phase-runner

Multi-agent orchestration prevents context window exhaustion by enforcing token-aware tool budgets and triggering graceful handoffs to sub-agents. This ensures long-running task phases maintain progress and recoverability across multiple sessions without exceeding limits.

Phased task execution for large plans is best handled through a four-file structure—context, role, tasks, and execution-plan—driving deterministic wave-based execution. This method enforces tool budgets and requires real-time task syncing with timestamps to track progress.

Yes, multi-session orchestration supports recoverability via graceful handoffs when token limits are reached. Sub-agents update phase files with completion timestamps in real-time, allowing the next wave to resume execution seamlessly from the last saved state.

Yes, wave-based automation requires a directory containing a four-file phase set: context, role, tasks, and execution-plan. Activating the orchestrator on this directory begins the deterministic execution of task waves with automatic token budget enforcement.

Real-time updates with timestamps are required to ensure progress tracking and recoverability across multiple sub-agents. This file syncing mechanism allows subsequent waves or handoff agents to verify completed tasks and resume the orchestration without losing context.

Token-aware budget enforcement limits the number of tool calls per phase to prevent context exhaustion, meaning very long task lists must be divided into multiple waves. If phase files are not updated with timestamps, subsequent sub-agents may lose orchestration state.