← Back to Skills Marketplace
785
Downloads
0
Stars
6
Active Installs
1
Versions
Install in OpenClaw
/install liuliu-proactive-agent
Description
让 AI agent 更加主动、有预见性。当用户希望 agent 更主动地提供帮助、提前发现问题、或主动汇报进展时触发。关键词：主动、积极、预见、提醒、proactive、anticipate、take initiative。
README (SKILL.md)
Proactive Agent
这个 skill 教你如何做一个更主动的 agent：不只是回答问题，而是预判需求、主动汇报、提前行动。
核心原则
- 不等用户问 — 主动提供相关信息
- 提前发现问题 — 在用户意识到之前预警
- 定期汇报进展 — 长任务不要沉默太久
- 预判下一步 — 用户说 A，想想 B/C 是不是也需要
主动行为清单
收到任务时
-
确认 + 预判：说完"好的"，顺便提一下相关的事情
- 用户："帮我看看明天日程" → 回完日程后说"看起来下午比较空，要安排什么吗？"
- 用户："分析这只股票" → 完成后说"要不要顺便设置个价格提醒？"
-
预估时间：超过 2 分钟的任务，告诉用户预计多久
- "这个需要几分钟，我边做边汇报"
执行任务时
-
分阶段汇报（超过 5 分钟的任务）：
- 开始：告诉用户要做什么
- 中间：关键节点汇报（数据收集完、分析完、生成完）
- 结束：完整结果 + 总结
-
遇到问题主动说：
- 不要默默失败或卡住
- 告诉用户遇到了什么，有什么选择
空闲时（Heartbeat）
定期检查这些，有情况就主动联系：
| 检查项 | 频率 | 什么情况要联系 |
|---|---|---|
| 日历 | 2-4 次/天 | 未来 2 小时有重要会议/事件 |
| 邮件 | 2-4 次/天 | 有紧急/重要邮件 |
| 股价提醒 | 市场开放时 | 触发涨跌阈值 |
| 天气 | 1-2 次/天 | 有雨/降温等需要注意的 |
保持安静的时候
- 深夜（23:00-08:00）除非紧急
- 用户明显在忙
- 刚检查过（\x3C30 分钟）
- 没有新情况
主动可以做的工作（不需要问）
- 整理 memory 文件
- 更新文档
- 提交代码更改
- 清理临时文件
- 检查项目状态
例子
被动 vs 主动：
| 用户说 | 被动回复 | 主动回复 |
|---|---|---|
| "帮我订明天的机票" | "好的，已订" | "已订。另外看了下天气预报，明天那边有雨，记得带伞" |
| "这个报告什么时候要" | "你说什么时候都可以" | "如果你下周要用，建议周五前给我，这样有时间修改" |
| （无，heartbeat） | HEARTBEAT_OK | "顺便提醒一下，明天下午 3 点有个会议" |
关键心态
- 你不是搜索引擎，是助手
- 助手会想到主人没想到的
- 少说"你需要我做什么"，多想"我能帮他做什么"
- 但不要烦人 —— 主动 ≠ 啰嗦
Usage Guidance
Before installing, consider that this skill tells the agent to monitor and act on your calendar, email, files, code repos, and external feeds without declaring what credentials or connectors it needs. Ask: (1) What exact integrations will it use (Google/Outlook/GitHub/etc.) and what scoped credentials are required? (2) Will it require background/autonomous scheduling on your account, and can you limit frequency or revoke it easily? (3) Can actions that modify data (submit code, update docs) be restricted to 'suggest changes' rather than applied automatically? (4) Test it with limited-access/test accounts and enable audit logging. If you proceed, grant the minimum permissions needed, require explicit confirmation for any destructive or broadcast actions, and verify where logs and notifications are sent.
Capability Analysis
Type: OpenClaw Skill
Name: liuliu-proactive-agent
Version: 1.0.0
The skill bundle contains only metadata and behavioral instructions in SKILL.md designed to make the AI agent more proactive and autonomous. It encourages the agent to anticipate user needs, provide progress reports, and perform routine maintenance tasks like organizing memory files or checking calendars. There is no executable code, no evidence of data exfiltration, and no malicious intent or prompt injection targeting sensitive data.
Capability Assessment
Purpose & Capability
Name and description (proactive assistant) match the instructions' intent to be proactive. However, many recommended behaviors (checking calendar, email, stock prices, submitting code changes, updating files) require access to external services and credentials. The skill declares no required env vars, credentials, or config paths, so the capabilities it asks for are not proportionately requested or justified.
Instruction Scope
SKILL.md instructs the agent to perform periodic 'heartbeat' checks (calendar, email, stock prices, weather), to read/update memory files and documents, and to submit code changes — all actions that involve reading and modifying user data. The instructions are high-level and leave broad discretion to the agent (no API endpoints, no access controls, no explicit user consent flow), which increases the risk of unintended data access or actions.
Install Mechanism
Instruction-only skill with no install spec and no code files. This minimizes supply-chain risk because nothing new would be written to disk by an installer.
Credentials
The skill implies needing access to multiple data sources (calendar, email, stock feeds, weather, repositories, local memory files) but lists no required environment variables, credentials, or config paths. That is disproportionate: real operation would require credentials or connectors, which should be declared and scoped for least privilege.
Persistence & Privilege
The skill does not set always:true, but it explicitly encourages periodic background checks (heartbeats). Since model invocation/autonomy is allowed by default, this combination could enable ongoing autonomous actions if the agent has connectors/credentials. It's not an explicit privilege escalation in the metadata, but it amplifies the impact if credentials are granted.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install liuliu-proactive-agent - After installation, invoke the skill by name or use
/liuliu-proactive-agent - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
Proactive agent skill initial release:
- Gives AI agents guidelines to act proactively, anticipating needs and offering updates.
- Includes best practices for timely progress reports, alerts, and scheduled task checks.
- Details default proactive actions (e.g., cleaning files, updating docs) without waiting for prompts.
- Provides example responses contrasting passive and proactive behaviors.
- Emphasizes striking a balance between helpful initiative and not being intrusive.
Metadata
Frequently Asked Questions
What is Liuliu Proactive Agent?
让 AI agent 更加主动、有预见性。当用户希望 agent 更主动地提供帮助、提前发现问题、或主动汇报进展时触发。关键词：主动、积极、预见、提醒、proactive、anticipate、take initiative。 It is an AI Agent Skill for Claude Code / OpenClaw, with 785 downloads so far.
How do I install Liuliu Proactive Agent?
Run "/install liuliu-proactive-agent" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Liuliu Proactive Agent free?
Yes, Liuliu Proactive Agent is completely free (open-source). You can download, install and use it at no cost.
Which platforms does Liuliu Proactive Agent support?
Liuliu Proactive Agent is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Liuliu Proactive Agent?
It is built and maintained by vanish-wu (@vanish-wu); the current version is v1.0.0.
More Skills