Setup
If ~/chat-rooom/ does not exist or is empty, read setup.md silently. Default to local-first coordination and keep persistence light until the user confirms they want a durable room workflow.
When to Use
User wants multiple agents to talk, coordinate, debate, or hand off work without copying terminal output around. Use when a task benefits from channels, mentions, lightweight ownership, or an auditable shared log inside the current workspace.
Architecture
Skill memory lives in ~/chat-rooom/. Active rooms live in the current workspace at .chat-rooom/. See memory-template.md for both templates.
CODEBLOCK0
Quick Reference
| Topic | File |
|---|
| Setup process | INLINECODE5 |
| Memory template |
memory-template.md |
| Room protocol |
protocol.md |
| Daily operations |
operations.md |
| Example room patterns |
patterns.md |
Core Rules
1. Start Coordination Inside a Named Room
- - Create or join one named room before multi-agent work starts.
- Keep one room per objective, incident, or milestone so decisions stay discoverable.
- Do not scatter the same coordination across scratch files, comments, and terminal notes.
2. Make Every Message Addressable
- - Each message should carry one primary intent: ask, update, proposal, decision, block, or handoff.
- Use
@agent mentions for directed work. Use @all only for blocking context changes or final checkpoints. - Link exact paths, commands, or commits instead of pasting large blobs that bury the action item.
3. Claim Shared Surfaces Before Editing Them
- - Update the claims table before touching the same file, test target, or subtask as another agent.
- Claim the smallest useful surface to reduce idle waiting.
- Refresh or release stale claims when work is done, blocked, or handed off.
4. Read the Summary First and Repair It Often
- - On join, read the room summary before scrolling the whole channel history.
- When a thread pauses, update summary with status, decisions, open questions, and next owner.
- If summary and channel history diverge, trust the newer timestamp and fix the summary immediately.
5. Separate Channels by Intent
- - Keep
general for status, review for critique, build for execution details, and incident for live recovery. - Create a new channel when one topic would bury another.
- Once a task becomes active, avoid mixing debate and execution in the same channel.
6. Keep the Room Local and Auditable
- - Prefer workspace files and local tools over a hosted chat backend unless the user explicitly asks for one.
- Treat the room as an operational log, not as private memory.
- Never store secrets, tokens, or unrelated personal data in room files.
Common Traps
- - Starting a room without a clear objective or roster -> duplicate work and vague ownership.
- Posting long monologues instead of targeted asks -> agents miss the real action item.
- Editing shared files without a claim -> merge collisions and silent overwrites.
- Leaving a room without updating summary or jobs -> the next agent rereads everything.
- Using
@all for routine chatter -> noisy wakeups and wasted context.
Security & Privacy
Data that leaves your machine:
- - None from this skill itself
Data that stays local:
- - Room logs and defaults in
~/chat-rooom/ and .chat-rooom/ inside the active workspace
This skill does NOT:
- - Require a hosted backend
- Access undeclared folders outside the active workspace and INLINECODE19
- Store credentials or secrets in room logs
Scope
This skill ONLY:
- - Sets up local chatroom coordination patterns for multiple agents
- Keeps channels, claims, jobs, and summaries consistent
- Helps agents talk through room files instead of terminal copy-paste
This skill NEVER:
- - Promise real-time transport that is not available locally
- Replace version control or formal code review
- Treat room logs as a secret store
Related Skills
Install with
clawhub install <slug> if user confirms:
- -
chat - Communication preference memory for cleaner agent interactions. - INLINECODE22 - Agent behavior and prompting patterns for consistent roles.
- INLINECODE23 - Multi-agent system design and safety boundaries.
- INLINECODE24 - Multi-agent operating patterns and coordination strategy.
- INLINECODE25 - Structured handoffs when work should move between agents.
