---

name: speculator-orchestration description: Multi-agent task orchestration using the spec CLI. Guides agents on claiming tasks, parallel execution, blocking, and coordination. Use when working with multiple agents, claiming tasks, or managing parallel work.

Speculator Orchestration

Core Principle

Claim before you code.

Never start work on a task without atomically claiming it first. This prevents duplicate effort and merge conflicts in multi-agent scenarios.

Quick Reference

Workflows

• Claiming Tasks - How to find and claim work
• Completing Tasks - When and how to mark work done
• Blocking Tasks - When and how to record blockers

Patterns

• Parallelization - Which tasks can run concurrently
• Decomposition - Breaking large tasks into subtasks

Examples

• Multi-Agent Session - Two agents working concurrently

The Orchestration Loop

Copy1. spec tasks ready → find available work
2. Evaluate conflicts → check for file overlap with other in-progress tasks
3. spec task claim <id> → atomically claim
4. Implement → do the work
5. spec task complete <id> | spec task block <id> --reason <reason> → finish or record blocker
6. Repeat

Important Notes

• Always check spec tasks ready before starting work
• If claim fails, another agent got it first - move on
• Block tasks with clear, actionable reasons
• Release tasks if you can't complete them
• Create subtasks for large pieces of work