← 返回
AI智能

Multi Workplace

Manage multiple workplaces (project directories) with multi-agent orchestration, isolated memory, and inter-agent communication. Use when the user mentions:...
利用多智能体编排、隔离记忆及智能体间通信,管理多个工作区(项目目录)。当用户提及……时使用。
dickwu
AI智能 clawhub v0.4.0 1 版本 99915 Key: 无需
★ 0
Stars
📥 1,176
下载
💾 25
安装
1
版本
#latest

概述

Workplace Skill

Manage multiple project workplaces with per-workspace agents, isolated memory, and Swarm-style agent orchestration.

/workplace Command (Telegram / Slash)

Hierarchical navigation with parent → child drill-down.

  • /workplace or /workplace list → Show top-level view: parent workspaces and standalone workplaces as buttons. Parents show (N) child count. Current workspace marked with ✓.
  • Click a parent button → Drill into children. Shows child buttons + "Use parent" + "← Back".
  • /workplace → If standalone or child, switch directly. If parent with children, show drill-in.
  • /workplace parent:child → Direct switch using colon syntax (e.g. log-stream:logstream).
  • /workplace status → Current workspace card with parent, linked, agents, deploy envs.
  • /workplace agents → Agent list with start/stop buttons.

Colon Syntax

/workplace log-stream:logstream resolves parent by name, then finds child under that parent. Supports quick switching without navigating menus.

Context Switching

When the user switches workplaces (via button click, name, or colon syntax):

  1. Update ~/.openclaw/workspace/.workplaces/current.json with the selected UUID and path
  2. Update lastActive in registry.json
  3. Load the new workspace's .workplace/config.json for context
  4. Send confirmation: name, path, parent (if any), linked workplaces, agent list
  5. Subsequent messages in the session should be aware of the active workspace context

Read current.json at the start of any workplace operation to know which workspace is active.

See telegram-ui.md for full button layouts, callback routing, and platform fallbacks.

Quick Reference

CommandAction
-----------------
workplace init [path]Initialize workplace (scan existing or set up new)
workplace listList all workplaces (inline buttons to switch)
`workplace switch uuid>`Switch active workplace
workplace scan [path]Discover .git workplaces in subdirectories
workplace link Link a related workplace
`workplace unlink uuid>`Remove a linked workplace
workplace statusCurrent workplace info + agent status
workplace agentsList agents in current workplace
workplace agent start Start an agent (runs as sub-agent)
workplace agent stop Stop a running agent
workplace kernel startStart persistent kernel agent
workplace kernel stopStop kernel agent
`workplace export [zip\json]`Export workplace config
workplace import Import workplace from export
`workplace delete uuid>`Remove from registry
workplace deploy Show/run deploy instructions
workplace sync Generate context for cursor/claude/opencode/all

Architecture

Registry

Central registry at ~/.openclaw/workspace/.workplaces/:

  • registry.json — all known workplaces with UUID, path, hostname, links
  • current.json — currently active workplace

Per-Workplace Structure

Each project gets a .workplace/ directory:

.workplace/
├── config.json          # UUID, name, path, hostname, linked, parent
├── agents/*.md          # Agent role definitions (kernel.md always present)
├── memory/              # Isolated daily logs (YYYY-MM-DD.md)
├── skills/              # Workplace-specific skills (user-managed via git)
├── chat.md              # Inter-agent communication
├── structure.json       # Auto-scanned file tree
├── full-tree.md         # Full tree with parent + linked workplaces (by hostname)
├── process-status.json  # Agent runtime states and errors
└── deploy/              # Deployment docs: dev.md, main.md, pre.md

Workplace Detection

  • Any directory with .git/ is a potential workplace
  • Submodules included as nested workplaces
  • Parent workplace auto-detected from parent directories
  • Manual linking via workplace link

Workflows

Initialize a Workplace

  1. Run scripts/init_workplace.sh [--name ] [--desc ]
  2. For existing projects: scan file structure, read *.md files, analyze project type, suggest agents
  3. For empty folders: ask project name, description, language/framework, roles needed
  4. Creates .workplace/ structure, registers in central registry, sets as current
  5. See init-guide.md for full flow details

Agent System

Agents are defined as .md files in .workplace/agents/ with YAML frontmatter (name, role, triggers, handoff_to). Run agents via sessions_spawn with system prompts built from their definitions + workplace context.

Inter-Agent Communication

Agents communicate via chat.md using a structured message protocol. The Rust file-watcher server monitors changes and outputs parsed messages as JSON lines.

Rust File-Watcher Server

Binary at assets/bin/workplace-server-{os}-{arch}. Build from source with scripts/build.sh.

# Start server for a workplace
workplace-server /path/to/project

# Server outputs JSON lines to stdout for each new chat.md message
{"timestamp":"...","sender":"coder","recipient":"reviewer","broadcast":[],"message":"...","line_number":1}

Export/Import

  • ZIP: Full .workplace/ folder (memory excluded by default)
  • JSON: Config + agent definitions + deploy docs as portable manifest
  • Import generates a new UUID to avoid collisions

Chat UI (Telegram / Discord)

On platforms with inline buttons, workplace list shows a clickable switcher. workplace agents shows start/stop buttons per agent. workplace deploy shows environment buttons.

See telegram-ui.md for message formats, button components, and callback handling.

Fallback: numbered text lists on platforms without button support (WhatsApp, Signal).

IDE Integration

Sync workplace context to external coding tools:

  • Cursor.cursor/rules/workplace.mdc (MDC with frontmatter)
  • Claude CodeCLAUDE.md (markdown, marker-based updates)
  • OpenCodeopencode.jsonc instructions field

Run workplace sync all to update all detected IDEs, or target one: workplace sync cursor.

See ide-sync.md for implementation details.

Scripts

ScriptPurpose
-----------------
scripts/init_workplace.shInitialize .workplace/ in a directory
scripts/scan_workplaces.shFind .git workplaces under a path
scripts/build.shBuild Rust server for current platform

Supermemory Integration

Each workplace uses its UUID as containerTag for supermemory operations:

  • Kernel agent saves structure summaries and project facts
  • All workplace memories are isolated by containerTag
  • Enables cross-session project state awareness

Command Details

See commands.md for full command reference with examples.

版本历史

共 1 个版本

  • v0.4.0 当前
    2026-03-29 06:44 安全 安全

安全检测

腾讯云安全 (Keen)

安全,无风险
查看报告

腾讯云安全 (Sanbu)

安全,无风险
查看报告

🔗 相关推荐

ai-intelligence

Proactive Agent

halthelobster
将AI智能体从任务执行者升级为主动预判需求、持续优化的智能伙伴。集成WAL协议、工作缓冲区、自主定时任务及实战验证模式。Hal Stack核心组件 🦞
★ 839 📥 213,456
ai-intelligence

self-improving agent

pskoett
捕获经验教训、错误和纠正,以实现持续改进。使用时机:(1)命令或操作意外失败;(2)用户纠正……
★ 4,062 📥 799,675
ai-intelligence

Self-Improving + Proactive Agent

ivangdavila
自我反思+自我批评+自我学习+自组织记忆。智能体评估自身工作、发现错误并持续改进。
★ 1,362 📥 319,004