Skill flagged — suspicious patterns detected

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

Self Evolution Pro

v1.0.0

增强型自我进化技能,集成自动技能提取、根因分析、知识图谱、跨会话同步、自动晋级机制。触发词:'总结这个经验'、'保存为技能'、'自我进化'、'学习这个'、'记录教训'。相比原版self-improving-agent,新增自动提取、多维度分析、进化追踪功能。

0· 253·0 current·0 all-time

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for windy-001-crypto/self-evolution-pro.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "Self Evolution Pro" (windy-001-crypto/self-evolution-pro) from ClawHub.
Skill page: https://clawhub.ai/windy-001-crypto/self-evolution-pro
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 self-evolution-pro

ClawHub CLI

Package manager switcher

npx clawhub@latest install self-evolution-pro
Security Scan
VirusTotalVirusTotal
Suspicious
View report →
OpenClawOpenClaw
Suspicious
medium confidence
Purpose & Capability
Name/description (self-improvement, skill extraction, cross-session sync) aligns with what the SKILL.md and scripts do: creating a ~/.openclaw/workspace, storing learnings, detecting recurrence, extracting skills and optionally publishing them. The behavior requested (read session history, create skills, schedule reviews) matches the stated purpose.
!
Instruction Scope
SKILL.md instructs the agent to read and write local files under ~/.openclaw/workspace, access other sessions via sessions_list/sessions_history/sessions_send/sessions_spawn, schedule cron jobs via cron_add, and run included scripts (extract.sh, review.sh). Those actions go beyond passive note-taking: they allow automated cross-session reading, spawning background subagents, and automated publishing. The spec also encourages automatic extraction/publishing when recurrence thresholds are met, which could cause sensitive conversation content to be promoted or shared without explicit user approval.
Install Mechanism
No install spec is declared (instruction-only), which is lower risk, but three executable shell scripts are bundled and intended to be run. The scripts create files under the user's home directory and call external CLI commands (clawhub). Because code is present, running these scripts writes to disk and can trigger network activity (clawhub publish). There is no external download or obfuscated installer, but presence of runnable scripts increases runtime risk vs. pure-documentation skills.
!
Credentials
The skill declares no required environment variables or primary credential, yet the extract script can call `clawhub publish` and SKILL.md shows use of sessions_* and cron_add platform APIs. Publishing to ClawHub or using platform session APIs typically requires authenticated credentials or platform capabilities; the skill does not declare or justify credential needs. More importantly, automatic extraction/publishing could expose sensitive conversation content or secrets recorded in .learnings to an external registry. The number and sensitivity of accessible data sources (local workspace files and other sessions) is high relative to the simple description.
Persistence & Privilege
always:false (good). The skill suggests scheduling recurring reviews with cron_add and spawning subagents via sessions_spawn, which grants it the ability to cause recurring or background activity if invoked or if the agent calls it autonomously. Autonomous invocation is the platform default; combined with publishing capability this increases blast radius. The skill does not request to change other skills' configs or system-wide settings.
What to consider before installing
This skill appears to do what it says (collect learnings, extract skills, sync across sessions), but it carries data-exposure and automation risks you should consider before installing: - It writes into ~/.openclaw/workspace and will read those files and your session histories. Inspect those files for any sensitive content you wouldn't want stored or published. - The extractor can call `clawhub publish` (network publish). If run, that could push learned content — potentially containing secrets — to ClawHub. Verify how ClawHub publishing is authenticated and require explicit consent before publishing; do not assume this is safe for confidential data. - The SKILL.md instructs spawning subagents and scheduling cron jobs. If the agent invokes this autonomously, it could run background tasks. Limit the agent's ability to run the skill autonomously or review scheduled jobs/cron entries the skill creates. - The skill does not declare required credentials but expects platform APIs/CLI (sessions_* functions, clawhub). Confirm what credentials/platform permissions are used and avoid supplying broad tokens unless you trust the code. Recommendations before enabling: - Review the three bundled scripts line-by-line and run them in a sandbox or VM first. - Disable automatic publishing (do not run extract.sh --publish) until you confirm outputs are safe. - Audit existing .learnings and session history for secrets; consider redaction policies before use. - Require manual approval for any publish/auto-extract actions and restrict agent autonomy for this skill. Given these ambiguities and the potential for unintended data exfiltration or background actions, treat this skill as suspicious until you validate the publish/auth flows and restrict its autonomy.

