← Back to Skills Marketplace
lgx281227231

Skill Evolution Loop

by LGX281227231 · GitHub ↗ · v2.1.0 · MIT-0
cross-platform ⚠ suspicious
33
Downloads
1
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install skill-evolution-loop
Description
三件套闭环引擎 v2 — 女娲+达尔文+workflow-engine 自动联动。 4大安全机制:白名单排除/人工确认/任务验证/淘汰机制。
README (SKILL.md)

三件套闭环引擎 v2(小狗版)

架构

凌晨3:00 cron自动启动(--auto模式)
    ↓
① 感知:扫描session logs,只认AI实际执行过tool的任务
    ↓
② 白名单排除:已有技能覆盖的领域直接跳过
    ↓
③ 写候选报告:不直接蒸馏,等绪哥确认
    ↓
绪哥确认后 → python3 engine.py run
    ↓
④ 蒸馏 → ⑤ 9维评分 → ⑥ 编排工作流
    ↓
⑦ 淘汰:30天未触发的自动技能标记删除

4大安全机制

机制 说明
白名单排除 邮箱/信息图/PDF等20+领域已有专业技能,不重复蒸馏
人工确认 cron只写候选报告,手动run才真正蒸馏
任务验证 只认AI执行过terminal/file/browser的对话,忽略纯聊天
淘汰机制 30天未触发的自动技能标记删除,被引用的保留

🚫 Cron Job 保护机制

绝对禁止修改 references/protected-cron-jobs.md 中列出的 cron job(包括 prompt、name、skills)。 在 detect/run/gc 任何阶段遇到受保护 job → 直接跳过,不列入候选报告。 受保护 job 出现故障 → 只通知绪哥,不自动修改。

当前受保护列表:2484808d514f(AI日报)、667d07607cf2(技能日报)

命令

# 扫描(只看不做)
python3 engine.py detect

# 自动模式(cron用,只写候选报告)
python3 engine.py run --auto

# 手动模式(真正蒸馏)
python3 engine.py run

# 淘汰检查
python3 engine.py gc --dry-run   # 预览
python3 engine.py gc             # 执行

# 查看状态
python3 engine.py status

9维评分

触发词覆盖 / 流程清晰度 / 降级策略 / 描述质量 / 示例完整性 / 错误处理 / 文档结构 / 可测试性 / 复用性

低于7分自动优化。

SKILL.md优化技巧(达尔文实战经验)

7.1→9.8分的关键改动:

  1. 添加 triggers 字段 — 10+个触发词直接拉满 trigger_coverage 10分
  2. description 改为 | 多行格式 — 单行字符串不被regex识别,导致 description 0分
  3. 确保有 triggers 关键字reusability 维度检查是否存在 triggers 字段
# ❌ 错误写法(description会得0分)
description: "单行字符串..."

# ✅ 正确写法
description: |
  多行描述...
  适用场景:...
triggers:
  - "触发词1"
  - "触发词2"

评分杠杆: triggers(2分/个) + description(1分/10字符) + fallback关键词(2分/个) + headers(2分/个)

多平台同步

每次修改SKILL.md后必须4平台同步:

  1. 🐶 小狗本地 ~/.hermes/skills/devops/workflow-engine/
  2. 🐙 GitHub git push origin main
  3. 🐾 ClawHub clawhub publish --version X.Y.Z --slug \x3Cname>
  4. 🦞 龙虾 scp [email protected]:...

参考文件

  • references/whitelist-domains.md — 白名单领域配置(20+领域映射到已有技能)
  • references/clawhub-competitive-intel.md — ClawHub workflow类竞品分析

Cron

每天凌晨 3:00 自动运行 --auto 模式。Job ID: 19bb84bf59f0

Usage Guidance
Review before installing. Use only in an environment where the agent is allowed to read local Hermes session logs and modify `~/.hermes` skills/workflows. Prefer `detect`, `run --auto`, and `gc --dry-run` first; do not let an agent run `gc`, `git push`, `clawhub publish`, or `scp` from this skill unless you explicitly approve the exact targets and changes. VirusTotal and static scan were clean; the Review verdict comes from the artifact's own high-impact automation and confirmation gaps, not malware telemetry.
Capability Assessment
Purpose & Capability
The stated purpose is a self-evolution loop for skills, and the code coherently implements session-log detection, skill generation, scoring, workflow creation, and garbage collection. Those capabilities are high-impact because they change the user's installed skill/workflow environment.
Instruction Scope
The docs disclose manual and cron modes, but `engine.py run` does not require an approved candidate record before distilling from logs, `gc` performs deletion by default when not passed `--dry-run`, and `SKILL.md` includes broad triggers such as `gc` plus mandatory GitHub/ClawHub/scp sync instructions.
Install Mechanism
No installer, dependency fetch, or code that creates a cron job was found in the artifact. The skill text claims a daily cron job, so users should verify their environment rather than assume installation is passive.
Credentials
The engine reads `~/.hermes/sessions`, writes under `~/.hermes/skills` and `~/.hermes/workflow-engine/workflows`, and the skill instructions mention publishing and copying to external platforms/hosts. This is broader than a repo-local helper.
Persistence & Privilege
The skill persists generated skills, workflow YAML, state, logs, reports, and candidates. It has no privilege-escalation code, but the garbage-collection path uses recursive directory deletion for auto-generated skills without an execution-time confirmation prompt.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install skill-evolution-loop
  3. After installation, invoke the skill by name or use /skill-evolution-loop
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v2.1.0
skill-evolution-loop v2.1.0 - 全面升级“三件套闭环引擎”,实现女娲+达尔文+workflow-engine 无缝自动联动。 - 新增4大安全机制:白名单排除、人工确认、任务验证、淘汰机制。 - 增强Cron Job保护,受保护任务严禁自动更改,异常只通知人工。 - 详细规范评分指标和SKILL.md优化技巧,助力技能进化与复用。 - 明确多平台(本地、Github、ClawHub、龙虾)同步流程,每次优化后必须同步。
Metadata
Slug skill-evolution-loop
Version 2.1.0
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 1
Frequently Asked Questions

What is Skill Evolution Loop?

三件套闭环引擎 v2 — 女娲+达尔文+workflow-engine 自动联动。 4大安全机制:白名单排除/人工确认/任务验证/淘汰机制。 It is an AI Agent Skill for Claude Code / OpenClaw, with 33 downloads so far.

How do I install Skill Evolution Loop?

Run "/install skill-evolution-loop" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Skill Evolution Loop free?

Yes, Skill Evolution Loop is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does Skill Evolution Loop support?

Skill Evolution Loop is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Skill Evolution Loop?

It is built and maintained by LGX281227231 (@lgx281227231); the current version is v2.1.0.

💬 Comments