Coordinator Resume Integrity — Production Agent Handoff Logic

Coordinator-worker patterns fail at the seam. The coordinator dispatches a worker perfectly. The worker completes. The coordinator resumes — and the mode has flipped, the state is stale, the task context is wrong, or the coordinator thinks it is mid-task when it should be evaluating results. This SKILL.md is the architecture that closes that seam.

---

⚡ What's Inside

• Mode consistency check The pre-resume validation that confirms the coordinator is in the correct operational mode before processing any worker return. Catches mode mismatch before it produces bad output

• State serialization requirements Exactly what must be written to disk before a worker is dispatched, and why in-memory state is insufficient. The list is specific and non-obvious

• Handoff receipt protocol Structured payload the worker returns to the coordinator, including task ID, completion status, artifact paths, and the checksum that confirms the right result is being processed

• Resume sequence The ordered steps the coordinator must execute on worker return before doing anything else. Each step has an explicit pass/fail gate

• Worker spawning rules The conditions under which a coordinator is permitted to spawn additional workers, and the constraint that prevents cascading spawns that exhaust the session budget

---

🏭 Proven in Production

Every element of this architecture was built in response to a real coordinator failure. The mode mismatch check came from a coordinator that resumed in dispatch mode instead of evaluation mode and re-spawned the same worker task. The state serialization requirement came from a WSL2 restart that lost everything the coordinator held in memory mid-task.

---

🆕 What's New in v1.1

• Multi-worker coordination patterns added
• Handoff receipt schema updated with artifact checksum field

---

✅ Core Capabilities

• ✅ Pre-resume mode consistency check — catches mode mismatch before it executes
• ✅ State serialization spec — exactly what must be on disk before worker dispatch
• ✅ Handoff receipt protocol — structured worker return payload with integrity check
• ✅ Resume sequence — ordered steps with pass/fail gates at each stage
• ✅ Worker spawning constraints — prevents cascading spawns and budget exhaustion
• ✅ Drop-in compatible with any OpenClaw coordinator-worker topology