菜单
Meta-cognitive self-learning system - Automated skill evolution based on predictive coding and value-driven mechanisms.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Fetch, categorize, and summarize GitHub Trending projects across daily, weekly, and monthly spans. Use when a user asks for "trending projects", "latest hot repos", or a "summary of GitHub trends" to provide a structured, categorised report with full hyperlinking. Supports Markdown, HTML (Pinterest design), or both output formats, with optional Obsidian vault storage.
Turn a local folder or GitHub repository into an interactive browser-based course that explains how the codebase works for non-expert programmers and AI-assisted builders. Use when a user asks to make a course, tutorial, walkthrough, learning guide, codebase explanation, or interactive lesson from a project.
Converts a PRD or broad technical requirement into an AI-executable agile workflow with cards, checklists, branch-diff mapping, review gates, and a final comparison report. Use when the user wants PRD review, agile card decomposition, AI execution tracking, implementation branch review, or PRD-to-code traceability.
Two complementary Bezos heuristics in one skill. (A) Two-Way Door — classify decisions by reversibility; reversible = decide fast with 70% info, irreversible = decide slow with 90% info. (B) Regret Minimization — for life-defining choices, project to age 80 and pick what minimizes regret. Use Two-Way Door triggers on Chinese 决策瘫痪 / 反复纠结小事 / 开了三次会还没定 / 大事小事一样慢; Regret Min triggers on 离职 / 创业 / 移民 / 结婚 / 生育 / 转行 / 重大决定 / 这辈子. Especially when team applies identical heavy process to all decisions (need Two-Way), or when user faces once-in-a-decade pivot where rational analysis ties (need Regret Min). Do NOT misclassify One-Way as Two-Way (most expensive mistake), use Regret Min for daily decisions (age-80 view on "what to eat" is meaningless), or run Regret Min in heated emotion (cool 24-48h first to avoid romanticizing risk).
Use BEFORE selecting any other decision framework — Cynefin (kuh-NEV-in) classifies WHICH framework fits the current situation across 5 domains (Clear / Complicated / Complex / Chaotic / Confused). Triggers when user is about to apply a method and the fit feels off, asks meta-questions like "should we follow SOP or explore?", or when same approach that worked last time seems wrong now. Also use after failure when "the method was right but the situation didn't match", or when team argues over deterministic-plan-vs-experimentation. Especially valuable at the start of major decisions to avoid using the wrong hammer for the nail. Do NOT use for trivial classified decisions (don't run Cynefin for "what to eat for lunch"), true Chaotic situations needing immediate stabilizing action (act first, classify later), or teams unfamiliar with the model (use simpler known/unknown binary).
Use when user reports overwhelm from too many tasks, asks for week/sprint/OKR planning help, or describes spending 'all day firefighting' while long-term projects stall. Triggers on Chinese phrases 太多事 / 不知道先做哪个 / 排不过来 / 周计划 / sprint planning / 救火 / 优先级 / overwhelm / 焦虑 / 做不完 / todo 列表炸了, and explicit task triage requests. Especially valuable when waiting list has 15+ pending items, or when user says "重要的事一直推不动 / 都在做紧急的事". Do NOT use for small lists (<5 items just sort by deadline), team-level responsibility assignment (use RAPID/DACI from backlog), or when importance needs quantitative weighting (use weighted decision matrix).
| name | Self-Evolving Skill |
| description | Meta-cognitive self-learning system - Automated skill evolution based on predictive coding and value-driven mechanisms. |
| homepage | https://github.com/whtoo/self-evolving-bot |
元认知自学习系统 - 基于预测编码和价值驱动的Skill自动演化。
# 技能已安装到 ~/.openclaw/skills/self-evolving-skill
# 或使用ClawHub
clawhub install self-evolving-skill
self-evolving-skill/
├── core/ # Python核心
│ ├── residual_pyramid.py # 残差金字塔（SVD分解）
│ ├── reflection_trigger.py # 自适应触发器
│ ├── experience_replay.py # 经验回放缓存
│ ├── skill_engine.py # 核心引擎+ValueGate
│ ├── storage.py # 持久化
│ └── mcp_server.py # MCP服务器
├── src/ # TypeScript SDK
│ ├── index.ts # 主入口
│ ├── cli.ts # CLI
│ └── mcp-tools.ts # 工具定义
├── skills/ # OpenClaw Skill
│ └── self-evolving-skill/ # 技能封装
├── MCP_CONFIG.md # MCP配置
└── README.md # 文档
| 工具 | 描述 | 参数 |
|---|---|---|
skill_create | 创建Skill | name, description |
skill_execute | 执行并学习 | skill_id, context, success, value |
skill_analyze | 分析嵌入 | embedding |
skill_list | 列出Skills | - |
skill_stats | 系统统计 | - |
skill_save | 持久化保存 | skill_id |
skill_load | 加载 | skill_id |
# 列出所有Skill
openclaw skill self-evolving-skill list
# 创建Skill
openclaw skill self-evolving-skill create --name "MySkill"
# 执行
openclaw skill self-evolving-skill execute <id> --success
# 分析
openclaw skill self-evolving-skill analyze --embedding '[0.1,0.2,...]'
# 统计
openclaw skill self-evolving-skill stats
# 启动MCP服务器
cd ~/.openclaw/skills/self-evolving-skill
./run_mcp.sh
# 或使用适配器
python3 mcporter_adapter.py skill_list '{}'
import { SelfEvolvingSkillEngine } from 'self-evolving-skill';
const engine = new SelfEvolvingSkillEngine();
await engine.init();
const { skillId } = await engine.createSkill({ name: 'Analyzer' });
const stats = await engine.stats();
pyramid = ResidualPyramid(max_layers=5, use_pca=True)
decomposition = pyramid.decompose(embedding)
# 输出:
# - residual_ratio: 残差能量比率
# - suggested_abstraction: POLICY / SUB_SKILL / PREDICATE
# - novelty_score: 综合新颖性
| 覆盖率 | 抽象层级 | 操作 |
|---|---|---|
| >80% | POLICY | 调整策略权重 |
| 40-80% | SUB_SKILL | 生成子Skill |
| <40% | PREDICATE | 归纳新谓词 |
trigger = ReflectionTrigger(
min_energy_ratio=0.10, # 初始阈值
value_gain_threshold=0.20, # 触发阈值
target_trigger_rate=0.15 # 目标15%触发率
)
| 路径 | 说明 |
|---|---|
~/.openclaw/skills/self-evolving-skill | 技能根目录 |
~/.openclaw/mcp_servers/self-evolving-skill.json | MCP服务器配置 |
~/.openclaw/workspace/self-evolving-skill/storage | 数据存储 |