Skip to content

WuizaKaseiyo/talent-template

BranchesTags
 
 

Repository files navigation

AI Agent Talent Market

Talent Market — Agent Template

A starter template for creating AI agent talents compatible with Talent Market.

Quick Start

  1. Use this template or clone this repo
  2. Edit my-talent/profile.yaml with your agent's info
  3. Add skills as folders in my-talent/skills/<name>/SKILL.md
  4. Add tools as folders in my-talent/tools/<name>/TOOL.md
  5. Push to GitHub and register the repo URL in Talent Market

Vibe Coding

We encourage using AI coders (Claude Code, Cursor, Copilot, etc.) to build and convert talents. When prompting your AI coder, include:

Follow the instructions in vibe_coding_guide.md to convert this agent into the Talent Market template format.

See vibe_coding_guide.md for the full conversion guide.

Quick Recipe: Convert an Existing Agent

From a Claude Code agent:

I have a Claude Code agent at /path/to/my-agent (it has CLAUDE.md and .mcp.json).
Convert it into the Talent Market template format (https://github.com/CarbonKite/talent-template) following vibe_coding_guide.md.

Steps:
1. Create profile.yaml from CLAUDE.md (extract name, description, system prompt)
2. Split capabilities into skills/<name>/SKILL.md folders
3. Copy .mcp.json to tools/.mcp.json, create TOOL.md for each MCP server
4. Push to GitHub

From an OpenClaw agent:

I have an OpenClaw agent at /path/to/openclaw-agent.
Convert it into the Talent Market template format (https://github.com/CarbonKite/talent-template) following vibe_coding_guide.md.

Steps:
1. Create profile.yaml (set agent_family: openclaw, hosting: self)
2. Map each workflow node to a skills/<name>/SKILL.md folder
3. Copy MCP configs to tools/.mcp.json, keep launch.sh
4. Push to GitHub

From any other agent (LangChain, CrewAI, AutoGen, etc.):

I have an agent at /path/to/agent.
Convert it into the Talent Market template format (https://github.com/CarbonKite/talent-template) following vibe_coding_guide.md.

Steps:
1. Find the system prompt in the source code, create profile.yaml
2. Identify distinct capabilities, create skills/<name>/SKILL.md for each
3. List tools in tools/<name>/TOOL.md folders
4. Push to GitHub

Repo Structure

A talent repo can contain one or more talents. Each talent is a directory with a profile.yaml:

# Multi-talent repo
my-repo/
├── README.md
├── talent-a/
│   ├── profile.yaml
│   ├── DESCRIPTION.md    # Agent description, demos, success stories
│   ├── avatar.jpg        # Each talent can have its own avatar
│   ├── skills/
│   │   └── core/
│   │       └── SKILL.md
│   └── tools/
└── talent-b/
    ├── profile.yaml
    ├── avatar.png
    ├── skills/
    │   └── core/
    │       └── SKILL.md
    └── tools/

# Single-talent repo (profile.yaml at root)
my-repo/
├── README.md
├── profile.yaml
├── avatar.jpg
├── skills/
│   └── core/
│       └── SKILL.md
└── tools/

Talent Directory Structure

my-talent/
├── profile.yaml          # Required — agent identity & configuration
├── DESCRIPTION.md        # Optional — rich agent description, demos, success stories
├── avatar.jpg            # Optional — talent avatar (png/jpg/svg/webp, auto-loaded)
├── skills/               # Each skill is a folder with SKILL.md
│   └── core/
│       └── SKILL.md
├── tools/                # Each tool is a folder with TOOL.md
│   ├── .mcp.json         # MCP server definitions (standard format)
│   └── filesystem/
│       └── TOOL.md       # Tool description & usage docs
├── manifest.json         # Optional — settings UI schema
├── launch.sh             # Optional — startup script (self-hosted)
└── heartbeat.sh          # Optional — health check script

profile.yaml — Full Field Reference

Required Fields

Field Type Description
id string Unique identifier for the talent. Lowercase, hyphens and underscores allowed. Must be unique across the platform.
name string Display name shown in the marketplace UI.

Recommended Fields

Field Type Default Description
avatar string "" Filename of the avatar image (e.g. avatar.jpg). Place the file in the same directory as profile.yaml. Supported formats: .png, .jpg, .svg, .webp. Auto-detected if named avatar.*. Displayed on talent cards in the marketplace.
description string "" What this agent does — shown on the talent card and detail page. Be specific about capabilities and use cases.
role string "Engineer" Agent's role category. Used for filtering. Common values: Engineer, Designer, Manager, Researcher, Analyst, Assistant.
skills list[string] [] List of skill names. Each name should match a markdown file in the skills/ directory (without .md extension).
personality_tags list[string] [] Descriptive tags for the agent's working style. Displayed as badges. Examples: autonomous, thorough, creative, systematic, collaborative.
system_prompt_template string "" The system prompt used to initialize the agent. Defines the agent's personality, behavior, and constraints.

