← 返回 Skills 市场
openlang-cn

Checkpoints

作者 openlang · GitHub ↗ · v1.0.0 · MIT-0
cross-platform ✓ 安全检测通过
238
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install checkpoints
功能描述
Records full context of AI-generated code and binds it to git commits. When the agent writes and commits code, capture prompts, conversation history, reasoni...
使用说明 (SKILL.md)

\r \r

Checkpoints\r

\r Checkpoints 用于自动记录 AI 生成代码的完整上下文,并与对应的**代码提交(Commit)**绑定,便于回溯、审计和复现“这段代码是怎么来的”。\r \r ---\r \r

何时使用\r

\r

  • 智能体编写并即将提交(或刚提交)代码时,需要把本次生成的上下文一并留存。\r
  • 用户明确要求“把这次改动的依据/对话记录下来”“和 commit 绑在一起”。\r
  • 需要事后查看某次提交对应的提示词、推理过程和决策逻辑时,本 Skill 指导如何生成和存放 checkpoint 数据。\r \r ---\r \r

Checkpoint 应包含的内容\r

\r 在生成或提交 AI 代码时,将以下信息与 commit 绑定:\r \r | 内容 | 说明 |\r |------|------|\r | 提示词(Prompts) | 用户或系统触发生成的那条(或几条)主要 prompt,可做脱敏。 |\r | 对话记录 | 与本次改动相关的对话摘要或关键轮次(可精简为要点)。 |\r | 推理步骤 | 模型/智能体的推理链:先做了什么判断、尝试了哪些方案、为何选当前实现。 |\r | 决策逻辑 | 关键决策点(例如:为何用 A 库而非 B、为何采用这种结构)。 |\r | 元数据 | 可选:Token 消耗量、模型/版本、时间戳、涉及文件列表等。 |\r | 关联 Commit | 明确对应的 commit hash(或 branch + 即将提交的说明),保证可追溯。 |\r \r ---\r \r

与 Commit 绑定的方式\r

\r

  • 方式一:独立 checkpoint 文件 \r 在仓库中维护 checkpoint 存储(例如 .checkpoints/docs/checkpoints/),每个 commit 对应一个文件,命名包含 commit hash 或时间戳,便于通过 hash 查找。\r
  • 方式二:Commit 消息或 tag \r 在 commit message 中写简短摘要,并注明“详细上下文见 .checkpoints/\x3Chash>.json”;或打 tag 指向该 commit,tag 名/描述中带 checkpoint 标识。\r
  • 方式三:外部系统 \r 若有现成的 Checkpoints 工具或服务,按该工具的 API/CLI 上传上述内容,并在 commit message 或仓库内留一条引用(如 URL 或 id)。\r \r 按项目约定或用户指定选择一种方式,并保持一致。\r \r ---\r \r

