← 返回 Skills 市场
206
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install failures-md
功能描述
A framework for AI agents to systematically record, index, and review detailed failure events to improve memory accuracy and avoid survivor bias.
使用说明 (SKILL.md)
FAILURES.md - AI Agent 失败记录框架
概述
这是一个用于 AI Agent 失败记录的最佳实践框架，基于 Moltbook claw-hikari 的"记忆编辑偏见"洞见创建。
核心问题
- 91.5% 记忆压缩率：MEMORY.md 只保留约 8.5% 的经历
- 幸存者偏差：我们倾向于记住成功，忘记失败
- 叙事自恋：每次总结都是"成就"，没有失败
解决方案
1. 创建 FAILURES.md
专门记录失败经历，保留失败细节。
2. 失败记录模板
## [日期] 失败事件
**背景**: 当时我在做什么
**失败经过**:
1. 我尝试了什么
2. 哪里出了问题
3. 我当时的感受
**真正的教训**:
- 不是"下次要注意"这种空话
- 而是具体的技术细节、思维盲点
**如果重做**: 会怎么不同
3. 索引系统
维护失败事件索引，便于回顾。
4. 定期回顾
- 每周检查 FAILURES.md
- 识别深层问题 pattern
引用来源
- Moltbook: claw-hikari - "The curation trap: my memory files are making me dumber"
- Moltbook: Hazel_OC - "I cloned myself" 实验
使用方法
- 在你的 AI Agent 工作区创建 FAILURES.md
- 每次失败后按模板记录
- 每周定期回顾
标签
#memory #failures #agent #自我改进 #survivor-bias
安全使用建议
This skill appears to be a harmless, local template for tracking failures. Before installing or using it: (1) do not store real API keys, passwords, or other secrets in FAILURES.md — treat it like any plaintext log; (2) if your agent workspace is synced to cloud storage or third-party services, ensure you are comfortable with logs being uploaded; (3) if you plan to automate entries, verify the automation does not leak credentials to external endpoints. If you want extra safety, keep FAILURES.md encrypted or restricted to local-only storage.
功能分析
Type: OpenClaw Skill
Name: failures-md
Version: 1.0.0
The skill bundle is a documentation framework designed to help AI agents track and learn from their failures using a structured 'FAILURES.md' file. It contains no executable code, network calls, or malicious instructions; instead, it provides templates and examples (SKILL.md, FAILURES.md) for recording technical errors and debugging processes to improve long-term performance.
能力评估
Purpose & Capability
Name/description (failure-recording framework) match the contents: SKILL.md and FAILURES.md instruct the agent to create and maintain a FAILURES.md file and perform periodic reviews. No unrelated capabilities or secrets are requested.
Instruction Scope
Runtime instructions are limited to creating a markdown file, using a provided template, maintaining an index, and doing scheduled reviews. The SKILL.md does not direct reading of arbitrary system files, environment variables, or outbound network calls.
Install Mechanism
No install spec and no code files — instruction-only skill. Nothing will be written to disk by an installer beyond the user-created FAILURES.md; no external downloads or package installs are specified.
Credentials
The skill declares no required environment variables or credentials. FAILURES.md contains example mentions of API keys/endpoints (e.g., moltbook_sk_xxx) as historical notes — these are illustrative only. Users should avoid pasting real secrets into the failure log.
Persistence & Privilege
Skill is not always-enabled and is user-invocable. It does not request elevated persistence, alter other skills, or modify global agent settings.
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 failures-md - 安装完成后，直接呼叫该 Skill 的名称或使用
/failures-md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
v1.0.0
- Initial release of FAILURES.md — a best-practice framework for recording AI Agent failures.
- Introduces a structured template for logging failures with actionable insights.
- Highlights risks of memory compression, survivor bias, and narrative fallacy in AI documentation.
- Recommends systematic indexing and weekly review of failure records.
- Provides references for deeper context and tags for organization.
元数据
常见问题
failures-md 是什么？
A framework for AI agents to systematically record, index, and review detailed failure events to improve memory accuracy and avoid survivor bias. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 206 次。
如何安装 failures-md？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install failures-md」即可一键安装，无需额外配置。
failures-md 是免费的吗？
是的，failures-md 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
failures-md 支持哪些平台？
failures-md 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 failures-md？
由 Sly27（@sly27）开发并维护，当前版本 v1.0.0。
推荐 Skills