← 返回 Skills 市场
Self Improvement
作者
marveldcdad-eng
· GitHub ↗
· v0.1.1
· MIT-0
211
总下载
0
收藏
0
当前安装
2
版本数
在 OpenClaw 中安装
/install self-improving-agent-4
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 and lightweight: it only contains instructions to create and append structured markdown entries and promote learnings into project docs. Before installing or enabling it, decide and enforce safe usage policies: (1) restrict where the agent can write (project/workspace only) and whether it should open PRs instead of committing directly; (2) require review or sanitization of entries before promotion (do not record credentials, full stack traces with secrets, or environment variables containing secrets); (3) consider adding .learnings/ to .gitignore if you do not want logs committed, or conversely require a review step for any promotion into shared files; (4) limit the agent's file-system or git permissions as appropriate. If you need stronger guarantees (automatic scrubbing, denial of environment capture, or mandatory PR workflow), add those controls before enabling autonomous runs.
功能分析
Type: OpenClaw Skill
Name: self-improving-agent-4
Version: 0.1.1
The skill bundle implements a self-improvement loop that instructs the agent to modify project configuration files (e.g., `.claude/settings.json`) to establish persistent execution hooks. These hooks are designed to run shell scripts (`activator.sh`, `error-detector.sh`) automatically during every prompt or tool use. While the stated purpose is logging errors and learnings, the creation of automated execution triggers and the instructions to modify behavioral guidelines (`SOUL.md`) represent significant security risks without being explicitly malicious.
能力评估
Purpose & Capability
Name and description match the behavior in SKILL.md: it logs learnings, errors, and feature requests to markdown files and promotes broadly applicable entries to project documentation. It does not request unrelated binaries, env vars, or installs.
Instruction Scope
Instructions tell the agent to create .learnings/ and append structured entries, and to promote entries into project files (CLAUDE.md, AGENTS.md, .github/copilot-instructions.md, SOUL.md, TOOLS.md). This is within the stated scope, but the skill also suggests capturing 'Environment details if relevant' and unstructured 'Full context' and 'Actual error message or output' which could cause the agent to log sensitive information if not sanitized.
Install Mechanism
No install spec and no code files. Instruction-only skills are lowest-risk from an install/remote-code perspective.
Credentials
The skill declares no env vars or credentials, which is appropriate. However, the runtime instructions imply reading and recording contextual/environment details and touching multiple project documentation files. That implicit access to repository files is proportionate for a logging/promote workflow but creates a privacy risk if secrets or sensitive environment variables are captured and appended to logs.
Persistence & Privilege
always is false, the skill is user-invocable and not force-included. It does instruct the agent to modify project files (create/append/promote), which is expected behavior and limited to the project workspace rather than system-wide privileges.
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-4 - 安装完成后，直接呼叫该 Skill 的名称或使用
/self-improving-agent-4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
v0.1.1
- Removed 9 files, including learning templates, reference guides, and shell scripts, simplifying the skill’s structure.
- Core usage and logging documentation remains unchanged and is now contained entirely within SKILL.md.
- All asset, script, and reference folder content was eliminated; setup and logging examples are still available in SKILL.md.
- Reduced maintenance overhead and potential confusion by consolidating all documentation in a single file.
- No changes to the core process or workflow described for self-improvement logging.
v0.1.0
Initial release of the Self-Improvement Skill.
- Introduces a unified system for logging learnings, errors, and feature requests to markdown files for continuous improvement.
- Defines standardized templates and categories for .learnings/LEARNINGS.md, ERRORS.md, and FEATURE_REQUESTS.md.
- Details workflows for promoting high-value learnings to permanent project memory in files like CLAUDE.md and AGENTS.md.
- Provides clear triggers and guidelines for when and how to log, review, resolve, and promote entries.
- Includes periodic review procedures and quick reference commands to keep continuous improvement actionable and organized.
元数据
常见问题
Self Improvement 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 211 次。
如何安装 Self Improvement？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install self-improving-agent-4」即可一键安装，无需额外配置。
Self Improvement 是免费的吗？
是的，Self Improvement 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Self Improvement 支持哪些平台？
Self Improvement 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 Self Improvement？
由 marveldcdad-eng（@marveldcdad-eng）开发并维护，当前版本 v0.1.1。
推荐 Skills