Problem

MeetingProtocolConfig.auto_create_tasks defaults to True. When enabled, MeetingOrchestrator._create_tasks() spawns Task objects from action items extracted from meeting minutes. If task execution emits events that match MeetingScheduler.trigger_event() patterns, new meetings are triggered.

MeetingScheduler.trigger_event() has no deduplication or cooldown -- any matching event fires the meeting immediately:

matching = tuple(mt for mt in self._config.types if mt.trigger == event_name)
async with asyncio.TaskGroup() as tg:
    tasks = [tg.create_task(self._execute_meeting(mt, context)) for mt in matching]

This creates a potential unbounded loop:

1. Sprint planning generates action items -> tasks created
2. Task completion emits events -> post-sprint meeting triggered
3. Post-sprint generates more action items -> more tasks
4. Repeat until budget hard stop

The only backstop today is the budget enforcer, which is blunt and expensive to hit.

Source: docs/research/multi-agent-failure-audit.md (R1), closes research #690.

Solution

Add min_interval_seconds: int | None = None to MeetingTypeConfig (in communication/meeting/config.py). Track last-triggered timestamps in MeetingScheduler per meeting type. In trigger_event(), skip execution if the meeting was triggered within min_interval_seconds.

Alternatively (or additionally): add max_tasks_per_meeting: int | None = None to MeetingProtocolConfig to cap how many tasks a single meeting can spawn.

Files

• src/synthorg/communication/meeting/config.py -- add min_interval_seconds field
• src/synthorg/communication/meeting/scheduler.py -- enforce cooldown in trigger_event()