long-run-implement

Long-run plan execution preserves quality by spawning a fresh subagent for each plan. This approach isolates tasks, maintains deterministic outcomes, and prevents context degradation across multiple implementation phases.

Autonomous task orchestration is managed through a phase-driven workflow covering initialization, plan transformation, execution, and finalization. The system handles deviations automatically while ensuring per-task commits and state persistence for resumable runs.

State persistence during plan execution is achieved through atomic checkpointed progress. Per-task commits are generated automatically, allowing runs to resume from the last saved state if execution is interrupted or deviates from the original spec.

Yes, multi-phase implementation workflows support atomic commits and deviation handling. The system guides per-plan execution, generates checkpoints, and automatically addresses deviations to ensure deterministic task outcomes across the initialization to finalization phases.

Yes, you need to provide a spec to trigger the long-run plan execution. The skill transforms the provided specification into executable steps, then spawns dedicated fresh subagents to carry out per-plan execution with automatic summaries and checkpoints.

You should avoid fresh subagent execution when tasks require shared mutable state across plans without checkpointing. This approach relies on isolated contexts and atomic commits, making it unsuitable for workflows needing continuous memory or non-deterministic state transitions.