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

agentvk97ccn5rsgnjw4xd5wnm0eqdfs83mt6tevolutionvk97ccn5rsgnjw4xd5wnm0eqdfs83mt6tlatestvk97ccn5rsgnjw4xd5wnm0eqdfs83mt6tlearningvk97ccn5rsgnjw4xd5wnm0eqdfs83mt6tmemoryvk97ccn5rsgnjw4xd5wnm0eqdfs83mt6tself-improvementvk97ccn5rsgnjw4xd5wnm0eqdfs83mt6t
253downloads
0stars
1versions
Updated 1mo ago
v1.0.0
MIT-0

Self Evolution Pro

增强型自我进化技能,让AI代理持续自我改进、自我学习。

核心升级(相比原版)

特性原版增强版
技能提取手动自动 + 一键
跨会话同步基础完整知识图谱
自动晋级基于复发次数自动晋级
根因分析发现模式 → 找根本原因
进化追踪版本历史 + 效果追踪
计划审查手动Cron自动化

触发词

当以下情况时激活:

  • 用户说"总结这个经验"、"保存为技能"
  • "学习这个"、"自我进化"、"记录这个"
  • 发现一个非显而易见的解决方案
  • 重复犯同一个错误超过2次
  • 解决了一个需要调查的问题

工作原理

会话中遇到问题/纠正
        ↓
记录到 .learnings/
        ↓
    ┌───┴───┐
 错误   纠正   发现
    ↓     ↓     ↓
根因分析 链接图谱 自动晋级
    ↓
技能提取 → 发布到 ClawHub
    ↓
跨会话同步 → 其他代理也能用

文件结构

~/.openclaw/workspace/
├── AGENTS.md              # 多代理工作流
├── SOUL.md               # 行为准则
├── TOOLS.md              # 工具能力
├── MEMORY.md             # 长期记忆
├── memory/               # 日常记忆
│   └── YYYY-MM-DD.md
├── .learnings/           # 本技能日志
│   ├── LEARNINGS.md      # 学习记录
│   ├── ERRORS.md         # 错误记录
│   ├── FEATURE_REQUESTS.md # 需求记录
│   └── KNOWLEDGE_GRAPH.md # 知识图谱(新增)
├── .skills/              # 提取的技能
│   └── <skill-name>/
└── .evolution/          # 进化追踪(新增)
    ├── metrics.md        # 效果指标
    ├── review-schedule.md # 审查计划
    └── version-history.md # 版本历史

快速参考

情况操作
命令/操作失败记录到 .learnings/ERRORS.md
用户纠正你记录到 LEARNINGS.md,类别=correction
发现更好方案记录到 LEARNINGS.md,类别=best_practice
根因分析分析根本原因,链接到 KNOWLEDGE_GRAPH.md
复发≥3次自动晋级到对应文件
技能提取使用 extract 命令
跨会话同步使用 sync 命令
计划审查查看 .evolution/review-schedule.md

自动晋级规则

当满足以下条件时自动晋级

目标文件条件
SOUL.md行为模式类学习,复发≥2次
TOOLS.md工具相关,复发≥2次
AGENTS.md工作流相关,已解决
CLAUDE.md项目约定,复发≥3次
技能提取跨3+个不同任务复发

根因分析流程

遇到错误时,不只是记录,要分析:

1. 直接原因(表象)
   → "文件权限不够"

2. 根本原因(深层)
   → "没检查当前用户权限配置"

3. 模式识别
   → "每次涉及系统配置都容易忽略权限"

