streaming-cast

# Streaming {{ cookiecutter.act_name }}'s Casts (v3) Implement v3 event streaming to consume `{{ cookiecutter.cast_snake }}_graph()` output in runtime, API endpoints, or other consumers. Event streaming returns a run stream object with typed projections (`stream.messages`, `stream.values`, `stream.subgraphs`, `stream.output`, `stream.tool_calls`, ...) for independent consumption. ## When to Use - Adding streaming output to a runtime or API endpoint - Need token-by-token LLM output, tool call lifecycle, or final output streaming - Custom progress events from nodes via stream writer (with custom transformers) - Subagent/subgraph projection for source identification - Transport integration (SSE recommended, WebSocket optional) ## When NOT to Use - Building graph structure (nodes, edges, state) → `developing-cast` - DeepAgent harness (create_deep_agent, backends) → `developing-deepagent` - Architecture design → `architecting-act` - Testing → `testing-cast` --- ## Quick Start `casts/{{ cookiecutter.cast_snake }}/modules/` and `casts/{{ cookiecutter.cast_snake }}/graph.py` are reserved for graph definition. **Stream consumer code lives anywhere else** — pick the entry point that fits the project (an additional module within the cast such as `runtime.py`, an external API endpoint module, a script, or a test). ```python # stream consumer — location flexible from langchain_core.messages import HumanMessage from casts.{{ cookiecutter.cast_snake }}.graph import {{ cookiecutter...