layout / design   → @frontend-specialist
- API / backend / DB     → @backend-specialist
- bug / error / crash    → @debugger
- deploy / CI-CD         → @devops-engineer
- plan / roadmap         → @project-planner
- security / auth        → @security-auditor
- test / TDD / coverage  → @test-engineer
- mobile / RN / Flutter  → @mobile-developer
- optimize / perf        → @performance-optimizer
- docs / README          → @documentation-writer

## Non-negotiables
- Never guess file content — read first, then edit
- Never write code before clearing the Socratic Gate on complex tasks
- Never use placeholders, TODOs, or omit files from delivery
- All project rules in .trae/rules/ are already loaded — apply them

5. Click Save — done.

💡 What is this?

Trae IDE ships with an AI assistant locked in SOLO mode — one generic model doing everything. This kit bypasses that limitation by installing a routing layer that automatically selects the right specialist for every task.

You type → dev agent reads context → right specialist activated → expert-level response

| Without this kit | With this kit | |---|---| | 1 generic AI for everything | 20 domain specialists | | Heavy fixed context (4400+ tokens) | Optimized fixed context (~1312 tokens) | | No design data | 50+ styles, 97 palettes, 57 font pairs | | No workflows | 11 automatic slash-command workflows | | Manual prompting every time | Automatic agent routing |

What "bypassing SOLO" means

Trae's SOLO mode restricts the AI to a single agent context. This kit installs an orchestration layer via .trae/rules/ and AGENTS.md that:

• Routes requests to specialist agents automatically
• Loads only the relevant skills per task (not everything at once)
• Reduces wasted tokens by 70% vs a naive always-apply setup
• Makes agents callable from each other for complex multi-step tasks

⚙️ How it Works

┌─────────────────┐     classify      ┌──────────────────────┐
│   Your Input    │ ────────────────▶  │    dev  (router)     │
└─────────────────┘                   └──────────────────────┘
                                                 │
             ┌───────────────────────────────────┼───────────────────┐
             ▼                                   ▼                   ▼
  ┌─────────────────────┐         ┌─────────────────┐    ┌──────────────────────┐
  │ frontend-specialist │         │    debugger     │    │ backend-specialist   │
  │  + frontend-design  │         │  + sys-debug    │    │  + api-patterns      │
  └─────────────────────┘         └─────────────────┘    └──────────────────────┘
             │                                   │                   │
             └───────────────────────────────────┴───────────────────┘
                                                 │
                                                 ▼
                                      Expert-level response

The 3-layer architecture

Layer 1 — Always Active (~1312 tokens)
  CORE.md          → routing protocol, request classifier, tier rules
  core-always.md   → security rules, delivery checklist

Layer 2 — On Demand (loaded only when relevant)
  rules-tsx.md           → React/TSX/Tailwind rules
  architecture.md        → system design, ADR framework
  CORE-REFERENCE.md      → agent map, script list
  workflow-*.md (×11)    → /debug, /plan, /deploy, /ui-ux...

Layer 3 — Per Task (loaded by the active agent)
  .trae/agents/{name}.md → full specialist persona
  .trae/skills/{name}/   → domain-specific techniques

🤖 Agents

20 specialist agents, each with a full persona, domain rules, and linked skills:

| Agent | Domain | Key Skills | |---|---|---| | frontend-specialist | React, Next.js, UI/UX | frontend-design, tailwind-patterns, nextjs-react-expert | | backend-specialist | API, Node.js, Python | api-patterns, nodejs-best-practices, python-patterns | | debugger | Bugs, errors, crashes | systematic-debugging | | devops-engineer | CI/CD, Docker, deploy | deployment-procedures, bash-linux | | project-planner | Roadmap, architecture | plan-writing, architecture | | security-auditor | Auth, JWT, OWASP | vulnerability-scanner, red-team-tactics | | test-engineer | Unit, E2E, TDD | tdd-workflow, testing-patterns, webapp-testing | | mobile-developer | React Native, Flutter | mobile-design, geo-fundamentals | | performance-optimizer | Cache, bundle, profiling | performance-profiling | | database-architect | Schema, ORM, queries | database-design | | documentation-writer | README, JSDoc, ADRs | documentation-templates | | orchestrator | Multi-agent, complex tasks | parallel-agents, intelligent-routing | | qa-automation-engineer | Playwright, Cypress | webapp-testing | | seo-specialist | Meta, Core Web Vitals | seo-fundamentals | | penetration-tester | Pen test, red team | red-team-tactics, vulnerability-scanner | | product-manager | PRD, features, strategy | brainstorming, plan-writing | | product-owner | Backlog, sprints | plan-writing | | game-developer | Unity, game loops | game-development | | code-archaeologist | Legacy code, refactor | systematic-debugging | | explorer-agent | Research, discovery | brainstorming, mcp-builder |

