---

name: agent-core description: "Use AtomicAgent, AgentConfig, BaseIOSchema, ChatHistory, SystemPromptGenerator, hooks, token counting, and multimodal content to build typed Atomic Agents." disable-model-invocation: true metadata: disco-role: operating license: MIT

Agent Core

Use this subskill when a task is about building or debugging the core Atomic Agents framework API: typed schemas, agent construction, history, prompts, streaming, async calls, hooks, multimodal content, or token counting.

Read first

• references/api-reference.md for the public classes, signatures, and object relationships.
• references/workflows.md for quickstart, memory, streaming, and context-provider recipes.
• references/troubleshooting.md for docstring, client-wrapper, history, role, and trimming failures.
• scripts/agent_core_smoke.py for a safe offline smoke check.

Owns

• AtomicAgent and AgentConfig construction.
• BaseIOSchema contracts and docstring requirements.
• ChatHistory, BaseChatHistory, SystemPromptGenerator, and BaseDynamicContextProvider.
• Sync, async, streaming, and async-streaming agent execution.
• Hooks, token counting, context trimming, and multimodal history serialization.
• VideoURL and other core content-shaping helpers that belong to the agent layer.

Does not own

• BaseTool, BaseResource, BasePrompt, Atomic Forge, or the atomic CLI; use ../tooling-and-forge/SKILL.md.
• MCP transport and dynamic tool discovery; use ../mcp-integrations/SKILL.md.
• Example project cataloging; use ../example-workflows/SKILL.md.
• Repo maintenance, tests, docs, or release commands; use ../repo-development/SKILL.md.

Common triggers

• "How do I create a typed agent?"
• "Why does my schema need a docstring?"
• "How do I keep chat history across turns or custom backends?"
• "Why are token counts or context trimming behaving oddly?"
• "How do I add a context provider or hook?"
• "How do I handle multimodal content in history?"
• "Why does the agent need different sync/async methods?"