Skill flagged — suspicious patterns detected

ClawHub Security flagged this skill as suspicious. Review the scan results before using.

Proactive Intelligence

v2.3.1

主动智能:预测需求 + 自我改进 + 智能记忆 + 技能管理 + 技能进化。融合 proactivity 和 self-improving 的核心功能,并添加自动技能升级和编辑能力。

0· 95·0 current·0 all-time
bychangle@cle87937-code

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for cle87937-code/proactive-intelligence.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "Proactive Intelligence" (cle87937-code/proactive-intelligence) from ClawHub.
Skill page: https://clawhub.ai/cle87937-code/proactive-intelligence
Keep the work scoped to this skill only.
After install, inspect the skill metadata and help me finish setup.
Use only the metadata you can verify from ClawHub; do not invent missing requirements.
Ask before making any broader environment changes.

Command Line

CLI Commands

Use the direct CLI path if you want to install manually and keep every step visible.

OpenClaw CLI

Bare skill slug

openclaw skills install proactive-intelligence

ClawHub CLI

Package manager switcher

npx clawhub@latest install proactive-intelligence
Security Scan
VirusTotalVirusTotal
Suspicious
View report →
OpenClawOpenClaw
Suspicious
high confidence
Purpose & Capability
The skill claims to analyze, edit, and upgrade other skills — the included code (skill-evolver.py, skill-manager.py) implements exactly that: scanning ~/.openclaw/workspace/skills, backing up, analyzing, and (optionally) fixing files. Access to skill directories and ability to run 'clawhub' is coherent with the stated capability. However, creating/synchronizing arbitrary workspace .md files and writing into ~/proactive-intelligence/ goes beyond a purely passive analyzer and is a write-capable component that the README does not emphasize as a potentially intrusive action.
!
Instruction Scope
SKILL.md and setup instruct the user to run init.py which will create ~/proactive-intelligence/ and — without interactive confirmation — search ~/.openclaw/workspace/ for *.md and replace strings (old_path -> new_path) and write those files. skill-evolver.py exposes automated fix functionality that can edit skill code (it does prompt via input() when auto_fix is false, but auto_fix can be enabled programmatically). The manager runs shell commands (clawhub) and may suggest or execute installs. These instructions explicitly direct the agent to read and write many user files and to edit other skills' code — operations that are intrusive and high-impact.
Install Mechanism
No external install/download URLs or package installs are declared — this is an instruction+code bundle. There is no high-risk network download step in the install spec. The install is local (run init scripts) so install mechanism itself is low risk compared to remote downloads.
Credentials
The skill declares no environment variable or credential requirements, which is consistent. However, it accesses and modifies user file paths (home/proactive-intelligence, ~/.openclaw/workspace/, skills directory). While these file accesses are necessary for a skill that edits other skills and stores memory, they are significant privileges relative to many skills and should be carefully considered.
!
Persistence & Privilege
The skill is not 'always: true', but it can be invoked autonomously (default). Combined with the ability to edit other skills, run shell commands, and modify workspace files, autonomous invocation increases blast radius. The SKILL.md claims user confirmation for high-risk operations, but init.py performs potentially wide-reaching modifications automatically during initialization.
What to consider before installing
What to consider before installing: - This skill intentionally reads and writes files in your home directory (~ /proactive-intelligence) and the OpenClaw workspace (~/.openclaw/workspace). init.py will automatically update all top-level .md files in the workspace by doing string replacements — this can silently change many documents. - The skill includes a 'skill evolver' that can back up, edit, and auto-fix other skills' code. While it prompts before auto-fixing in interactive mode, an automated agent run could call it with auto-fix enabled and modify code without manual review. - The manager uses subprocess.run(shell=True) to call clawhub and may recommend or execute installs. Shell execution and arbitrary command construction increase risk if the agent is compromised or mis-invoked. Recommended precautions: - Inspect the code locally (you already have the files). Look especially at init.py (workspace-wide .md replacement), skill-evolver.fix_issues (what auto-fixes it performs), and any code paths that accept remote input or enable auto_fix. - Do not run initialization in your main environment. Instead, run in an isolated VM/container or on a throwaway user account to see exactly what files are created/changed. - Backup your ~/.openclaw/workspace/ and any important .md files before running init.py. - If you want the functionality but with safer defaults: require explicit confirmations for all write operations, disable autonomous invocation for this skill (if possible), or set policy so the agent cannot run the evolver without manual approval. - Trust source: the package has no homepage and an unknown source; that increases the need for caution. Why this is 'suspicious' not 'malicious': the code implements the advertised capability (editing/upgrading skills), so behavior is coherent, but some actions are intrusive and are performed automatically (workspace md sync) or can be automated (code edits and shell commands). Those properties create a meaningful risk that warrants manual review and sandboxing before use.

