moo.agent.response_model

Pydantic response model for one LLM cycle. PydanticAI validates the model’s reply against AgentResponse; a failed validation is re-asked automatically. See docs/source/explanation/agent-internals.md (Structured Responses).

Stage-2: tool calls are dispatched via PydanticAI’s native tool loop, so the old actions field, per-tool ActionBase subclasses, and ToolName Literal are gone. AgentResponse now carries only meta-state (goal, plan, done, soul_patches, build_plan) — actions live in the tool-call channel.

Classes

AgentResponse(**data)	The single validated meta-response shape for one LLM cycle.
SoulPatch(**data)	A learned-knowledge patch appended to SOUL.patch.md (was SOUL_PATCH_*).