Self Improving Agent Cn.Skip

AI自我改进与记忆系统 - 解决'同类错误反复犯、用户纠正不长记性'的痛点。自动捕获错误、用户纠正、最佳实践，并转化为长期记忆。

Security Scan

The skill only reads/writes local JSONL memory files and asks for no credentials, but the documentation promises automatic triggers and cross‑project syncing that are not implemented in the shipped code — an inconsistency you should understand before installing.

Purpose & Capability

The stated purpose is a local self-improvement/memory store for the agent, which matches the included scripts that append/read JSONL files under ~/.openclaw/memory/self-improving. However the SKILL.md repeatedly describes "自动触发" (automatic capture) and "自动同步" to project files (e.g., .learnings, AGENTS.md, MEMORY.md) that the provided scripts do not implement. The capability to automatically hook into shell commands or sync to other locations is claimed but not present in code.

Instruction Scope

Runtime instructions in SKILL.md instruct the agent to auto-record command failures, user corrections, and to check memories before executing commands. The shipped Python scripts only provide manual CLI entrypoints to append/read entries and do not include any listener/daemon, webhooks, or integration code that would automatically monitor commands or perform cross‑project writes. The doc's examples imply the agent will autonomously modify behavior (e.g., '自动改用 sudo'), but the scripts never execute or alter commands — they only write/read local files. This mismatch could lead to false expectations about what the skill will actually do; it also means an integrating agent would need additional glue that could introduce risk.

Install Mechanism

No install spec and no external downloads; the skill is instruction-plus-local Python scripts only. That minimizes install-time risk — nothing will be fetched from remote hosts during install.

Credentials

The skill requests no environment variables or credentials and only accesses files under the user's home (~/.openclaw/memory/self-improving). That access is proportionate to a local memory store. Note: memory files may contain sensitive information if the agent records raw command text, errors, or user corrections; there is no encryption or access control in the code.

Persistence & Privilege

The skill does not request always:true, does not change other skills' configs, and only writes to per-user paths. It persists data in user home but does not elevate privileges or force installation into all agents.

What to consider before installing

What to consider before installing: - Expectation vs reality: the SKILL.md promises automatic capture/sync behavior, but the included scripts are simple CLI tools that only append/read JSONL files; there is no daemon, shell hook, or sync implementation. If you rely on "自动触发", ask the author how that is implemented or review any integration glue the agent will use. - Sensitive data: memories may include command lines, errors, or user corrections (potentially secrets). The code stores data in plaintext under ~/.openclaw; consider whether you want that on disk, and enforce filtering/desensitization or encryption if needed. - Behavior testing: run the scripts in a sandboxed account and inspect created files (errors.jsonl, corrections.jsonl, best_practices.jsonl) before enabling any agent automation that would call them. - Integration risk: if you or your agent adds automation to call these scripts on every command failure, ensure the automation does not inadvertently capture tokens/passwords or automatically run corrective commands (the current code does not execute fixes, but an integration might). - Ask the author for details about the claimed sync to project files and automatic triggers; absent that, treat the skill as a manual local memory store only.

Self-Improving Agent

让AI从错误中学习，越用越聪明。

核心解决的问题

✅ 命令/操作莫名失败，下次还用同样的错方式
✅ 反复纠正AI的写法、偏好、风格，它下个会话又忘了
✅ 同一个项目里反复踩同一个坑
✅ 发现更好的做法/更优解法，却没有系统化记住
✅ 外部工具/API变动，AI还在用旧知识
✅ 重要经验只存在于当前会话，跨天/跨项目就丢失

安装

# 创建记忆目录
mkdir -p ~/.openclaw/memory/self-improving

# 使用本skill
cat ~/.openclaw/skills/self-improving-agent/SKILL.md

使用方法

1. 记录错误（自动）

当命令失败时，自动记录到错误库：

# 脚本会自动捕获并记录
python3 ~/.openclaw/skills/self-improving-agent/log_error.py \
  --command "npm install xxx" \
  --error "permission denied" \
  --fix "use sudo or check permissions"

2. 记录用户纠正

当用户说：

• "不对，应该..."
• "错了，要用..."
• "不对，我之前说过..."

自动记录：

python3 ~/.openclaw/skills/self-improving-agent/log_correction.py \
  --topic "代码风格" \
  --wrong "用了双引号" \
  --correct "项目要求单引号" \
  --context "AGENTS.md 第23行"

3. 记录最佳实践

发现更好的做法时：

python3 ~/.openclaw/skills/self-improving-agent/log_best_practice.py \
  --category "security" \
  --practice "安装skill前必须审计代码" \
  --reason "防止供应链投毒"

4. 查看记忆（执行前自动检查）

# 执行命令前，先检查是否有相关记忆
python3 ~/.openclaw/skills/self-improving-agent/check_memory.py \
  --command "npm install"

输出示例：

⚠️ 发现相关记忆:
  [错误] 2026-03-05: npm install 权限失败 → 改用 sudo 或在用户目录安装
  [纠正] 2026-03-04: 项目使用单引号而非双引号

文件结构

~/.openclaw/memory/self-improving/
├── errors.jsonl          # 错误记录
├── corrections.jsonl     # 用户纠正
├── best_practices.jsonl  # 最佳实践
├── knowledge_gaps.jsonl  # 知识盲区
└── index.json           # 快速索引

自动触发场景

场景	自动记录	触发词
命令失败	✅ errors	退出码非0
用户纠正	✅ corrections	"不对" "错了" "应该"
更好的方法	✅ best_practices	"更好的" "更高效" "最优"
知识过时	✅ knowledge_gaps	"过时了" "已废弃" "新版是"

跨项目同步

重要记忆自动同步到：

• ~/.openclaw/memory/self-improving/ (全局)
• 当前项目 .learnings/ (项目级)
• AGENTS.md (用户偏好)
• MEMORY.md (长期记忆)

与现有Skill对比

功能	原self-improvement	本skill
记录错误	✅	✅
记录纠正	✅	✅ 增强版
自动触发	❌ 手动	✅ 自动
跨项目记忆	❌ 项目级	✅ 全局+项目
执行前检查	❌	✅
知识过时检测	❌	✅

示例工作流程

场景1：用户纠正代码风格

用户: "不对，我说过项目里要用单引号"

AI动作:

1. 自动记录纠正到 corrections.jsonl
2. 更新 AGENTS.md 添加代码风格规则
3. 回复: "已记录，下次会使用单引号"

场景2：命令反复失败

AI: 执行 npm install -g xxx 结果: 权限失败

AI动作:

1. 记录错误到 errors.jsonl
2. 下次执行前检查记忆
3. 自动改用: sudo npm install -g xxx 或本地安装

场景3：发现更好做法

用户: "其实用 pip install -e . 更高效"

AI动作:

1. 记录最佳实践到 best_practices.jsonl
2. 后续安装Python包时优先使用 editable 模式

注意事项

• 记忆文件定期备份到git
• 敏感信息脱敏后记录
• 定期review并清理过时记忆

---

Created: 2026-03-05 by 老二