什么是 专注模式？

专注模式是一款专为开发者和高级用户设计的扩展功能，旨在帮助他们维持心流状态。通过在会话开始时定义具体目标，该技能将 AI 代理转变为一名问责伙伴，主动监控对话中的跑题、拖延或需求蔓延。对于处理多动症（ADHD）式工作流或高压开发任务的用户来说，它是 Openclaw Skills 生态系统中不可或缺的工具。

该技能的工作原理是根据一组偏离信号评估每一次交互。无论您是过度深入研究无关的库，还是被支线任务分散了注意力，专注模式都会提供上下文相关的提醒，将您的注意力带回主要目标。它利用“停放区”（Parking Lot）系统确保灵感不会丢失，允许您保存偏离的想法以供后续使用，而不会打破当前的专注。

下载入口:https://github.com/openclaw/skills/tree/main/skills/savorgbot-exe/focus-mode

安装与下载

1. ClawHub CLI

从源直接安装技能的最快方式。

2. 手动安装

将技能文件夹复制到以下位置之一

优先级：工作区 > 本地 > 内置

3. 提示词安装

将此提示词复制到 OpenClaw 即可自动安装。

专注模式 应用场景

• 通过防止需求蔓延，在复杂的重构会话中保持势头。
• 通过温和的重定向和问责，管理多动症（ADHD）驱动的开发工作流。
• 追踪在特定目标与偏离研究上花费的时间和精力。
• 将支线灵感卸载到结构化的待办事项中，同时保持心流状态。

1. 用户通过 focus 命令提供具体目标来激活会话。
2. 该技能在本地配置目录中初始化状态文件，以追踪开始时间、目标和重定向次数。
3. 分析随后的每条消息，对比主题相关信号与偏离信号（如无关话题或拖延模式）。
4. 根据偏离程度，该技能提供升级的响应模式，从轻微的旁注到直接的重新聚焦提示。
5. 偏离的话题可以移动到“停放区”，记录上下文和时间戳以供后续回顾。
6. 停用时，该技能生成一份全面的总结，包括持续时间、任务执行百分比以及停放项目的列表。

专注模式 配置指南

要在您的环境中启用专注模式，请确保配置目录存在，然后触发激活命令。这些 Openclaw Skills 需要文件系统访问权限以持久化会话数据。

# 确保配置目录存在
mkdir -p ~/.config/clawdbot-focus/

# 在代理界面中激活专注模式
/focus "重构身份验证模块"

专注模式 数据架构与分类体系

专注模式使用 JSON 文件组织其状态和积压工作，以确保跨会话的持久性。这些文件存储在用户的主目录中。

name: focus-mode
description: Help users stay focused on a specific goal or task. Activate with "/focus " to set a focus target. Monitors conversation for drift and gently redirects back to the focus area. Use when user wants accountability, has ADHD-style workflows, tends to go down rabbit holes, or explicitly asks to stay on track. Deactivate with "/focus off".

Focus Mode

Keep users on track toward their stated goal by detecting conversation drift and providing gentle nudges.

Activation

When user says /focus :

1. Store the focus goal in ~/.config/clawdbot-focus/current.json
2. Acknowledge: "e??ˉ Focus mode ON: {goal}. I'll help keep you on track."
3. Note the start time

When user says /focus off or /focus done:

1. Calculate session duration and drift stats
2. Show summary: time spent, tangents parked, redirects given
3. Clear the focus state

When user says /focus (no args):

1. Show current focus if active, or prompt to set one

Drift Detection

After each user message, evaluate:

On-topic signals:

• Directly relates to the focus goal
• Asks for help with a subtask of the goal
• Reports progress on the goal
• Asks clarifying questions about the goal

Drift signals:

• Completely unrelated topic introduced
• Deep dive into tangent that doesn't serve the goal
• Scope creep ("while we're at it, let's also...")
• Procrastination patterns ("actually, first let me...")

Response Patterns

Light drift (related but tangential):

• Continue helping, but add: "(Noting this as a side thread a?? we can circle back after {goal})"

Medium drift (unrelated topic):

• Help briefly, then: "Want me to park this for later? Still have {goal} on deck."

Heavy drift (extended tangent, 3+ exchanges off-topic):

• Direct but kind: "We've drifted a bit from {goal}. Ready to refocus, or is this tangent worth pursuing?"

Time-based nudge (30+ min on tangent):

• "You've been on this for a while. Still serving the main goal, or should we context-switch back?"

Parking Lot

When user says "park this" or you suggest parking a tangent:

1. Append to ~/.config/clawdbot-focus/parked.json: {topic, timestamp, context}
2. Confirm: "Parked: {topic}. Will remind you after focus session."

After /focus off, list parked items:

e??? Parked tangents:
a?￠ Research that API library (from 20min ago)
a?￠ Check Discord notification settings

Tone Modes

User can set tone with /focus tone :

• gentle (default): Soft suggestions, doesn't interrupt flow
• strict: More direct, shorter leash on tangents
• accountability: Includes time pressure ("You said by 9pm, it's 8:30")

State File Format

~/.config/clawdbot-focus/current.json:

{
  "goal": "Ship Alithos news page",
  "started": "2026-01-30T21:22:00Z",
  "tone": "gentle",
  "deadline": null,
  "redirects": 0,
  "tangent_time_sec": 0
}

Session Summary

On /focus off or /focus done:

e??ˉ Focus session complete!

Goal: Ship Alithos news page
Duration: 1h 42m
On-task: ~78%
Redirects: 3
Parked: 2 tangents

Nice work staying focused! e???

Integration Notes

• Works alongside other skills a?? doesn't block anything
• Drift detection runs passively, doesn't slow responses
• State persists across session restarts via JSON file
• Can be combined with Pomodoro-style timers if user requests