4. 系统性预防
   → 创建技能或添加到 SOUL.md

格式:

## [RCA-YYYYMMDD-XXX] 问题标题

**Root Cause**: 根本原因描述
**Pattern**: 识别到的模式
**Prevention**: 系统性预防措施
**Files**: 相关文件
**Skills**: 相关技能

### Why-Tree
- Why 1: 原因A
  - Why 2: 原因B
    - Why 3: 根本原因 ←

知识图谱

.learnings/KNOWLEDGE_GRAPH.md 链接相关学习:

# 知识图谱

## 节点
| ID | 类型 | 标题 | 关联 |
|----|------|------|------|
| N001 | error | Docker权限问题 | N002, N003 |
| N002 | learning | M1 Docker平台问题 | N001 |
| N003 | skill | docker-m1-fixes | N001 |

## 关系
- N001 → causes → N002
- N001 → solved_by → N003

进化指标

.evolution/metrics.md 追踪效果:

# 进化指标

## 记录统计
- 本周新增:5条
- 已解决:3条
- 已晋级:2条
- 技能提取:1个

## 效果追踪
| 学习 | 记录日期 | 复发次数 | 节省估计 |
|------|----------|----------|----------|
| Docker M1修复 | 2025-01-15 | 0 | ~30分钟/次 |
| pnpm优先 | 2025-01-18 | 2 | ~5分钟/次 |

计划审查

.evolution/review-schedule.md 安排定期审查:

# 审查计划

## 每日 (Heartbeat时)
- 检查高优先级待处理项
- 检查新复发的模式

## 每周
- 完整审查所有pending项
- 识别可晋级项
- 更新知识图谱

## 每月
- 技能版本更新
- 效果指标复盘
- 清理过时项

技能提取流程

方式1:自动提取(推荐)

当检测到复发≥3次,自动触发:

# 自动执行
./skills/self-evolution-pro/scripts/extract.sh skill-name

方式2:手动提取

触发:"保存为技能" / "这个可以提取"
操作:
1. 创建 .skills/<skill-name>/SKILL.md
2. 填写模板
3. 发布到 ClawHub(可选)
4. 更新知识图谱
5. 记录到 version-history.md

方式3:跨会话提取

场景:在会话A发现问题,在会话B需要同样知识
操作:
1. 在会话A:记录 + 标记为 shared
2. 在会话B:使用 sessions_history 读取
3. 提取到共享位置

跨会话同步

使用 OpenClaw 的会话工具:

发送学习到其他会话

sessions_send({
  sessionKey: "session:project-alpha-daily",
  message: "新学习:Docker M1平台问题解决方案已记录到 .learnings/ERRORS.md"
})

读取其他会话的学习

// 查看最近会话
sessions_list({ activeMinutes: 60, messageLimit: 3 })

// 读取特定会话历史
sessions_history({ sessionKey: "session-id", limit: 50 })

Spawn子代理做背景研究

sessions_spawn({
  task: "研究这个错误并提出系统性解决方案",
  label: "error-research",
  runtime: "subagent",
  mode: "run"
})

日志格式

学习记录

## [LRN-YYYYMMDD-XXX] category

**Logged**: ISO-8601
**Priority**: low | medium | high | critical
**Status**: pending | in_progress | resolved | promoted | promoted_to_skill
**Area**: frontend | backend | infra | tests | docs | config
**Recurrence-Count**: 1
**First-Seen**: YYYY-MM-DD
**Last-Seen**: YYYY-MM-DD

### Summary
一句话描述学到了什么

### Root Cause(新增)
根本原因分析(如果是错误)

### Solution
具体解决方案

### Pattern(新增)
如果复发,识别到的模式

### Suggested Action
具体修复或改进建议

### Metadata
- Source: conversation | error | user_feedback | self_discovered
- Related: N001, N002(知识图谱节点)
- See Also: LRN-YYYYMMDD-YYY
- Estimated Time Saved: X minutes(估算节省时间)

错误记录

## [ERR-YYYYMMDD-XXX] skill_or_command

**Logged**: ISO-8601
**Priority**: high | critical
**Status**: pending | resolved | wont_fix
**Root Cause Analysis**: [RCA ID 如果已分析]
**Area**: ...

