Auto-discover activities from workflow classes (remove need for collect_activities)

[bdsaglam]

Problem

Currently, users must explicitly collect and pass activities to the worker:

# Current approach - verbose and error-prone
activities = collect_activities(
    temporal_multitool,
    calculator_tool,
    search_web_tool,
    get_current_time_tool,
)

async with Worker(
    client,
    task_queue="queue",
    workflows=[MultiToolAgentWorkflow],
    activities=activities,  # Must manually list all activities
    ...
)

This is:

• Error-prone: Easy to forget a tool or module
• Verbose: Duplicates information already in the workflow
• Not DRY: Workflow knows its modules, why list them again?

Pydantic AI provides a plugin (AgentPlugin) that automatically collects and passes activities to the worker.

from pydantic_ai.durable_exec.temporal import AgentPlugin

agent = dspy.ReAct(
    "question -> answer",
    tools=[evaluate_math, search_web_tool, get_current_time_tool],
)
temporal_agent = TemporalModule(agent, name="temporal_agent")

async with Worker(
    client,
    task_queue="queue",
    workflows=[MultiToolAgentWorkflow], # Some workflows using DSPy programs
    plugins=[AgentPlugin(temporal_agent)],
    ...
)

Benefits

1. Less boilerplate: No collect_activities() calls

Priority

High (Developer Experience)

References

• Pydantic AI implementation: pydantic_ai/durable_exec/temporal/__init__.py:97-120
• Current example: examples/multitool_agent.py:230-235

[bdsaglam]

@claude implement this