About this skill

Problem

Local multi-agent development teams are easy to conceptualize but tedious to wire up: a coordinator, developer, and tester need separate workspaces, tool paths, model defaults, and report conventions. This skill materializes a local OpenClaw team under ~/.openclaw/ and routes work from main to my_dev and my_test.

How It Works

Incremental configuration: reads the current model from openclaw.json, asks for confirmation, and applies only a patch instead of overwriting existing agent settings.
Team structure: creates main, my_dev, and my_test, then populates SOUL.md, AGENTS.md, TOOLS.md, and IDENTITY.md for the dev and test roles.
Task loop: the developer writes reports/{task_id}-dev.md; the tester writes reports/{task_id}-test.md; main reads recommendation to complete or retry, with a maximum of 3 retries.
Stack templates: covers common local full-stack scenarios such as Vue 3 + TypeScript, Spring Boot 3 + Java 17, and MyBatis-Plus.

Boundaries

The skill operates only on local OpenClaw configuration and workspaces. It does not make external network requests or upload API keys. It should not be used to provision external services. Its shutdown rules target only frontend and backend processes started for a test run, and explicitly avoid killing openclaw, databases, or other infrastructure.

Screenshots

Use Cases

After a user asks to set up a dev team, generate local OpenClaw main, dev, and test agents with workspaces.
Before building a LLM full-stack app, write Vue 3 + TypeScript and Spring Boot 3 dev/test tool templates locally.
After a dev handoff, have my_test produce a test report and use recommendation to complete or retry up to 3 times.
When updating an existing team, patch only the relevant OpenClaw agent config without resetting main's existing files.

Best For

AI engineers maintaining local OpenClaw multi-agent setups who want less manual routing and workspace setup.
Full-stack or backend engineers building Vue 3 and Spring Boot 3 apps who want dev agent stack and report presets.
QA or test engineers doing local acceptance who want my_test to output fixed completion or retry recommendations.
Project leads using LLM-assisted development who want a skill to split PM, dev, and test roles into local agents.