五大核心能力，让 Agent 从"建议者"变成"完成者"。每一项都建立在权限可控、本地优先的工程基础之上。 Five core capabilities that turn the agent from a suggester into a finisher — built on permission gating and local-first storage.
任务既能直接在你的电脑上跑（速度快，可访问真实文件），也能放进 OpenClaw 沙箱跑（隔离危险命令）。每一次执行前你都能看到完整指令。 Tasks run either directly on your machine (fast, full file access) or inside the OpenClaw sandbox (isolated for risky commands). You see the full command every time.
Office 文档生成、Web 搜索、Playwright 自动化、Remotion 视频生成 — 每个技能都是独立的可执行模块，按需加载、可自由扩展。 Office authoring, web search, Playwright automation, Remotion video. Each skill is a self-contained module — loaded on demand, easy to extend.
把指令发到微信、企业微信、钉钉、飞书、Telegram… Agent 在桌面上干完活，把结果（PPT、图片、链接）回推到聊天里。 Send instructions through WeChat, WeCom, DingTalk, Lark, Telegram, and others. The agent finishes the work on your desktop and posts the result (deck, image, link) back to the chat.
对话中提到的偏好、身份、常用项目，会自动沉淀成结构化记忆条目。下次开新会话时，Agent 不必再问一遍。 Preferences, identity, recurring projects mentioned in chat are auto-distilled into structured memory entries. Next session, the agent doesn't ask again.
对话式发起，或在 GUI 里手动添加。每日新闻、邮箱整理、周期性报告 — 时间一到 Agent 自动开干，干完通过 IM 推送结果。 Set up by chat or via the GUI. Daily news, inbox triage, periodic reports — the agent runs on time and pushes results to your IM.
