Purpose
Long sessions (6+ hours) accumulate massive context, degrading response quality and increasing cost.
Solution
Add recommendation to parallel-orchestrator and taskMaestro docs:
- Compact after every 2 waves
- Compact before starting a new batch of waves
- Save critical state to context.md before compact
References