The Problem

Standing up a multi-agent system on OpenClaw means hand-writing agent definitions, routing rules, and channel bindings for every deployment. The setup is repetitive, error-prone, and easy to get wrong. This repo packages that configuration so a working fleet deploys in one command.

What This Does

openclaw-agents is a configuration kit for OpenClaw. It provisions 9 specialized agents (coder, critic, ideator, planner, reviewer, scout, surveyor, writer) plus workflows, each with an agent.md, soul.md, and user.md under .agents/. The setup.sh script handles installation and configuration merging, and agents.yaml defines the agent roster.

The repo ships 4 workflow templates in .agents/workflows/ (brainstorm, daily-digest, paper-pipeline, rebuttal) and example channel configs in examples/ for Feishu, local, Telegram, and WhatsApp. It supports two deployment modes: channel-based routing to chat platforms, or local workflow mode where agents communicate via the agentToAgent tool.

How It Is Wired

This repository is configuration-heavy, not code-heavy. Static analysis found 1 code file (setup.sh, Shell) with 0 internal modules and 0 import edges. The wiring has not been mapped for this repository beyond that; the control flow lives in OpenClaw's runtime, not in this repo.

The entry point is setup.sh. It clones the repo, verifies the OpenClaw CLI is installed (installing via npm install -g openclaw@latest if missing), asks the user which deployment mode they want, then merges the agent configuration into OpenClaw's existing config. The merge is designed to append, not overwrite.

File-by-file responsibility map:

  • setup.sh — installation, CLI verification, config merge
  • agents.yaml — agent roster and routing rules
  • .agents/<agent>/agent.md — each agent's behavior definition
  • .agents/<agent>/soul.md — each agent's identity/persona
  • .agents/<agent>/user.md — each agent's user-facing instructions
  • .agents/workflows/*.md — multi-agent workflow templates
  • examples/*.json — channel-specific configuration presets
  • docs/installation.md — detailed setup guide for LLM agents

The blast radius is concentrated in setup.sh — a bug there corrupts the user's OpenClaw config. Everything else is declarative and low-risk to change.

How To Use It

Setup:

git clone https://github.com/shenhao-stu/openclaw-agents.git
cd openclaw-agents
chmod +x setup.sh
./setup.sh

Configuration: The script prompts for deployment mode (channel vs. local workflow). For channel mode, use the presets in examples/openclaw.feishu.json or similar. Model selection defaults to zai/glm-5, with per-agent overrides via --model-map.

Running it: After setup, agents activate within OpenClaw. In local workflow mode, invoke workflows directly via CLI. In channel mode, agents respond in their bound chat groups.

Real-World Use

A research team wants a paper-review pipeline in their Feishu group. They run setup.sh, choose channel mode, bind the Feishu channel, and the fleet deploys. The writer drafts, the critic applies SHARP quality gates, the reviewer flags issues, and the rebuttal workflow handles responses — all without manual agent configuration.

Code Health & Issues

Measured analysis (static, not opinion):

  • Medium/SDLC — No test files detected; untested code paths repository-wide
  • Medium/SDLC — No CI/CD pipeline; no automated build/test gate (no .github/ or CI config)

Additional observations from the file structure:

  • Low — No lockfile committed; dependency versions for OpenClaw CLI are unpinned
  • Low — No Dockerfile; deployment assumes a host with Node.js/npm

License is present (MIT). No committed secrets found.

The Bottom Line

A practical, well-organized configuration kit for OpenClaw multi-agent deployments. The value is in the curated agent definitions and workflows, not the code — setup.sh is the only executable and it's straightforward. Worth using if you run OpenClaw and want a working multi-agent fleet without hand-writing configs. The lack of tests and CI is acceptable for a config repo, but pinning the OpenClaw version would reduce setup risk.