首页openclaw插件 › 会下指令 TaskFit — TaskFit — 指令优化插件

代码插件 安全

会下指令 TaskFit — TaskFit — 指令优化插件

v0.1.1

OpenClaw插件,仅使用从同一设备成功历史任务中本地学习的匿名结构模式,将用户的粗略请求重写为更清晰的执行简报,不上传本地历史或发送原始任务历史文本到模型。

0· 1·0 当前
下载插件包 项目主页
最后更新
2026/4/8
安全扫描
VirusTotal
无害
查看报告
OpenClaw
安全
medium confidence
该插件的代码、文档和运行时指令内部一致:它访问本地OpenClaw历史DB学习模式,本地重写用户请求,在可见代码中不请求外部凭证或网络访问。
安全有层次,运行前请审查代码。

版本

latestv0.1.12026/4/8
● 无害

安装命令 点击复制

官方npx clawhub@latest install @sheygoodbai/openclaw-taskfit
镜像加速npx clawhub@latest install @sheygoodbai/openclaw-taskfit --registry https://cn.clawhub-mirror.com

插件文档

会下指令 TaskFit

会下指令 TaskFit is an OpenClaw plugin that rewrites the user's rough request into a clearer execution brief before it reaches the agent.

It is designed for the opposite side of the same problem that NoJargon solves. NoJargon makes the agent speak clearly back to people. TaskFit makes rough user intent easier for the agent to execute correctly.

Canonical install page:

Why it is different

  • OpenClaw-native: it works inside the prompt build path instead of as copy/paste
  • local-history-aware: it learns only anonymous structure patterns from successful local tasks on the same device
  • privacy-first: it does not upload local task history, local files, or raw task-history text to any plugin service
  • read-only by design: it opens the local task-history file in read-only mode
  • no raw history exposure: it never forwards raw local task-history text directly to the model
  • command-light: it can help even when the user gives a rough, short, or underspecified request

What it does

  • reads successful local OpenClaw task history on the same device in read-only mode
  • filters out noisy or trivial tasks
  • finds structurally similar successful tasks locally
  • extracts anonymous high-signal patterns such as evidence-first checks, explicit deliverables, and verified/unverified boundaries
  • rewrites the current user request into a clearer execution brief with /taskfit on|off|status|adaptive|always

Privacy boundary

TaskFit is intentionally designed so ClawHub and the plugin author do not get the user's raw local task history.

  • local history stays on the user's device
  • the history file is opened read-only
  • similar-task matching happens on-device
  • TaskFit makes no network request to process history
  • raw history task text is not uploaded anywhere
  • raw task-history snippets are not injected into the model prompt
  • only the current turn gets rewritten, using anonymous structural patterns learned locally

Install

openclaw plugins install clawhub:@sheygoodbai/openclaw-taskfit
openclaw plugins enable taskfit

Then turn optimization on in chat:

/taskfit adaptive

openclaw plugins enable taskfit only loads the plugin. The actual optimization switch stays off until you run /taskfit on or /taskfit adaptive.

If ClawHub returns 429 Rate limit exceeded, use the source fallback:

git clone https://github.com/Sheygoodbai/taskfit.git
cd taskfit
openclaw plugins install -l .
openclaw plugins enable taskfit

Commands

  • /taskfit on
  • /taskfit off
  • /taskfit adaptive
  • /taskfit always
  • /taskfit status
  • /taskfit help

Config

{
  "plugins": {
    "entries": {
      "taskfit": {
        "config": {
          "enabled": true,
          "mode": "adaptive",
          "toneMatch": true,
          "respectStrongPrompts": true,
          "historyLimit": 600,
          "maxMatches": 3,
          "minSimilarity": 0.22,
          "minTaskSeconds": 20,
          "maxTaskSeconds": 5400,
          "trackStats": true
        }
      }
    }
  }
}
数据来源:ClawHub ↗ · 中文优化:龙虾技能库
OpenClaw 技能定制 / 插件定制 / 私有工作流定制

免费技能或插件可能存在安全风险,如需更匹配、更安全的方案,建议联系付费定制

了解定制服务