skills.rest
BrainBlend-AI
BrainBlend-AICOMMUNITY
Explore all skills by this author →
GitHub

create-atomic-agent

Builds and wires AtomicAgent instances with schemas, provider clients, and system prompts.

6.2k|536|Updated Jun 3, 2024
Search Tags:#Pydantic#Python#lLM#instructor#Agent#Structured Output#atomic-agents
One-click install
npx skills add https://github.com/BrainBlend-AI/atomic-agents --skill create-atomic-agent
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: create-atomic-agent
Source: https://github.com/BrainBlend-AI/atomic-agents/tree/main/claude-plugin/atomic-agents/skills/create-atomic-agent
Command: npx skills add https://github.com/BrainBlend-AI/atomic-agents --skill create-atomic-agent

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires atomic-agents, instructor.

What problem does it solve?

Creating a new agent in the Atomic Agents framework requires coordinating schemas, provider clients, Instructor modes, system prompts, and history configuration, and small mistakes like a missing max_tokens or wrong mode silently break every call. This Skill walks through the full agent-authoring workflow so the resulting AtomicAgent works on the first run.

Core Features & Use Cases

  • Guided agent authoring: Clarifies purpose, input/output schemas, provider, conversational state, and context providers before writing any code.
  • Provider wiring matrix: Supplies correct Instructor-wrapped client setup for OpenAI, Anthropic, Gemini, Groq, Ollama, and MiniMax, including mode and model_api_parameters gotchas.
  • Verification and handoff: Includes a smoke-test command and guidance on run, run_async, run_stream, hooks, and follow-up skills for tools, schemas, and context providers.
  • Use Case: A developer asks to add a ticket-classification router agent; the Skill produces the typed AtomicAgent[Input, Output] with a SystemPromptGenerator, Anthropic client with max_tokens set, and a passing smoke test.

Quick Start

Ask the assistant to create an Atomic Agents agent that classifies incoming support tickets by category and urgency using your preferred LLM provider.

Frequently Asked Questions about create-atomic-agent

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I create an agent with the Atomic Agents framework?▼

Define input and output schemas extending BaseIOSchema, wrap your provider client with an instructor.from_* factory, then instantiate AtomicAgent[In, Out] with an AgentConfig containing the client, model, and a SystemPromptGenerator. Add ChatHistory only if the agent is conversational.

How do I configure Anthropic or Gemini providers in Atomic Agents?▼

Anthropic requires Mode.TOOLS and max_tokens inside model_api_parameters or every call fails. Gemini requires Mode.GENAI_TOOLS in the instructor factory and assistant_role set to "model" in AgentConfig to avoid role mismatch errors.

Why does my Atomic Agents structured output stop working?▼

The most common cause is forgetting to wrap the client with instructor.from_openai or the equivalent factory, which silently disables structured outputs. Also check that AgentConfig mode matches the Instructor factory mode and that schemas extend BaseIOSchema, not plain BaseModel.

Should I use ChatHistory for a single-shot agent?▼

No, omit ChatHistory from AgentConfig for stateless single-shot transformers like extractors or classifiers. Include it only for conversational agents, and monitor token utilization or set max_messages in long-running services.

When should I use custom schemas instead of BasicChatInputSchema?▼

Use BasicChatInputSchema and BasicChatOutputSchema for free-form chat agents. For structured tasks like extraction, classification, planning, or routing, define a custom BaseIOSchema pair so the generics AtomicAgent[MyInput, MyOutput] carry the typed contract.

Related Skills

Explore More Software Engineering Skills→
AgentientAgentient
2

mcp-pydantic-tool-definition

Define MCP tool schemas with Pydantic models and strict validation.

Official
Advanced
ai-sand-castlesai-sand-castles

team-orchestration

Coordinate multi-agent team definitions, spawning, and lifecycle management.

Official
Advanced
jaydubya818jaydubya818

context-fundamentals

Explains context management fundamentals for AI agent systems.

Community
Intermediate
dformosodformoso

test-driven-dev

Implement test-first development for user-facing behavior with integration and end-to-end tests.

Community
Advanced