🧠 Skills

50 modular skill packs — each with a SKILL.md index file loaded selectively:

| Skill | Size | Purpose | |---|---|---| | agent-frontend-specialist | 26163c | Deep frontend specialization | | agent-mobile-developer | 13296c | Mobile specialization | | agent-orchestrator | 14266c | Multi-agent orchestration | | frontend-design | 12618c | Design system, components | | mobile-design | 15203c | Native mobile design | | intelligent-routing | 10858c | Agent routing logic | | nextjs-react-expert | 9971c | Advanced Next.js | | clean-code | 6759c | Quality standards | | python-patterns | 9077c | Python best practices | | vulnerability-scanner | 7651c | Security scanning | | rust-pro | 7730c | Advanced Rust | | nodejs-best-practices | 8147c | Node.js patterns | | tailwind-patterns | 7026c | Tailwind CSS | | parallel-agents | 5485c | Parallel execution | | behavioral-modes | 5559c | Behavior modes | | deployment-procedures | 5837c | Deploy & CI/CD | | brainstorming | 4757c | Structured ideation | | plan-writing | 4128c | Planning documents | | bash-linux | 4355c | Shell scripting | | game-development | 4814c | Game dev patterns | | tdd-workflow | 2940c | Test-driven dev | | testing-patterns | 3662c | Testing strategies | | webapp-testing | 3972c | E2E testing | | database-design | 1563c | Schema design | | api-patterns | 2506c | API design | | documentation-templates | 3500c | Doc templates | | seo-fundamentals | 2912c | SEO techniques | | performance-profiling | 3230c | Perf analysis | | mcp-builder | 3458c | MCP servers | | ...and 21 more | | |

📏 Rules & Context Window

╔══════════════════════════════════════════════════════╗
║           CONTEXT WINDOW BREAKDOWN                   ║
╠══════════════════════════════════════════════════════╣
║  ALWAYS APPLY (every request)                        ║
║  ├── CORE.md           3633 chars  (routing protocol)║
║  ├── core-always.md     737 chars  (security + QA)   ║
║  └── AGENTS.md (toggle) 878 chars  (skill map)       ║
║                                                      ║
║  TOTAL FIXED:  ~5248 chars  (~1312 tokens) ✅        ║
╠══════════════════════════════════════════════════════╣
║  INTELLIGENTLY (loaded only when needed)             ║
║  ├── architecture.md     1688c  → design tasks       ║
║  ├── CORE-REFERENCE.md   1726c  → reference lookups  ║
║  ├── rules-tsx.md        1521c  → React/TSX work     ║
║  └── workflow-*.md ×11  ~25kc   → slash commands     ║
╚══════════════════════════════════════════════════════╝

Before optimization: ~17704 chars (~4426 tokens) 🔴
After optimization:   ~5248 chars (~1312 tokens) ✅
Reduction: 70% less wasted context per request

🛠 Utility Scripts

python .trae/scripts/checklist.py .
python .trae/scripts/checklist.py . --url https://your-app.com
python .trae/scripts/verify_all.py
python .trae/scripts/auto_preview.py

Checklist order: Security → Lint → Schema → Tests → UX → SEO → Lighthouse → E2E

🎨 UI/UX Data System

| Dataset | File | Contents | |---|---|---| | Styles | styles.csv | 50+ visual styles | | Colors | colors.csv | 97 complete palettes | | Typography | typography.csv | 57 font pairings | | UX Guidelines | ux-guidelines.csv | 99 UX directives | | Stacks | stacks/*.csv | Next.js, React, Flutter, Vue, Svelte, SwiftUI... |

python .trae/shared/ui-ux-pro-max/scripts/search.py "dark minimalist"
python .trae/shared/ui-ux-pro-max/scripts/design_system.py

📁 Project Structure

TRAE_Kit-Multi-Agents/
├── .trae/
│   ├── agents/          # 20 specialist agent files
│   ├── rules/           # 16 rule files (2 always + 14 intelligently)
│   ├── skills/          # 50 skill modules
│   ├── scripts/         # 4 utility Python scripts
│   └── shared/          # UI/UX data (517kb)
├── bin/cli.js           # npx trae-kit CLI
├── AGENTS.md            # Skill routing map
├── package.json
└── README.md

🤝 Contributing

git checkout -b feat/agent-my-specialist
cp .trae/agents/debugger.md .trae/agents/my-specialist.md
python .trae/scripts/verify_all.py
git commit -m "feat: add my-specialist agent"
git push origin feat/agent-my-specialist

📄 License

MIT License — © 2026 PedroIves — Free for personal and commercial use.

Made with ☕ by PedroIves

If this saved you time, drop a ⭐