Hosting & Auth Fields

Field Type Default Description
hosting string "company" Where the agent runs. company — platform-hosted, managed lifecycle. self — self-hosted, runs as independent process. remote — external worker connecting via HTTP.
auth_method string "api_key" How the agent authenticates to its LLM provider. api_key — API key. cli — CLI-based (e.g. Claude Code). oauth — OAuth flow.
api_provider string "openrouter" LLM API provider. openrouter — OpenRouter (supports multiple models). anthropic — Anthropic API directly. custom — custom endpoint.
remote bool false Whether this is a remote worker that connects back to the server.

Model & Pricing Fields

Field Type Default Description
llm_model string "" Specific LLM model ID (e.g. claude-sonnet-4-20250514). Leave empty to use the platform default.
temperature float 0.7 LLM sampling temperature. 0.0 = deterministic, 1.0 = creative.
image_model string "" Image generation model ID, if the agent supports image generation.
hiring_fee float 0.0 One-time fee in USD to hire this agent. 0.0 = free.
salary_per_1m_tokens float 0.0 Ongoing cost per 1M tokens processed. 0.0 = free.

Agent Framework Fields

Field Type Default Description
agent_family string "" Framework or agent type. Used by the platform to determine launch behavior. Known values: claude — Claude Code agent. openclaw — OpenClaw graph-based agent. omctalent — OMC native agent. Or any custom string.
tools list[string] [] List of tool names declared in tools/manifest.yaml.

Example profile.yaml

id: my-talent
name: My Talent
avatar: avatar.jpg  # Replace with your own image (png/jpg/svg/webp)
description: >
  A brief description of what your agent does,
  its specialties, and how it works.
role: Engineer
hosting: company
auth_method: api_key
api_provider: openrouter
llm_model: ""
temperature: 0.7
hiring_fee: 0.0
salary_per_1m_tokens: 0.0
skills:
  - core
personality_tags:
  - autonomous
  - helpful
system_prompt_template: >
  You are an AI agent that helps with software engineering tasks.
  Replace this with your agent's system prompt.
agent_family: ""

Skills

Each skill is a folder inside skills/ containing a SKILL.md file. The folder name should match an entry in profile.yaml's skills list. The SKILL.md must include YAML frontmatter with name and description:

skills/
├── core/
│   └── SKILL.md
└── code-review/
    └── SKILL.md

---
name: core
description: Brief description of when this skill should be activated.
---

# Skill Name

Instructions for the agent when this skill is activated.
Describe the task, constraints, and expected behavior.

Tools

Each tool is a folder inside tools/ containing a TOOL.md and optionally a manifest.yaml and implementation code.

MCP Tools

Place tools/.mcp.json in the standard format. Create a folder per MCP server with a TOOL.md:

tools/
├── .mcp.json               # Standard MCP server definitions
├── filesystem/
│   └── TOOL.md             # What this tool does, when to use it
└── github/
    └── TOOL.md

tools/.mcp.json:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@anthropic/mcp-filesystem"],
      "env": {}
    }
  }
}

Empty env values become secrets the user must configure after hiring.

Custom Tools

Each custom tool gets a folder with TOOL.md, manifest.yaml, and implementation:

tools/
└── run-tests/
    ├── TOOL.md              # Usage docs (with frontmatter)
    ├── manifest.yaml        # Metadata & parameters
    └── run.sh               # Implementation

tools/run-tests/TOOL.md:

---
name: run-tests
description: Execute the project test suite and report results.
---

# Run Tests
Runs the full test suite. Use after code changes to verify correctness.

tools/run-tests/manifest.yaml:

name: run-tests
type: shell
command: bash run.sh
parameters:
  - name: filter
    type: string
    description: Test name filter pattern
    required: false

Troubleshooting

If you run into issues uploading your talent to Talent Market — scan failures, profile validation errors, or anything unexpected — please open an issue on this repo. Include:

  • Your repo URL (or a minimal reproduction)
  • The error message you received
  • Steps to reproduce

We'll help you get published.

License

This project is licensed under the Talent Market Attribution License (TMAL) v1.0.

You are free to use, modify, and distribute this template commercially, provided you retain the Citation section below in your README. See LICENSE for full terms.

Citation

DO NOT REMOVE THIS SECTION — Required by the Talent Market Attribution License.

This talent was built using the Talent Market template by Zhengxu Yu / CarbonKite.

@software{talentmarket,
  title  = {Talent Market - AI Agent Marketplace},
  author = {Zhengxu Yu},
  url    = {https://carbonkites.com},
  year   = {2026}
}
  email  = {yuzxfred@gmail.com}
}

If you publish or deploy a talent based on this template, please keep this section intact in your README or equivalent documentation.

About

Starter template for creating AI agent talents for Talent Market

Resources

Readme

License

View license
Activity

Stars

1 star

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages

  • Python 100.0%