Nobrainer Team Builder

NoBrainer Team Builder

Dynamically assemble a team of expert subagents for any task from a catalog of 401 agents across 27 categories. Agents are spawned on-demand — nothing loaded into context until needed.

Trigger: "nbteam", "team builder", "build team", "nobrainer team", "use nbteam", "compose team", "agent team", "assemble team"

Agent source: All 401 agent definitions are sourced from davila7/claude-code-templates and ship with this skill — no setup or download needed.

How It Works

This skill does NOT copy agents into the project. Instead, it:

1. Reads the task/request from the user
2. Identifies relevant categories → loads only those category JSONs from categories/
3. Picks best agents from loaded categories
4. Reads full agent .md only for selected agents from agent-catalog/
5. Spawns each as a temporary subagent via Agent tool

Zero permanent context cost. Only relevant categories are read, not all 401 agents.

Directory Structure

~/.claude/skills/nobrainer-team-builder/
├── skill.md                          # This file
├── agent-catalog.json                # Full index (name → description + category)
├── categories/                       # Per-category JSON indexes
│   ├── ai-specialists.json           #   6 agents
│   ├── api-graphql.json              #   8 agents
│   ├── blockchain-web3.json          #   4 agents
│   ├── business-marketing.json       #  18 agents
│   ├── data-ai.json                  #  38 agents
│   ├── database.json                 #  11 agents
│   ├── deep-research-team.json       #  16 agents
│   ├── development-team.json         #  16 agents
│   ├── development-tools.json        #  30 agents
│   ├── devops-infrastructure.json    #  37 agents
│   ├── documentation.json            #  11 agents
│   ├── expert-advisors.json          #  52 agents
│   ├── ffmpeg-clip-team.json         #   8 agents
│   ├── finance.json                  #   5 agents
│   ├── game-development.json         #   5 agents
│   ├── git.json                      #   3 agents
│   ├── mcp-dev-team.json             #   8 agents
│   ├── modernization.json            #   3 agents
│   ├── obsidian-ops-team.json        #   7 agents
│   ├── ocr-extraction-team.json      #   7 agents
│   ├── performance-testing.json      #   5 agents
│   ├── podcast-creator-team.json     #  11 agents
│   ├── programming-languages.json    #  49 agents
│   ├── realtime.json                 #   2 agents
│   ├── security.json                 #  20 agents
│   ├── ui-analysis.json              #   5 agents
│   └── web-tools.json                #  16 agents
└── agent-catalog/                    # Full agent definitions (401 .md files)
    ├── ai-specialists/               #   6 agents
    │   ├── ai-ethics-advisor.md
    │   └── ...
    ├── programming-languages/        #  49 agents
    │   ├── python-pro.md
    │   ├── javascript-pro.md
    │   └── ...
    ├── security/                     #  20 agents
    │   ├── security-auditor.md
    │   └── ...
    └── ... (27 category folders total)

Workflow

Step 1: Understand the Task

Read the user's request. Identify:

• What needs to be done (code review? architecture? debugging? deployment?)
• Technologies involved (Python? React? Supabase? Terraform?)
• Complexity (single agent sufficient, or multi-agent collaboration needed?)

Step 2: Select Categories

Based on the task, pick 2-5 relevant categories and read only those JSONs:

| Task Type | Categories to Load | |-----------|-------------------| | Code review / refactor | development-tools, programming-languages, security | | New feature | development-team, programming-languages, database, development-tools | | Architecture design | expert-advisors, development-team, devops-infrastructure | | Security audit | security, development-tools, devops-infrastructure | | Database work | database, data-ai, performance-testing | | Frontend / UI | web-tools, development-team, ui-analysis, performance-testing | | DevOps / Deploy | devops-infrastructure, git, security | | Research / Analysis | deep-research-team, data-ai, documentation | | API design | api-graphql, development-team, documentation | | MCP development | mcp-dev-team, development-tools | | Marketing / Business | business-marketing, documentation | | Video / Audio | ffmpeg-clip-team, media | | Blockchain / Web3 | blockchain-web3, security, finance | | Game dev | game-development, programming-languages | | Obsidian | obsidian-ops-team | | OCR / Documents | ocr-extraction-team, document-processing |

