Summary: Add an opt-in compiler/export layer that turns supported Workflow and CompiledExecution graphs into equivalent artifacts for external agent frameworks such as LangGraph and smolagents.
Why:
- Author study workflows once in
design_research_agents.workflow, then run the same logic in other agent ecosystems.
- Make interoperability and cross-framework benchmarking easier without hand-rewriting nodes, edges, and state handoff rules.
- Reuse framework-specific tooling and runtime surfaces while keeping this repo's workflow-first API as the source of truth.
Tasks:
- Define a small public compiler/adapter surface for exporting supported workflow graphs.
- Start with a clearly documented supported subset of deterministic workflow semantics and reject unsupported features with clear validation errors.
- Implement a LangGraph compiler that maps workflow steps and dependencies into a
StateGraph or equivalent runnable.
- Implement a smolagents compiler/adapter for the subset that can be represented there, even if that requires a wrapper around framework-native agents/tools.
- Preserve enough metadata to keep
ExecutionResult provenance, step IDs, and tracing correlation intelligible after compilation.
- Add parity tests and examples showing the same internal workflow compiled into both frameworks.
Acceptance Criteria:
- A user can export a supported workflow into LangGraph and smolagents from a documented public entry point.
- Unsupported step types or control-flow semantics fail fast with explicit validation messages.
- Examples and docs show the supported subset, dependency extras, and known behavior differences.
- Tests cover equivalence for at least one linear workflow and one branched workflow within the supported subset.
Summary: Add an opt-in compiler/export layer that turns supported
WorkflowandCompiledExecutiongraphs into equivalent artifacts for external agent frameworks such as LangGraph and smolagents.Why:
design_research_agents.workflow, then run the same logic in other agent ecosystems.Tasks:
StateGraphor equivalent runnable.ExecutionResultprovenance, step IDs, and tracing correlation intelligible after compilation.Acceptance Criteria: