Skill flagged — suspicious patterns detected

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

QClaw Self-Evolver

v1.0.1

AI Agent 自我进化引擎。基于 OpenSpace + HermesAgent 理念,实现技能质量监控、学习记录自动处理、工作流模式提炼、自动创建新技能。触发场景: (1) 用户说「升级/完善/优化」相关 → 触发技能自进化 (2) 用户想了解「还记得什么/搜一下记忆」→ 触发语义记忆搜索 (3) 用户想「...

0· 69·0 current·0 all-time
by理想菇@qshan1

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for qshan1/qclaw-self-evolver.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "QClaw Self-Evolver" (qshan1/qclaw-self-evolver) from ClawHub.
Skill page: https://clawhub.ai/qshan1/qclaw-self-evolver
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 qclaw-self-evolver

ClawHub CLI

Package manager switcher

npx clawhub@latest install qclaw-self-evolver
Security Scan
VirusTotalVirusTotal
Suspicious
View report →
OpenClawOpenClaw
Benign
high confidence
Purpose & Capability
Name/description (self‑evolver for agent learning and generating skills) matches the included scripts and runtime actions: creating a workspace .learnings/, recording corrections, measuring skill metrics, running SEA (Sense/Assess/Evolve), and generating skill candidates. No unrelated env vars or external services are required.
Instruction Scope
SKILL.md and the scripts instruct the agent to create and write files under the workspace (LEARNINGS.md, ERRORS.md, pending.md, skill_metrics.json, skills/...), register a cron job (via openclaw.cron if available), and auto-generate SKILL.md files for new skills. These behaviors are consistent with the stated functionality but do grant the skill write access to a persistent workspace and the ability to create new skill descriptors automatically — review generated skills before trusting them.
Install Mechanism
No network downloads or package installs; code is included in the skill bundle and the install script copies local files into ~/.qclaw/workspace. This is a low-risk install mechanism (local file writes only).
Credentials
No credentials or sensitive environment variables are requested. The only environment usage is an optional QW_WORKSPACE path override (defaults to ~/.qclaw/workspace). No attempts to read system-wide config or secrets were detected.
Persistence & Privilege
The skill registers a recurring task (cron every 3 days) and writes persistent files to the user's workspace. always:true is not set. Persistent presence via cron and on-disk artifacts is expected for this functionality but increases persistence and should be considered before installation.
Assessment
This skill appears internally consistent and performs only local operations (no networking or credential access). Before installing: (1) back up any existing ~/.qclaw/workspace data you care about; (2) inspect the generated files under ~/.qclaw/workspace/.learnings and ~/.qclaw/workspace/skills to confirm generated SKILL.md contents are safe before using them as active skills; (3) if you don't want automatic cron tasks, skip the cron registration step or register manually; (4) consider running the scripts in a sandboxed environment first. If you need higher assurance, request code signatures or a trusted upstream source for this skill package.

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

latestvk97cd8hn4p7a0qexavzmtmg1qn84rr6e
69downloads
0stars
2versions
Updated 2w ago
v1.0.1
MIT-0

Self Evolver — 自我进化引擎

版本: 1.0.0 适用对象: QClaw 等 AI Agent 依赖: Python 3.8+, 无需外部库


核心能力

帮助 AI Agent 建立「从经验中持续自我进化」的能力,核心三大模块:

模块做什么触发方式
SEA 进化循环从错误/纠正中提炼改进,3天周期扫描cron
纠正自动记录被用户纠正时自动记录到学习档案关键词检测
技能自进化从任务历史自动发现可固化的技能关键词检测

触发条件

当用户说「帮我安装自我进化」或「安装 self-evolver」或「你要有自我进化能力」时,触发本 skill。

安装后 AI 会:

  1. 创建 .learnings/ 目录结构
  2. 注册每 3 天一次的 SEA 进化扫描 cron
  3. 激活纠正自动记录机制

使用方法

1. 安装(首次)

skillhub_install install_skill self-evolver

或在 OpenClaw 中对 AI 说:

「安装自我进化能力」

AI 会自动完成:

  • 创建 .learnings/ 目录(含 LEARNINGS.md / ERRORS.md / FEATURE_REQUESTS.md)
  • 注册 cron:每3天运行 SEA 进化扫描
  • 生成 SKILL.md 覆盖本文件

2. 日常使用

场景操作
被纠正了AI 自动检测纠正关键词 → 写入 pending → 下次 heartbeat 合并
想要手动进化说「自我反思」「进化一下」「帮我分析最近的错误」
查看学习记录说「我最近学到了什么」「看一下记忆」
新技能候选出现AI 自动检测关键词 → 生成候选技能写入 skills/

3. 手动触发进化

python ~/.qclaw/workspace/scripts/evolve.py

核心脚本说明

evolve.py(SEA 进化引擎)

核心流程:

Sense(感知)→ Assess(评估)→ Evolve(进化)

Sense(感知):

  • 扫描 7 天内所有学习记录(LEARNINGS.md / ERRORS.md)
  • 识别高优先级(Priority=高)未处理项
  • 识别连续重复错误(出现 ≥3 次)

Assess(评估):

  • 技能健康度诊断(fallback_rate / completion_rate / effective_rate)
  • 行为模式识别(从对话历史提取)
  • 学习效率评分

Evolve(进化):

  • 对连续错误 → 更新系统提示词 / AGENTS.md
  • 对重复任务 → 触发技能自进化生成新 skill
  • 生成进化报告

auto_learn.py(纠正自动记录)

触发关键词:

  • 打断/纠正:不對、错了、不是、等等、重新、更正、纠正
  • 质疑:不對、我没说过、不用、不需要

工作流:

  1. 检测到纠正 → 立即写入 pending.md
  2. 下次 heartbeat → flush 到 LEARNINGS.md
  3. 高优先级纠正 → 立即触发 SEA 扫描

skill_evolution.py(技能自进化)

检测关键词(任务后分析):

  • 研究、开发、分析、优化、自动化、监控、集成、部署、测试、重构

生成规则:

  • 同一模式出现 ≥3 次 → 生成候选技能
  • 生成 skills/[pattern]-skill/SKILL.md
  • cron 定期 Nudge 提醒优化

目录结构

~/.qclaw/workspace/
├── .learnings/
│   ├── LEARNINGS.md          # 纠正与改进记录
│   ├── ERRORS.md             # 命令/操作失败记录
│   ├── FEATURE_REQUESTS.md   # 用户请求的功能
│   ├── pending.md            # 待合并的纠正(心跳时处理)
│   └── assessment-report.json # 技能健康度报告
├── scripts/
│   ├── evolve.py             # SEA 进化引擎
│   ├── auto_learn.py         # 纠正自动记录
│   ├── skill_evolution.py    # 技能自进化
│   └── assess.py             # 技能质量评估
├── skills/                   # 自生成的技能目录
│   └── [pattern]-skill/
│       └── SKILL.md
└── skill_metrics.json        # 技能使用指标

记录格式(LEARNINGS.md)

| 时间 | 错误做法 | 正确做法 | 区域 | Priority |

进化报告格式

## 进化扫描报告 — YYYY-MM-DD

### Sense(感知)
- 待处理纠正:N 条
- 连续错误:N 条
- 高优先级:N 条

### Assess(评估)
- 总技能数:N
- 健康技能:N
- 需进化技能:N

### Evolve(进化)
- 已生成技能:N
- 已更新规则:N
- 待确认改进:N

安装检查清单

  • .learnings/ 目录存在
  • scripts/evolve.py 存在
  • scripts/auto_learn.py 存在
  • scripts/skill_evolution.py 存在
  • cron 任务已注册(每3天)
  • skill_metrics.json 可写

与 QClaw 原版的区别

本 skill 是给新 Agent 的完整版,包含 QClaw 经过多次迭代后的所有最佳实践。新 Agent 安装后即可获得:

  1. SEA 三阶段进化循环
  2. 纠正即学习(自动记录 + 合并)
  3. 技能自生长(从任务历史自动生成)
  4. 三层记忆(HOT/WARM/COLD)
  5. 技能健康度监控

无需额外配置,开箱即用。

Comments

Loading comments...