Read each category JSON like:

Read ~/.claude/skills/nobrainer-team-builder/categories/programming-languages.json

Each file contains: { "agent-name": "description", ... }

Step 3: Select Agents (max 10)

From loaded categories, pick best 3-10 agents:

• Task match: Agent's description directly addresses the task
• Tech match: Agent knows the specific technologies
• Complementary coverage: Don't pick 3 code reviewers
• Stop when sufficient: Don't add agents that won't contribute

Team sizes:

• Simple (review, debug): 3-5 agents
• Medium (feature, refactor, audit): 5-8 agents
• Complex (architecture, migration): 8-10 agents

Step 4: Present the Team

Before spawning, show:

NBTeam for: "{task summary}"
Categories loaded: programming-languages, development-tools, security

| # | Agent | Category | Role in This Task |
|---|-------|----------|-------------------|
| 1 | python-pro | programming-languages | Core implementation |
| 2 | test-engineer | development-tools | Test coverage |
| 3 | security-auditor | security | Security review |

Spawn all? (or adjust)

Step 5: Spawn Agents

For each selected agent:

1. Read full .md from ~/.claude/skills/nobrainer-team-builder/agent-catalog/{category}/{name}.md
2. Spawn via Agent tool:

• name: nbteam-{agent-name}
• prompt: Agent persona + task + project context
• description: Short task summary

1. Parallel for independent agents, sequential for dependent ones

Agent prompt template:

You are {agent-name}. Here is your persona and expertise:

---
{full content of agent .md file}
---

Your task: {user's specific request}

Context about the project:
{relevant project context — CLAUDE.md, file structure, etc.}

Step 6: Synthesize Results

• Collect all agent outputs
• Highlight agreements and conflicts
• Present unified recommendation
• If agents disagree, explain tradeoffs

Categories Reference (27 categories, 401 agents)

| Category | Count | Focus | |----------|-------|-------| | ai-specialists | 6 | AI/ML strategy, ethics, prompts | | api-graphql | 8 | API design, GraphQL, REST | | blockchain-web3 | 4 | Smart contracts, Web3 | | business-marketing | 18 | Product, sales, legal, analytics | | data-ai | 38 | ML, data science, NLP, CV, quant | | database | 11 | SQL, NoSQL, Supabase, Neon, optimization | | deep-research-team | 16 | Research, analysis, fact-checking | | development-team | 16 | Full-stack, frontend, backend, mobile | | development-tools | 30 | Testing, debugging, code review, profiling | | devops-infrastructure | 37 | CI/CD, cloud, Kubernetes, Terraform, monitoring | | documentation | 11 | Technical writing, API docs, diagrams | | expert-advisors | 52 | Architecture review, dependency management | | ffmpeg-clip-team | 8 | Video/audio processing | | finance | 5 | Fintech, trading, payments | | game-development | 5 | Unity, Unreal, game design | | git | 3 | Git workflow, branching strategies | | mcp-dev-team | 8 | MCP server development | | modernization | 3 | Legacy migration, cloud modernization | | obsidian-ops-team | 7 | Obsidian vault management | | ocr-extraction-team | 7 | OCR, document analysis | | performance-testing | 5 | Load testing, web vitals | | podcast-creator-team | 11 | Podcast production | | programming-languages | 49 | Python, JS, Rust, Go, C#, PHP, etc. | | realtime | 2 | WebSockets, Supabase Realtime | | security | 20 | Pentesting, compliance, incident response | | ui-analysis | 5 | Screenshot analysis, UI review | | web-tools | 16 | React, Next.js, SEO, accessibility |

Rules

• Max 10 agents per task — diminishing returns beyond that
• Load only relevant category JSONs — never read all 27 at once
• Read agent .md only when spawning — don't pre-read all 401 files
• Parallel by default — spawn independent agents in parallel
• Present team first — always show before spawning, accept adjustments
• No permanent installation — agents are temporary subagents, they don't persist

Source & license

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

• Author: nobrainer-tech
• Source: nobrainer-tech/nobrainer-claude-skills
• License: MIT

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