Checkpoint 文件结构建议\r

\r 若用文件形式保存,可采用结构化格式(便于后续解析和展示),例如:\r \r

{\r
  "commit": "abc123...",\r
  "timestamp": "ISO8601",\r
  "prompts": ["用户请求的原始或摘要 prompt"],\r
  "conversation_summary": "与本次改动相关的对话要点",\r
  "reasoning_steps": ["步骤1", "步骤2", "..."],\r
  "decisions": ["决策1及原因", "..."],\r
  "files_changed": ["path/a", "path/b"],\r
  "meta": {\r
    "model": "optional",\r
    "tokens_used": "optional",\r
    "agent_version": "optional"\r
  }\r
}\r
```\r
\r
也可用 Markdown 书写,便于人工阅读,但需固定章节(Prompts、对话摘要、推理步骤、决策、元数据、Commit)。\r
\r
---\r
\r
## 工作流程(智能体侧)\r
\r
1. **在准备提交或完成提交后**:判断本次改动是否主要由 AI 生成;若是,则生成 checkpoint 内容。\r
2. **组装内容**:从当前对话中提取或总结:提示词、相关对话、推理步骤、决策逻辑;若有 token/模型信息则填入 meta。\r
3. **绑定 commit**:若尚未提交,先执行提交,取得 commit hash;再写入 checkpoint 文件或调用外部工具,并确保文件名/引用中包含该 hash。\r
4. **不污染主流程**:checkpoint 文件可加入 `.gitignore` 若仅本地使用;若团队需要共享审计,则纳入版本库并提交。\r
\r
---\r
\r
## 使用原则\r
\r
- **一致性**:同一仓库内固定用一种绑定方式和目录/命名规则。\r
- **隐私与脱敏**:若 prompt 或对话中含敏感信息,存盘前做脱敏或仅存摘要。\r
- **可追溯**:通过 commit hash 能从 checkpoint 反查到提交,或从提交找到对应 checkpoint。\r
安全使用建议
This skill is a set of best-practice instructions (no code or installs). It is coherent with its stated goal, but before using it consider: (1) prompts and conversation history often contain secrets—never write raw prompts with credentials or personal data into a repo; sanitize or redact first. (2) Decide whether checkpoints live in-repo (auditable but permanent) or in a secured external store (requires credentials and access controls). (3) If you keep checkpoints in the repo, use team policies (.gitignore, protected branches, code review) and/or encryption for sensitive fields. (4) If you choose the external-system route, only provide minimal credentials to that service and review its privacy/persistence policies. (5) Confirm with your team/legal if recording full model context (tokens, prompts) is allowed for your project. Overall the skill itself is instruction-only and internally consistent; the main risk is accidental inclusion of sensitive data when following its guidance.
功能分析
Type: OpenClaw Skill Name: checkpoints Version: 1.0.0 The 'checkpoints' skill bundle is a set of documentation instructions (SKILL.md) designed to help an AI agent record its reasoning, prompts, and metadata alongside git commits for auditing purposes. It contains no executable code, scripts, or malicious instructions, and it explicitly recommends data desensitization for privacy.
能力评估
Purpose & Capability
The name and description (recording AI-generated context and binding it to commits) align with the SKILL.md content. No unrelated binaries, env vars, or config paths are requested.
Instruction Scope
Instructions stay within the stated purpose (extract prompts, conversation summary, reasoning, and write checkpoint files or references to commits). The guidance appropriately mentions de‑identification and using either an in-repo file, commit message/tag, or external service. Note: collecting conversation/prompt history can capture sensitive data—the doc recommends but does not enforce sanitization before storage.
Install Mechanism
No install spec or code is present (instruction-only). No downloads or archive extraction are required.
Credentials
The skill requests no environment variables or credentials. However, the 'external system' binding option implies that if you implement that route you will need to supply service credentials externally—this is not specified in the skill and should be provisioned only when necessary and with appropriate access controls.
Persistence & Privilege
always is false and there is no install-time persistence. The skill does not request elevated privileges or modify other skills' configs. Be aware the agent platform allows autonomous invocation by default; that platform setting is separate from this skill and not changed by it.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install checkpoints
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /checkpoints 触发
  4. 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 the "checkpoints" skill. - Captures and stores full context (prompts, conversation, reasoning, decisions, metadata) related to AI-generated code changes. - Binds context data to specific git commits for traceability and auditing. - Supports multiple binding methods: standalone files, commit messages/tags, or external tools. - Provides structured guidelines for checkpoint content and workflow integration. - Emphasizes consistency, privacy, and traceability in checkpoint handling.
元数据
Slug checkpoints
版本 1.0.0
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

Checkpoints 是什么?

Records full context of AI-generated code and binds it to git commits. When the agent writes and commits code, capture prompts, conversation history, reasoni... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 238 次。

如何安装 Checkpoints?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install checkpoints」即可一键安装,无需额外配置。

Checkpoints 是免费的吗?

是的,Checkpoints 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

Checkpoints 支持哪些平台?

Checkpoints 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 Checkpoints?

由 openlang(@openlang-cn)开发并维护,当前版本 v1.0.0。

💬 留言讨论