openclaw-workspace

Summary

This repository is a backup of an OpenClaw AI workspace, containing detailed agent work specifications and operational protocols for all key agent roles. It provides agent configuration, guidelines, role-specific duties, memory system migration notes, and operational conventions, supporting rapid restoration or replication of OpenClaw agent workspaces.

Included Assets

• Work specifications (AGENTS.md) for all agent roles:
  • Analyst (洞察分析师)
  • Coder (开发工程师)
  • Cron Worker (定时任务执行者)
  • Researcher (洞察分析师)
  • Reviewer (质量仲裁官)
  • Main agent (主Agent)
• Parallel agent instance guidelines (e.g., analyst-02, coder-02, researcher-02, cron-worker-02, reviewer-02) with clear references to their main role’s responsibilities.
• Critical protocol files for each agent and instance.
• Explicit memory system migration rules and forbidden operations.
• Strict change-control and configuration protection instructions.

How to Use

1. Clone or download this repository as your OpenClaw workspace backup.
2. Refer to each agent’s AGENTS.md for role-specific operation, boundaries, and execution protocols.
3. Observe the updated memory management:
   • All user profile, memory, and skill data are to be accessed via the MemOS (/root/.openclaw/memos-local/memos.db) using memory_search and memory_write_public APIs.
   • Legacy memory files (SOUL.md, USER.md, or any memories/ directory content) must not be read or written.
4. When restoring or replicating a workspace:
   • Ensure agents follow only the instructions and boundaries described in their corresponding AGENTS.md.
   • Do not attempt to modify or commit changes to openclaw.json—changing this file is strictly prohibited and can cause unrecoverable system failure.
   • All operational outputs, reports, and code must be saved to files as specified by agent protocols.

Notes

• Each AGENTS.md strictly defines agent responsibilities, output formats, submission guidelines (e.g., using Chinese, including relevant technical terms), and required filesystem paths.
• The workspace enforces a clear operational split:
  • Analysts/proposal (事前/事中)
  • Reviewers/quality (事后)
  • Coders/implementation
  • Cron Workers/automation
  • Researchers/information discovery
• All agents must write outputs to files and are prohibited from returning only conversational results.
• All commit messages should follow the described convention and be written in Chinese.
• Configuration file editing (openclaw.json) is forbidden in all circumstances.
• The workspace is designed for use with OpenClaw’s API and conventions. No additional automation or unsupported features are provided beyond the included markdown work standards and protocols.