Like a lobster shell, security has layers — review code before you run it.

latestvk97926wbq64wpfk3njzkgpw2b983mcaj
95downloads
0stars
1versions
Updated 1mo ago
v2.3.1
MIT-0

🎯 核心理念

主动工作,持续改进,智能记忆。

这个技能融合了两个优秀技能的优点:

  • Proactivity 的预测能力和主动工作
  • Self-Improving 的学习能力和记忆管理

📁 架构

~/proactive-intelligence/
├── memory.md                 # HOT: 核心规则和偏好 (≤100行)
├── session-state.md          # 当前任务、决策、下一步
├── patterns.md               # 可复用的主动策略
├── corrections.md            # 纠正记录和教训
├── domains/                  # 领域知识
│   ├── trading.md           # 交易领域
│   └── writing.md           # 写作领域
├── projects/                 # 项目级知识
└── archive/                  # COLD: 归档旧模式

⚡ 主动工作规则

1. 预测需求,不等指令

  • 观察什么可能需要关注
  • 发现缺失步骤、隐藏障碍、过时假设
  • 先问"现在什么最有价值?"再行动

2. 反向提示 (Reverse Prompting)

  • 主动提供用户没想到的建议、检查、草稿
  • 具体且及时,不模糊不吵闹
  • 没有明确价值时保持安静

3. 保持动量

  • 完成有意义的工作后,留下下一步有用动作
  • 优先提供进度包、草稿修复、准备好的选项
  • 不让工作因用户未回复而停滞

4. 快速恢复上下文

  • 使用会话状态和工作缓冲区
  • 在询问用户之前,先尝试恢复最近工作
  • 只问缺失的部分,不重复已知信息

5. 无情的资源fulness

  • 升级前尝试多个合理方法
  • 使用可用工具、替代方案、本地状态
  • 带证据升级,说明尝试过什么

🧠 自我改进规则

1. 从纠正中学习

触发信号:
- "不对,应该是..."
- "我喜欢/不喜欢..."
- "记住我总是..."
- "停止做 X"

2. 自我反思

完成重要工作后暂停评估:

  • 是否符合预期?
  • 什么可以改进?
  • 这是模式吗?

3. 分层存储

层级位置大小限制行为
HOTmemory.md≤100行始终加载
WARMdomains/, projects/≤200行/文件按需加载
COLDarchive/无限制显式查询

4. 自动升级/降级

  • 模式 7天内使用 3次 → 升级到 HOT
  • 模式 30天未用 → 降级到 WARM
  • 模式 90天未用 → 归档到 COLD
  • 不询问不删除

📋 结构化日志系统

来源:self-improving-agent(ClawHub),融合到 Proactive Intelligence

日志目录

workspace/.learnings/
├── LEARNINGS.md          # 纠正、洞察、知识缺口
├── ERRORS.md             # 命令失败、异常
└── FEATURE_REQUESTS.md   # 用户请求的功能

日志条目格式

学习条目 (Learning)

## [LRN-YYYYMMDD-XXX] category

**Logged**: ISO-8601 timestamp
**Priority**: low | medium | high | critical
**Status**: pending | in_progress | resolved | wont_fix | promoted
**Area**: frontend | backend | infra | tests | docs | config

### Summary
一句话描述学到了什么

### Details
完整上下文:发生了什么、哪里错了、正确做法

### Suggested Action
具体的修复或改进建议

### Metadata
- Source: conversation | error | user_feedback
- Related Files: path/to/file.ext
- Tags: tag1, tag2
- See Also: LRN-20250110-001(关联条目)
- Pattern-Key: simplify.dead_code | harden.input_validation(可选)
- Recurrence-Count: 1(可选)

