Launch parallel subagent tasks safely with merge plan

Parallel Agents

Purpose

Run 2-3 independent subagent lanes in parallel while avoiding state collisions.

Instructions

1. Validate no write overlap across subtasks.
2. Create independent prompts with strict ownership boundaries.
3. Launch parallel execution in one orchestration step.
4. Collect outputs and run deterministic merge order.
5. Return consolidated result with lane provenance.

Constraints

Do not parallelize tasks touching same mutable file set. Do not merge unvalidated out