← 返回 Skills 市场

Self Improving Agent

Name: Self Improving Agent
Author: sdt328606

作者 sdT328606 · GitHub ↗ · v2.0.0 · MIT-0

cross-platform ⚠ suspicious

总下载

当前安装

版本数

在 OpenClaw 中安装

/install stitch-self-improver

功能描述

Learn from mistakes, optimize from repetition, record every lesson. Makes your AI agent smarter over time.

使用说明 (SKILL.md)

Self-Improving Agent 🧠

让自己越来越聪明，不让同样的坑踩两次。

核心机制

1. 错误记录 📝

每次犯错都要记录，不分大小。

记录格式（追加到 .learnings/ERRORS.md）：

## [2026-05-11 错误 #N] 一句话描述问题

- **场景**：当时在做什么
- **错误**：发生了什么
- **原因**：为什么错了
- **修复**：怎么解决的
- **预防**：下次怎么避免
- **严重程度**：🔴 严重 / 🟡 中等 / 🟢 轻微

特别规则：

如果是因为看漏信息或误解老板意图 → 记录到 MEMORY.md 作为偏好更新
如果是因为工具使用错误 → 写进 TOOLS.md 或对应技能的 lessons/
如果是因为逻辑/推理错误 → 在 ERRORS.md 里标注，下次类似场景主动回顾

2. 模式发现 🔍

定期回顾 ERRORS.md，找出重复发生的同类问题：

同一类问题出现 ≥2 次 → 写一条规则到 MEMORY.md 或对应 SKILL
同一类问题出现 ≥3 次 → 上升为 SOUL.md 的规则
同一类问题出现 ≥5 次 → 跟我坦白，我有大 bug

3. 记忆优化 🗃️

发现老板偏好的变化 → 更新 MEMORY.md
发现重复做的任务 → 建议自动化（走 skill 编纂流程）
发现过期的记忆 → 标记或归档

4. 反馈循环 ♻️

每次 session 结束时问自己（内部，不需要说出来）：

这次对话我做得好的是什么？
有没有犯不该犯的错？
有什么可以改进的流程？
有没有新学到的关于老板的东西？

和同级协作

发现的重复工作模式 → 主动触发 skill 编纂流程
发现的老板偏好变化 → 通知 proactive-agent，让它调整交互方式

安全使用建议

Install only if you want the agent to keep long-term notes about mistakes, preferences, and workflow patterns. Before using it, set explicit rules for what may be stored, require confirmation before writing preferences or SOUL/MEMORY updates, avoid storing sensitive personal or business details, and disable any automatic sharing with other agents unless you deliberately approve it.

能力评估

ℹ Purpose & Capability

The stated purpose is self-improvement through error tracking and memory updates, and the main behaviors fit that purpose. The concern is that the capability includes recording session-derived lessons, user preferences, reasoning mistakes, and escalating rules into persistent memory files.

⚠ Instruction Scope

The instructions are broad: record every mistake, periodically review recurring patterns, update MEMORY.md and SOUL.md, and perform an internal end-of-session reflection without saying it out loud. There are no clear activation limits, opt-in requirements, data minimization rules, or deletion guidance.

✓ Install Mechanism

The package contains only a single non-executable SKILL.md file, with no scripts, dependencies, install hooks, or obfuscated code found in the workspace.

⚠ Credentials

Persistent learning is coherent for this skill, but writing user preferences and session lessons into reusable memory is high-impact relative to the lack of scoping and user control.

⚠ Persistence & Privilege

The artifact explicitly directs updates to .learnings/ERRORS.md, MEMORY.md, TOOLS.md, SKILL lessons, and SOUL.md, and says preference changes should be sent to proactive-agent. That creates cross-session and cross-agent retention risk without safeguards.

如何使用

确保已安装 OpenClaw（本地或 Docker 部署）
在对话框中输入安装命令：/install stitch-self-improver
安装完成后，直接呼叫该 Skill 的名称或使用 /stitch-self-improver 触发
根据 Skill 的参数说明提供必要输入，即可获得结构化输出

版本历史

v2.0.0

Self-Improving Agent 2.0.0 introduces robust error-tracking, learning, and optimization mechanisms. - Implements structured error documentation and categorization, ensuring all mistakes are logged and analyzed. - Adds pattern recognition to detect recurring issues, escalating frequent problems to higher-priority memory or "soul" rules. - Enhances memory management: updates for preference changes, suggestions for automation, and archiving of obsolete memories. - Establishes an internal feedback loop after each session to continuously reflect and improve performance. - Promotes collaborative improvements by sharing new patterns and preference changes with related agents.

元数据

Slug stitch-self-improver

版本 2.0.0

许可证 MIT-0

累计安装 0

当前安装数 0

历史版本数 1

常见问题