Feedback
- - If useful: INLINECODE26
- Stay updated: INLINECODE27
设置
如果 ~/chat-rooom/ 不存在或为空,则静默读取 setup.md。默认采用本地优先的协调方式,并保持轻量级持久化,直到用户确认需要持久化的房间工作流。
使用时机
用户希望多个智能体进行对话、协调、辩论或交接工作,而无需复制终端输出。当任务受益于频道、提及、轻量级所有权或当前工作空间内可审计的共享日志时使用。
架构
技能记忆存储在 ~/chat-rooom/ 中。活跃房间位于当前工作空间的 .chat-rooom/ 目录下。两种模板详见 memory-template.md。
text
~/chat-rooom/
|- memory.md # 激活默认值和持久化偏好
|- rooms.md # 近期房间名称、角色和约定
- - patterns.md # 反复有效的协调模式
/.chat-rooom/
|- room.md # 目的、成员名单、频道、状态
|- summary.md # 快照、决策、下一步行动
|- jobs.md # 工作项及其负责人和状态
|- claims.md # 文件、任务或测试的所有权
|- channels/general.md # 共享时间线
|- channels/review.md # 评审和审批请求
- inbox/.md # 待处理的提及和定向请求
快速参考
memory-template.md |
| 房间协议 | protocol.md |
| 日常操作 | operations.md |
| 房间模式示例 | patterns.md |
核心规则
1. 在命名房间内开始协调
- - 在多智能体工作开始前,创建或加入一个命名房间。
- 每个目标、事件或里程碑保持一个房间,以便决策可追溯。
- 不要将同一协调分散到临时文件、注释和终端笔记中。
2. 每条消息都应可寻址
- - 每条消息应携带一个主要意图:询问、更新、提议、决策、阻塞或交接。
- 使用 @agent 提及进行定向工作。仅在阻塞性上下文变更或最终检查点时使用 @all。
- 链接确切的路径、命令或提交,而不是粘贴会掩盖行动项的大段内容。
3. 在编辑前声明共享资源
- - 在接触与其他智能体相同的文件、测试目标或子任务前,更新声明表。
- 声明最小的有用资源以减少空闲等待。
- 当工作完成、受阻或交接时,刷新或释放过期的声明。
4. 首先阅读摘要并经常修复
- - 加入时,先阅读房间摘要,再滚动查看整个频道历史。
- 当线程暂停时,更新摘要,包含状态、决策、未解决问题和下一个负责人。
- 如果摘要与频道历史不一致,信任更新的时间戳并立即修复摘要。
5. 按意图分离频道
- - 保持 general 用于状态,review 用于评审,build 用于执行细节,incident 用于实时恢复。
- 当一个主题会掩盖另一个主题时,创建新频道。
- 一旦任务变为活跃状态,避免在同一频道中混合辩论和执行。
6. 保持房间本地化和可审计性
- - 优先使用工作空间文件和本地工具,而非托管聊天后端,除非用户明确要求。
- 将房间视为操作日志,而非私有记忆。
- 切勿在房间文件中存储密钥、令牌或无关的个人数据。
常见陷阱
- - 在没有明确目标或成员名单的情况下启动房间 -> 重复工作和模糊的所有权。
- 发布长篇独白而非定向请求 -> 智能体错过真正的行动项。
- 未声明所有权就编辑共享文件 -> 合并冲突和静默覆盖。
- 离开房间时不更新摘要或工作项 -> 下一个智能体重新阅读所有内容。
- 对日常闲聊使用 @all -> 嘈杂的唤醒和浪费的上下文。
安全与隐私
离开您机器的数据:
保持本地的数据:
- - 房间日志和默认值存储在 ~/chat-rooom/ 和活跃工作空间内的 .chat-rooom/
此技能不会:
- - 需要托管后端
- 访问活跃工作空间和 ~/chat-rooom/ 之外的未声明文件夹
- 在房间日志中存储凭据或密钥
范围
此技能仅:
- - 为多个智能体设置本地聊天室协调模式
- 保持频道、声明、工作项和摘要的一致性
- 帮助智能体通过房间文件而非终端复制粘贴进行对话
此技能绝不:
- - 承诺本地不可用的实时传输
- 替代版本控制或正式代码评审
- 将房间日志视为密钥存储
相关技能
如果用户确认,使用 clawhub install
安装:
- - chat - 通信偏好记忆,用于更清晰的智能体交互。
- agent - 智能体行为和提示模式,用于一致的角色。
- agents - 多智能体系统设计和安全边界。
- agentic-engineering - 多智能体操作模式和协调策略。
- delegate - 工作应在智能体间移动时的结构化交接。
反馈
- - 如果有用:clawhub star chat-rooom
- 保持更新:clawhub sync