Provides an end-to-end orchestration harness for OpenClaw agents using a Plan→Work→Review cycle. The harness automates agent spawning, stage transitions, and a bridge component that tracks agent state and emits real-time channel notifications. It also contains model-routing guidelines so different tasks can be routed to GLM/GPT/Claude variants (including GLM-5.1) based on language and task complexity.
Use this skill when you need structured multi-agent workflows: breaking work into planning, execution, and review phases with automated notifications and failure handling. Ideal for CI-style automation, multi-step coding tasks, code review pipelines, or orchestrated audits where state tracking and escalation are required.
Best fit for OpenClaw runtime agents that support sessions_spawn and model selection (GLM/Claude/GPT families).
OpenClaw Harness is a comprehensive multi-agent orchestration skill with Plan→Work→Review pipeline, bridge notifications, account pool management with OAuth round-robin, and model routing. SKILL.md is detailed (in Korean) with clear agent definitions and routing tables. Scripts are well-structured CLI tools with proper argument handling, but all 10 failed in dry-run isolation due to missing config/directory dependencies — this is expected since they need a full harness setup. The account-pool and bridge subsystems are sophisticated with cooldown, fanout, and gap-detection logic.
Large skill (10 scripts) covering orchestration, notifications, and account pooling. Well-designed P1-P4 principles. Gap-detection loop with automatic feedback is innovative. Scripts handle missing python3 gracefully. The Korean-only documentation limits accessibility significantly.