← 返回 Skills 市场
自我改进智能体
作者
tigertamvip
· GitHub ↗
· v0.1.0
· MIT-0
106
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install self-improving-agent-5
功能描述
Captures learnings, errors, and corrections to enable continuous improvement. Use when: (1) A command or operation fails unexpectedly, (2) User corrects Clau...
安全使用建议
This skill is coherent for collecting and promoting learnings, but before installing or enabling hooks: (1) Inspect the referenced GitHub repo and any hook scripts you copy to ~/.openclaw/hooks to ensure they don't perform unexpected actions; (2) Understand what your OpenClaw 'sessions_*' tools can access — sessions_history and sessions_send could expose or forward other sessions' transcripts; (3) Back up any workspace files that might be overwritten and consider limiting the directories the skill touches to a project-specific .learnings if you don't want cross-session/global effects; (4) Only enable the hook if you trust the skill and its source.
功能分析
Type: OpenClaw Skill
Name: self-improving-agent-5
Version: 0.1.0
The skill bundle facilitates an agent's self-modification by instructing it to log 'learnings' and promote them to core behavioral and workflow files (e.g., SOUL.md, AGENTS.md, and CLAUDE.md), which effectively act as system prompts. It also encourages the use of high-risk tools to access transcripts from other sessions (sessions_history) and references external shell scripts for hook-based execution (activator.sh, error-detector.sh) that are not included in the provided files. While these capabilities align with the stated goal of 'self-improvement,' they introduce significant risks regarding persistent prompt injection and unauthorized cross-session data access.
能力评估
Purpose & Capability
Name/description match the instructions: all actions are about collecting and promoting learnings/errors/feature-requests into .learnings and workspace docs. No unrelated environment variables, binaries, or surprising external services are requested.
Instruction Scope
Instructions direct the agent (and the user) to create and append to files under ~/.openclaw/workspace/.learnings and potentially promote entries to other workspace files (SOUL.md, AGENTS.md, TOOLS.md). They also reference session-level tools (sessions_list, sessions_history, sessions_send, sessions_spawn) — these are coherent for cross-session sharing but have privacy implications because they can read or forward session transcripts if available. The scope is appropriate for the stated purpose but you should confirm what 'sessions_history' and 'sessions_send' can access in your environment before enabling automatic behavior.
Install Mechanism
No install spec in the skill bundle (instruction-only). The README suggests optional manual git clone from a GitHub repo and a ClawdHub install command; these are typical and optional. There is no automatic download/extract or opaque URL in the skill itself.
Credentials
The skill declares no required environment variables, credentials, or config paths. The file-write/read operations are limited to OpenClaw workspace paths (.learnings and other workspace files) which are proportionate to the logging/promotion purpose.
Persistence & Privilege
always:false and model invocation is allowed (default). The skill recommends enabling a hook in ~/.openclaw/hooks to run reminders at session start; enabling that would give persistent behavior across sessions and could influence agent prompts. This is expected for an auto-notification skill but is a persistent change the user must opt into.
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 self-improving-agent-5 - 安装完成后，直接呼叫该 Skill 的名称或使用
/self-improving-agent-5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
v0.1.0
Initial release of the self-improvement skill for agents—log and promote errors, learnings, and feature requests for continuous improvement.
- Introduces structured logging for errors, learnings, and feature requests in markdown files under `.learnings/`
- Provides detailed templates for each log entry type, including priority, context, suggested actions, and metadata
- Describes workflows for promoting broadly applicable learnings to project memory files like `CLAUDE.md`, `AGENTS.md`, and others
- Includes guidance for setup with OpenClaw, as well as generic agent compatibility
- Offers examples of entry formats and guidance for resolving or promoting learnings
元数据
常见问题
自我改进智能体 是什么？
Captures learnings, errors, and corrections to enable continuous improvement. Use when: (1) A command or operation fails unexpectedly, (2) User corrects Clau... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 106 次。
如何安装 自我改进智能体？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install self-improving-agent-5」即可一键安装，无需额外配置。
自我改进智能体 是免费的吗？
是的，自我改进智能体 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
自我改进智能体 支持哪些平台？
自我改进智能体 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 自我改进智能体？
由 tigertamvip（@tigertamvip）开发并维护，当前版本 v0.1.0。
推荐 Skills