📦 飞鸿

Name: 飞鸿
Rating: 18

v1.0.0

记录学习心得、错误及修正，以实现持续改进。适用于：(1) 命令或操作意外失败，(2) 用户纠正 Clau...

18· 18·0 当前·0 累计

by @sunyue1977

命令行工具教育学习即时通讯

下载技能包项目主页

安全扫描

VirusTotal

无害

查看报告

OpenClaw

安全

high confidence

该技能的代码和说明与其既定目的（捕获和推广学习成果）一致；它不含隐藏的网络安装或凭据请求，但你仍需审查并选择是否启用提供的 hooks，因为某些脚本会在被显式调用时写入文件。

评估建议

This skill appears to do what it says: lightweight reminders and local logging of learnings. Before enabling hooks or running scripts: (1) review the included scripts (activator.sh, error-detector.sh, extract-skill.sh) to ensure you’re comfortable with local filesystem writes, (2) enable hooks only in the scope you intend (project vs user-level) to avoid noisy global runs, (3) prefer the extract-skill.sh --dry-run before creating files, and (4) back up workspace files (CLAUDE.md, AGENTS.md, .lea...

详细分析 ▾

✓ 用途与能力

Name/description match the included files: documentation, templates, and small helper scripts for reminders, error detection, and skill extraction. The files operate on local workspace files (.learnings/, AGENTS.md, etc.) which is exactly what a 'self-improvement' skill would need.

ℹ 指令范围

SKILL.md stays within scope: it instructs creating .learnings/, copying templates, enabling optional hooks, and manually promoting learnings to workspace files. The runtime hooks read CLAUDE_TOOL_OUTPUT and print reminders; they do not exfiltrate data. Note: promoting learnings or running the extract script will create/modify files in your workspace — that is intended but worth awareness before enabling hooks globally.

✓ 安装机制

No install spec / no remote downloads embedded in the package. Manual install suggestions reference a GitHub repo (a normal optional source). All included scripts run locally; the only script that writes files is extract-skill.sh and it restricts absolute/.. paths and offers --dry-run.

✓ 凭证需求

The skill requests no credentials or env vars. The error-detector script reads CLAUDE_TOOL_OUTPUT (a platform-provided variable) which is consistent with its purpose and is declared in the docs. No unrelated secrets or external service tokens are requested.

ℹ 持久化与权限

always:false and user-invocable (normal). Hooks described are opt-in and run with the session/user permissions. Because hooks can be enabled globally (~/.claude or ~/.openclaw), enabling them gives the scripts recurring execution in your sessions — review and enable deliberately.

安全有层次，运行前请审查代码。

运行时依赖

无特殊依赖

版本

latestv1.0.0

1. 重新生成客户端：`pnpm run generate:api`

● 无害

安装命令

点击复制

官方npx clawhub@latest install feihong-self-improving-agent

镜像加速npx clawhub@latest install feihong-self-improving-agent --registry https://cn.longxiaskill.com

技能文档

自我提升技能将学习与错误记录到 markdown 文件，持续改进。后续编码代理可据此修复，重要学习将晋升至项目记忆。

速查情境 → 动作命令/操作失败 → 记入 .learnings/ERRORS.md 用户纠正 → 记入 .learnings/LEARNINGS.md，类别 correction 用户需要缺失功能 → 记入 .learnings/FEATURE_REQUESTS.md API/外部工具失败 → 记入 .learnings/ERRORS.md，附集成详情知识过时 → 记入 .learnings/LEARNINGS.md，类别 knowledge_gap 发现更优方案 → 记入 .learnings/LEARNINGS.md，类别 best_practice 简化/加固重复模式 → 更新 .learnings/LEARNINGS.md，Source: simplify-and-harden，并给出稳定 Pattern-Key 类似已有记录 → 用 See Also 链接，考虑提升优先级广泛适用的学习 → 晋升至 CLAUDE.md、AGENTS.md 和/或 .github/copilot-instructions.md 工作流改进 → 晋升至 AGENTS.md（OpenClaw workspace）工具陷阱 → 晋升至 TOOLS.md（OpenClaw workspace）行为模式 → 晋升至 SOUL.md（OpenClaw workspace）

OpenClaw 设置（推荐） OpenClaw 是本技能主平台，采用 workspace 提示注入，自动加载技能。

安装 ClawdHub（推荐）：clawdhub install self-improving-agent 手动：git clone https://github.com/peterskoett/self-improving-agent.git ~/.openclaw/skills/self-improving-agent 基于原仓库为 openclaw 重制：https://github.com/pskoett/pskoett-ai-skills

Workspace 结构 OpenClaw 每会话注入： ~/.openclaw/workspace/ ├── AGENTS.md # 多代理工作流、委托模式 ├── SOUL.md # 行为准则、个性、原则 ├── TOOLS.md # 工具能力、集成陷阱 ├── MEMORY.md # 长期记忆（仅主会话） ├── memory/ # 每日记忆文件 │ └── YYYY-MM-DD.md └── .learnings/ # 本技能日志 ├── LEARNINGS.md ├── ERRORS.md └── FEATURE_REQUESTS.md

创建学习文件

mkdir -p ~/.openclaw/workspace/.learnings

复制模板或新建： LEARNINGS.md — 纠正、知识缺口、最佳实践 ERRORS.md — 命令失败、异常 FEATURE_REQUESTS.md — 用户所需能力

晋升目标学习类型 → 晋升至 → 示例行为模式 → SOUL.md → “保持简洁，免免责声明” 工作流改进 → AGENTS.md → “长任务派生子代理” 工具陷阱 → TOOLS.md → “Git push 需先配置认证”

跨会话通信 OpenClaw 提供： sessions_list — 查看活动/近期会话 sessions_history — 读取他会话记录 sessions_send — 向他会话发送学习 sessions_spawn — 派生子代理后台运行

可选：启用钩子会话启动自动提醒：

cp -r hooks/openclaw ~/.openclaw/hooks/self-improvement
openclaw hooks enable self-improvement

详见 references/openclaw-integration.md

通用设置（其他代理） Claude Code、Codex、Copilot 等，在项目内建 .learnings/：

mkdir -p .learnings

复制模板或新建文件并加头。在 AGENTS.md、CLAUDE.md 或 .github/copilot-instructions.md 中添加引用，提醒自己记录。

自我提升工作流出错或被纠正 → 记入 .learnings/ERRORS.md、LEARNINGS.md 或 FEATURE_REQUESTS.md 审查并晋升广泛适用的学习至： CLAUDE.md — 项目事实与约定 AGENTS.md — 工作流与自动化 .github/copilot-instructions.md — Copilot 上下文

记录格式学习条目追加到 .learnings/LEARNINGS.md：

## [LRN-YYYYMMDD-XXX] category
Logged: ISO-8601 时间戳  
Priority: low | medium | high | critical  
Status: pending  
Area: frontend | backend | infra | tests | docs | config  
Summary  
一句话总结所学  
Details  
完整背景：原状、问题、正解  
Suggested Action  
具体修复或改进  
Metadata  
Source: conversation | error | user_feedback  
Related Files: path/to/file.ext  
Tags: tag1, tag2  
See Also: LRN-20250110-001  
Pattern-Key: simplify.dead_code | harden.input_validation  
Recurrence-Count: 1  
First-Seen: 2025-01-15  
Last-Seen: 2025-01-15

错误条目追加到 .learnings/ERRORS.md：

## [ERR-YYYYMMDD-XXX] skill_or_command_name
Logged: ISO-8601 时间戳  
Priority:

数据来源：ClawHub ↗ · 中文优化：龙虾技能库