僅限 macOS 的命令行錄製器，把示範過的桌面工作流轉成證據（session.json / events.jsonl），供 Agent 生成可複用技能。不是 dsh 外掛（無 `dsh plugin` 安裝）；透過內建宿主 Agent SKILL.md 配合 DeepSeek Harness 使用。需要 macOS + Node.js 18+。
DSH 整合
生態系相關
作者聲明
安全稽核
未稽核
最後核實
2026-08-23
授權條款
MIT
01它能幫你完成什麼？
Record a macOS desktop workflow once and hand the evidence to an agent to build a reusable skill
Structured artifacts (session.json, events.jsonl) plus a validated skill-input package (README.md, events.jsonl, session.json) the agent's own skill-creation flow consumes
Developers/agents who want to teach a Computer Use agent a UI workflow (file pickers, drag-drop, multi-app) that has no clean API, by demonstrating it once
02如何將外掛接入 DeepSeek Harness？
先決條件
- macOS
- Node.js 18+, Swift toolchain / Xcode Command Line Tools, Accessibility and Input Monitoring permissions
安裝步驟
- 01
Clone the repo: `git clone https://github.com/humblebanana/open-record-replay.git && cd open-record-replay`
$ git clone https://github.com/humblebanana/open-record-replay.git && cd open-record-replay
- 02
Run `npm install` then `npm run check`; grant macOS permissions with `node bin/orr.js permissions request`
$ npm install
驗證整合成功
作者未說明
復原
- The README documents install/check and a host-agent skill handoff, but no explicit uninstall command beyond removing the cloned directory.
03DSH 整合程度與能力邊界
Not a dsh plugin: a macOS CLI recorder that captures desktop workflows into session.json/events.jsonl so a coding agent (incl. DeepSeek Harness) can turn a demonstration into a reusable Computer Use skill.
macOS workflow recorder
A demonstrated desktop workflow→session.json + events.jsonl capturing window.changed, mouse.click/drag, keyboard.text_input/submit, selection.changed, app/window attribution, UI targets, accessibility tree
The core recorder does not require Screen Recording; recordings are local by defaultValidation & skill packaging
A recorded session→`session validate-recording` checks quality; `skill prepare` writes a skill-input package (README.md, events.jsonl, session.json) for the agent
The final skill is created by the agent's own skill system, not by this toolHost-agent skill
Agent instructions→An included SKILL.md tells the agent when to start/stop recording, how to inspect events.jsonl, and how to hand off the evidence package
The agent should stop its turn while the user demonstrates, then resume after
04適合誰？何時不該用？
適合
- Developers/agents who want to teach a Computer Use agent a UI workflow (file pickers, drag-drop, multi-app) that has no clean API, by demonstrating it once
不適合
- Open Record/Replay is a macOS-only, alpha-stage CLI; it requires Node.js 18+, the Swift toolchain / Xcode Command Line Tools, and Accessibility + Input Monitoring permissions. The core recorder does not need Screen Recording.
- events.jsonl can contain sensitive data (window titles, URLs, typed/selected text, file names, local paths, accessibility-tree text). Review recordings before sharing; do not publish raw recordings with secrets or private data.
05相容性、維護與安全提醒
- Open Record/Replay is a macOS-only, alpha-stage CLI; it requires Node.js 18+, the Swift toolchain / Xcode Command Line Tools, and Accessibility + Input Monitoring permissions. The core recorder does not need Screen Recording.
- events.jsonl can contain sensitive data (window titles, URLs, typed/selected text, file names, local paths, accessibility-tree text). Review recordings before sharing; do not publish raw recordings with secrets or private data.
- This is a generic agent tool, not a DeepSeek Harness plugin — there is no `dsh plugin` install path. It works with any coding agent's skill-creation flow, including DeepSeek Harness, via the included host-agent SKILL.md.
MIT · 由 facts pack 抓取（README 分支 main）
06常見問題
Open Record/Replay 是 DeepSeek Harness 外掛嗎？
不是。它是一個獨立的 macOS CLI 工具，沒有 `dsh plugin` 安裝路徑。它配合任何編碼 Agent 的技能建立流程使用——包括 DeepSeek Harness——透過內建的宿主 Agent SKILL.md 告訴 Agent 何時錄製、如何交接證據。
怎麼安裝？
複製倉庫，執行 `npm install`，然後 `npm run check`。用 `node bin/orr.js permissions request` 授予 macOS 輔助功能與輸入監控權限。需要 macOS、Node.js 18+ 與 Swift 工具鏈 / Xcode Command Line Tools。
一次錄製會產出什麼？
session.json 與 events.jsonl，記錄視窗變化、滑鼠點擊 / 拖曳、鍵盤輸入、選取內容與輔助功能樹上下文。`session validate-recording` 校驗品質，`skill prepare` 寫出供 Agent 使用的技能輸入包。
我的資料安全嗎？
錄製預設保存在本機，但 events.jsonl 可能包含敏感資訊（視窗標題、URL、輸入文字、檔案路徑、輔助功能樹文字）。分享前請先檢查，切勿公開含密鑰或隱私的原始錄製。
07相關的 DSH 工作流程
archify
作者 tt-a1i
為編碼智慧體生成美觀可驗證的架構圖、時序圖與資料流圖，輸出自包含 HTML，支援動效與清晰匯出。
openviking
作者 volcengine
為 AI 智慧體打造的自進化上下文資料庫，統一智慧體記憶、知識 RAG 與技能。
nocobase
作者 nocobase
開源的 AI + 無程式碼應用搭建平臺：AI 在久經生產驗證的底層基礎設施與所見即所得的無程式碼介面之上工作，幫你快速構建 CRM、ERP 等業務系統，兼顧速度與可靠性。
learn-harness-engineering
作者 walkinglabs
Harness 工程新手教程，從 0 到 1 系統學習智慧體工作流框架。
08資料與來源
Teach Computer Use agents by demonstration.
git clone https://github.com/humblebanana/open-record-replay.git
此頁面根據專案公開文件、儲存庫中繼資料與 DSH Plugins 的結構化解析所產生；最後核實於 2026-08-23。發現錯誤？提交更正。
最佳 DeepSeek Harness 外掛
從全目錄挑出的 12 個值得優先安裝的外掛，涵蓋各個分類。