### Summary
简短描述什么失败了

### Error

实际错误信息


### Context
- 尝试的命令/操作
- 使用的输入或参数
- 环境详情

### RCA
**Direct Cause**: 直接原因
**Root Cause**: 根本原因
**Pattern**: 识别到的模式

### Resolution
- **Resolved**: ISO-8601
- **Method**: how it was fixed
- **Prevention**: 如何预防再次发生

Cron 自动化

设置定期自我审查:

// 每周一早上审查学习
cron_add({
  name: "self-review",
  schedule: { kind: "cron", expr: "0 9 * * 1" },
  payload: {
    kind: "agentTurn",
    message: "执行 .learnings/ 审查:1) 高优先级pending项 2) 识别可晋级项 3) 更新知识图谱"
  },
  delivery: { mode: "announce" }
})

晋级决策树

学习可以晋级吗?
        ↓
是否项目特定?
├── 是 → 留在 .learnings/
└── 否 → 是行为/风格相关?
    ├── 是 → 晋级到 SOUL.md
    └── 否 → 是工具相关?
        ├── 是 → 晋级到 TOOLS.md
        └── 否 → 是工作流相关?
            ├── 是 → 晋级到 AGENTS.md
            └── 否 → 是项目约定?
                ├── 是 → 晋级到 CLAUDE.md/AGENTS.md
                └── 否 → 考虑技能提取

效果追踪

记录每个学习/技能节省的时间:

### Time Tracking
- First Occurrence: YYYY-MM-DD
- Estimated Time per Incident: 15 minutes
- Recurrence Count: 5
- Total Time Saved (if resolved): 75 minutes
- ROI: 本技能投资回报率

发布技能到 ClawHub

# 1. 提取技能后
cd ~/.openclaw/workspace

# 2. 发布
clawhub publish .skills/<skill-name> --version 1.0.0

# 3. 更新版本历史
./scripts/update-version-history.sh <skill-name>

触发器检测

自动检测以下信号:

纠正 → 学习(correction类别)

  • "不,那是错的..."
  • "实际上应该是..."
  • "你说的不对..."

功能需求 → 功能请求

  • "你能也做...吗"
  • "我希望你能..."
  • "有办法...吗"

知识差距 → 学习(knowledge_gap类别)

  • 用户提供了你不知道的信息
  • 参考的文档已过时

错误 → 错误记录

  • 命令返回非零退出码
  • 异常或堆栈跟踪
  • 意外输出或行为

发现更好方案 → 学习(best_practice类别)

  • 改进最初方案
  • 发现更高效的方法

最佳实践

  1. 立即记录 - 上下文最清晰的时候
  2. 包含根本原因 - 不只是表象
  3. 具体解决方案 - 未来需要能直接使用
  4. 追踪复发 - 用 Recurrence-Count
  5. 更新知识图谱 - 链接相关项
  6. 追踪时间节省 - 量化价值
  7. 积极晋级 - 存疑时优先晋级
  8. 定期审查 - 设置cron自动化
  9. 发布分享 - 有价值的发布到ClawHub

与原版 self-improving-agent 的区别

  1. 知识图谱 - 新增 KNOWLEDGE_GRAPH.md 链接相关学习
  2. 根因分析 - RCA 模板,分析根本原因
  3. 进化指标 - metrics.md 追踪节省的时间
  4. 版本历史 - version-history.md 记录技能演进
  5. 计划审查 - review-schedule.md + Cron 自动化
  6. 自动晋级 - 明确的自动晋级规则
  7. 跨会话同步 - 更完整的同步机制
  8. 时间追踪 - Estimated Time Saved 字段

Comments

Loading comments...