Harness
Run Plan→Work→Review agent loops with per-agent model routing, Korean→GLM auto-routing, and live channel bridge notifications in OpenClaw.
Overview
harness (ohmyclaw) is a journey-wide agent skill that orchestrates Plan→Work→Review agent runs with multi-model routing and channel bridges—usable whenever a solo builder needs governed agent execution before trusting ad
Install
npx skills add https://github.com/jkf87/ohmyclaw --skill SKILL.mdWhat is this skill?
- Plan→Work→Review orchestration loop modeled on Claude Code harness patterns
- Model routing across GLM (including GLM-5.1), GPT, and Claude with Korean detection→GLM auto-routing
- sessions_spawn to assign a different model per spawned agent
- Real-time channel notifications via bridge integration
- OpenClaw-focused harness informed by Claude Code harness ecosystem analysis
- 3-phase orchestration loop: Plan, Work, and Review
- 3 model families called out: GLM, GPT, and Claude with GLM-5.1 support
Adoption & trust: 60 GitHub stars.
What problem does it solve?
Your OpenClaw agents plan and code in one blobby session with no review gate, wrong model for the locale, and no live signal back to your team channel.
Who is it for?
OpenClaw users who want Claude-Code-style harness discipline, GLM/GPT/Claude mix-and-match, and Korean-aware routing without building orchestration from scratch.
Skip if: Builders on a single-model, single-session Cursor workflow who do not use OpenClaw, sessions_spawn, or external channel bridges.
When should I use this skill?
OpenClaw sessions need governed multi-phase agent runs, locale-aware GLM routing, or channel-visible progress.
What do I get? / Deliverables
You get separated plan, work, and review phases with routed models per agent and bridged updates so the next human or agent step starts from an audited checkpoint.
- Plan artifact from planning phase
- Work output from implementation agent
- Review summary and routed channel updates
Recommended Skills
Journey fit
Useful at every journey phase - explore requirements and options before committing to a direction.
Where it fits
Run a plan-only GLM pass on a Korean product brief before spawning a Claude worker for code.
Execute Plan→Work→Review on a feature branch with per-agent model assignments.
Dedicated review agent on GPT while implementation stayed on a faster coder model.
Route Korean drafts to GLM and English polish to Claude within one harnessed pipeline.
Bridge failure summaries from an incident work agent to your ops channel in real time.
How it compares
An orchestration harness skill, not a framework tutorial—compare to raw agent chat rather than to LangGraph library docs.
Common Questions / FAQ
Who is harness (ohmyclaw) for?
Solo and small-team builders on OpenClaw who need structured Plan→Work→Review loops, multi-model routing (including GLM-5.1), and realtime channel alerts during agent work.
When should I use harness (ohmyclaw)?
Before large Build features to force a plan artifact, during Ship for automated review passes, during Grow content pipelines that mix Korean and English, and during Operate when incident agents must notify a channel mid-run.
Is harness (ohmyclaw) safe to install?
It can spawn agents, route to external model APIs, and bridge channels—review the Security Audits panel on this page and limit secrets and network scopes in your OpenClaw config.
SKILL.md
READMESKILL.md - Harness
OpenClaw 하네스 — Plan→Work→Review 에이전트 오케스트레이션 + 모델 라우팅 + 채널 브릿지. Claude Code 하네스 생태계 분석 기반. GLM/GPT/Claude 모델 지원. GLM-5.1 포함. 한국어 감지→GLM 자동 라우팅. sessions_spawn으로 에이전트별 모델 별도 지정. 브릿지로 실시간 채널 알림. # harness { "name": "harness", "description": "OpenClaw 하네스 — Plan→Work→Review 에이전트 오케스트레이션 + 모델 라우팅 + 채널 브릿지. Claude Code 하네스 생태계 분석 기반. GLM/GPT/Claude 모델 지원. GLM-5.1 포함. 한국어 감지→GLM 자동 라우팅. sessions_spawn으로 에이전트별 모델 별도 지정. 브릿지로 실시간 채널 알림." }