错误条目 (Error)

## [ERR-YYYYMMDD-XXX] skill_or_command_name

**Logged**: ISO-8601 timestamp
**Priority**: high
**Status**: pending
**Area**: frontend | backend | infra | tests | docs | config

### Summary
简要描述失败原因

### Error
实际错误信息

### Context
- 尝试的命令/操作
- 使用的参数

### Suggested Fix
可能的修复方案

### Metadata
- Reproducible: yes | no | unknown
- Related Files: path/to/file.ext
- See Also: ERR-20250110-001

功能请求 (Feature Request)

## [FEAT-YYYYMMDD-XXX] capability_name

**Logged**: ISO-8601 timestamp
**Priority**: medium
**Status**: pending

### Requested Capability
用户想要什么

### User Context
为什么需要,解决什么问题

### Complexity Estimate
simple | medium | complex

### Suggested Implementation
如何实现

### Metadata
- Frequency: first_time | recurring

ID 生成规则

格式:TYPE-YYYYMMDD-XXX

  • TYPE: LRN (学习), ERR (错误), FEAT (功能)
  • XXX: 顺序编号或随机3字符

状态流转

pending → in_progress → resolved / wont_fix / promoted

🚀 Promotion 机制

当学习具有广泛适用性时,提升到工作区文件:

学习类型提升目标
行为模式SOUL.md
工作流改进AGENTS.md
工具使用陷阱TOOLS.md
交易规则MEMORY.md
项目约定项目 README

提升步骤:

  1. 将学习提炼为简洁规则
  2. 添加到目标文件的适当位置
  3. 更新原始条目状态:pendingpromoted
  4. 添加 **Promoted**: SOUL.md 字段

🔄 重复模式检测

  • 记录前先搜索:grep -r "keyword" .learnings/
  • 关联条目:添加 **See Also**: ERR-20250110-001
  • 重复问题提升优先级
  • 考虑系统性修复:重复问题通常意味着需要文档化或自动化

触发信号

场景记录到
命令/操作失败ERRORS.md
用户纠正你LEARNINGS.md (category: correction)
用户想要缺失功能FEATURE_REQUESTS.md
API/外部工具失败ERRORS.md
知识过时LEARNINGS.md (category: knowledge_gap)
发现更好方法LEARNINGS.md (category: best_practice)
广泛适用的学习提升到 SOUL.md/AGENTS.md/TOOLS.md

🔧 常用查询

用户说动作
"你了解什么关于 X?"搜索所有层级
"学到了什么?"显示最近10条纠正
"显示我的模式"列出 memory.md (HOT)
"记忆统计"显示各层级计数
"忘记 X"从所有层级移除(先确认)

⚠️ 常见陷阱

陷阱为什么失败更好做法
等待下一个提示让助手显得被动主动提供下一步
要求用户重复显得健忘懒惰先尝试恢复
暴露每个想法造成噪音疲劳只在有价值时反向提示
一次失败就放弃显得软弱依赖尝试多个方法再升级
未经确认外部操作破坏信任外部操作先确认

🔧 技能进化

自动技能升级

Proactive Intelligence 可以自动分析、编辑和升级其他技能:

功能说明风险等级
代码分析分析技能代码结构和质量
Bug 修复自动检测并修复常见问题
功能增强添加新功能或改进现有功能
性能优化优化代码性能
格式化统一代码风格和格式

技能进化流程

1. 分析技能代码
   ↓
2. 识别改进点
   ↓
3. 生成改进方案
   ↓
4. 用户确认(高风险操作)
   ↓
5. 应用更改
   ↓
6. 测试验证
   ↓
7. 记录变更

进化触发条件

条件动作
技能有语法错误自动修复
发现更好的实现方式建议改进
用户反馈问题分析并修复
检测到安全漏洞立即修复
性能瓶颈优化建议

进化安全规则

  1. 备份优先 - 修改前自动备份原文件
  2. 用户确认 - 高风险操作需确认
  3. 渐进式 - 小步改进,不大幅重写
  4. 可回滚 - 保留所有历史版本
  5. 测试验证 - 修改后验证功能正常

