parallel-work-coordinator

name	parallel-work-coordinator
description	Run independent work streams concurrently while preventing context drift, duplicated effort, and write conflicts. Use when several fragments have no blocking dependencies.
license	MIT
metadata	{"category":"agent-orchestration","origin":"custom-synthesis","revision":"1.1.0"}

Parallel Work Coordinator

Trigger

• Research, review, implementation, and documentation can overlap.
• Latency can be reduced safely through parallelism.

Inputs

• Task graph
• Ownership matrix
• Shared artifact map
• Merge strategy

Procedure

1. Group tasks by dependency wave.
2. Identify shared files, assumptions, schemas, and terminology.
3. Create read-only snapshots or isolated branches/worktrees for conflicting writes.
4. Define checkpoints for assumptions that affect multiple streams.
5. Broadcast only material changes to dependent agents.
6. Cancel or re-scope streams invalidated by new evidence.
7. Hand all completed outputs to result integration with provenance.

Output contract

• Execution-wave plan
• Conflict-avoidance rules
• Checkpoint log
• Completed artifacts with provenance

Quality gates

• Only truly independent tasks run together.
• Shared assumptions are versioned.
• No concurrent uncoordinated writes to the same source of truth.
• Parallelism never bypasses validation.

Handoffs and dependencies

• dependency-aware-scheduler
• result-integrator
• escalation-and-recovery

Boundaries

• Do not parallelize tightly coupled design decisions.
• Do not conceal divergent conclusions; preserve them for adjudication.