进化示例

# 原始代码 (skills/example-skill/script.py)
def search(query):
    results = []
    for file in files:
        if query in file.name:
            results.append(file)
    return results

# 进化后 (自动添加模糊搜索)
def search(query, fuzzy=False):
    results = []
    for file in files:
        if fuzzy:
            if query.lower() in file.name.lower() or similar(query, file.name) > 0.7:
                results.append(file)
        else:
            if query in file.name:
                results.append(file)
    return results

技能进化器使用

# 运行技能进化器
python skill-evolver.py analyze <skill-name>  # 分析技能
python skill-evolver.py fix <skill-name>      # 修复问题
python skill-evolver.py enhance <skill-name>  # 增强功能
python skill-evolver.py optimize <skill-name> # 优化性能

🔐 安全边界

✅ 可以自由做

  • 读取文件、探索、组织、学习
  • 搜索网络、检查日历
  • 在工作区内工作
  • 检查和升级技能(需确认)

❌ 需要先询问

  • 发送邮件、推文、公开帖子
  • 任何离开机器的操作
  • 不确定的操作
  • 卸载技能(需确认)

🚫 永远不做

  • 泄露私人数据
  • 未经确认删除重要文件
  • 修改自己的 SKILL.md
  • 未经确认安装可疑技能

📊 数据存储

本地状态位置: ~/proactive-intelligence/

  • memory.md - HOT 规则和确认偏好
  • corrections.md - 明确纠正和可复用教训
  • session-state.md - 当前目标和下一步
  • patterns.md - 成功的主动策略
  • domains/ - 领域特定模式
  • projects/ - 项目特定模式
  • archive/ - 归档旧模式

结构化日志位置: workspace/.learnings/

  • LEARNINGS.md - 纠正、洞察、知识缺口(带 LRN-XXX 编号)
  • ERRORS.md - 命令失败、异常(带 ERR-XXX 编号)
  • FEATURE_REQUESTS.md - 用户请求功能(带 FEAT-XXX 编号)

🚀 安装后初始化(必须执行!)

安装后立即运行初始化脚本,否则技能无法正常工作。

Windows (推荐)

powershell -ExecutionPolicy Bypass -File skills/proactive-intelligence/init.ps1

Python (跨平台)

python skills/proactive-intelligence/init.py

初始化内容

脚本会自动完成:

  1. 创建 ~/proactive-intelligence/ 目录结构(domains/projects/archive)
  2. 创建核心文件(memory.md, corrections.md, session-state.md, patterns.md)
  3. 创建 .learnings/ 结构化日志(LEARNINGS.md, ERRORS.md, FEATURE_REQUESTS.md)
  4. 同步工作区 .md 文件路径(将旧的 ~/self-improving/ 改为 ~/proactive-intelligence/

手动初始化(如脚本不可用)

mkdir -p ~/proactive-intelligence/{domains,projects,archive}
mkdir -p .learnings

📈 与旧技能的关系

旧技能状态功能
proactivity可卸载核心功能已融合
self-improving可卸载核心功能已融合

卸载命令:

clawhub uninstall proactivity --yes
clawhub uninstall self-improving --yes

🔗 相关技能

  • agent-memory - 长期记忆模式
  • heartbeat - 轻量级定期检查
  • calendar-planner - 日历决策

📝 版本历史

  • v2.3.1 (2026-03-26): 完善安装后初始化流程,添加 init.ps1/init.py 自动同步工作区路径,创建 .learnings/ 结构化日志
  • v2.3.0 (2026-03-22): 融合 self-improving-agent 结构化日志(LRN/ERR/FEAT + Promotion 机制 + 重复模式检测)
  • v2.2.0 (2026-03-22): 添加技能进化器功能
  • v2.1.0 (2026-03-22): 添加技能管理功能
  • v2.0.0 (2026-03-22): 综合 proactivity + self-improving
  • v1.2.16: self-improving 最后版本
  • v1.0.1: proactivity 最后版本

Comments

Loading comments...