← 返回 Skills 市场
312
总下载
0
收藏
2
当前安装
4
版本数
在 OpenClaw 中安装
/install skill-polisher
功能描述
当用户需要持续打磨和进化技能、追踪技能质量、收集反馈改进、检查技能规范合规性时使用。每次技能执行后自动收集用户评分，沉淀使用数据，定期分析并输出改进建议。包含技能健康度检查、最佳实践沉淀、标准化评分卡等功能。🔒 只读分析，不修改任何技能文件。
安全使用建议
This skill appears coherent: it only reads installed skills and stores feedback/analysis under ~/.openclaw/workspace/.skill-polisher/. It does not request credentials or network access. Before installing, consider: 1) confirm you are comfortable with the agent invoking the bundled scripts (they run locally and may prompt for interactive input when run without flags); 2) inspect the scripts if you want assurance that no external network calls are added later — currently there are none; 3) be aware of a functional bug: polish-suggest imports get_all_skills from health-report but that function is not defined there, so some commands may fail at runtime; 4) review tracking.json/expectation files if you want to control which skills are tracked (only tracked skills can collect feedback by default). Overall the behavior matches the description and there are no disproportionate permissions or hidden endpoints.
功能分析
Type: OpenClaw Skill
Name: skill-polisher
Version: 1.0.3
The 'skill-polisher' bundle is a meta-utility designed to track, analyze, and improve the quality of other OpenClaw skills. The scripts (e.g., check-spec.py, health-report.py, and collect-feedback.py) focus on local file analysis, regex-based specification checking, and JSON-based data storage within a dedicated workspace directory (~/.openclaw/workspace/.skill-polisher/). There is no evidence of network activity, data exfiltration, or unauthorized file modification; the bundle explicitly enforces a read-only policy regarding other skills' source code and follows established best practices for data/code separation.
能力评估
Purpose & Capability
Name/description (skill polishing, feedback, health checks) match the included scripts and SKILL.md. The code reads SKILL.md and scripts from ~/.openclaw/workspace/skills/ and writes only to ~/.openclaw/workspace/.skill-polisher/, which is appropriate for a feedback/analysis tool.
Instruction Scope
SKILL.md and scripts consistently describe read-only analysis and local storage of suggestions. The runtime instructions instruct the agent to run local Python scripts (collect-feedback, health-report, polish-suggest, etc.) against skills in the workspace. There is no code that tries to access remote endpoints or unrelated system paths. Note: collect-feedback enforces that a skill must be in the tracking list unless --force is used; polish-suggest attempts to call functions (e.g., get_all_skills from health-report) that may not exist at runtime (functional bug), so some commands may error.
Install Mechanism
No install spec — instruction-only with bundled scripts. Nothing is downloaded or written outside the declared .skill-polisher data directory and the workspace skills directory.
Credentials
No environment variables, no external credentials requested. File reads and writes are limited to the user's OpenClaw workspace (skills and a hidden .skill-polisher directory), which aligns with the stated purpose of local feedback collection and analysis.
Persistence & Privilege
always:false (default). The skill can be invoked autonomously by the agent (platform default) but does not declare forced inclusion. It writes only to its own data directory and does not attempt to modify other skills.
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 skill-polisher - 安装完成后，直接呼叫该 Skill 的名称或使用
/skill-polisher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
v1.0.3
## skill-polisher 1.0.3 Changelog
- Removed obsolete script: `scripts/polish.py`.
- Updated: Main suggestion generation now handled by `scripts/polish-suggest.py`.
- No changes to skill description or user documentation.
- This update streamlines the suggestion workflow by consolidating code improvements into a single script.
v1.0.2
- Introduced strict "read-only analysis" mode—skill-polisher now analyzes and suggests improvements but never modifies any skill file.
- Updated documentation and description to clarify that all outputs (e.g., recommendations) are saved to user space for manual review and adoption.
- Added explicit permissions metadata, specifying file system read/write boundaries and ensuring no changes to other skills.
- All polishing actions (health analysis, suggestions) now output to `.skill-polisher/polish-history/`; applying changes remains opt-in by the user.
- Refined usage guides, workflows, and warnings to stress the new non-destructive approach.
v1.0.1
Skill-polisher 1.0.1
- 大幅重构功能与架构，明确“技能打磨”需明确加入追踪列表后才启用。
- 新增 tracking.py 支持技能追踪列表的添加/移除/查询。
- 优化 collect-feedback.py, health-report.py 等脚本，仅作用于追踪列表中的技能，提升隐私与数据安全。
- 更新使用说明，强调用户数据与技能代码强隔离，杜绝隐私数据泄露到技能目录。
- 文档大幅完善，详细说明典型流程、Agent 调用方式和数据存储规范。
v1.0.0
Initial release of skill-polisher: a system for continuous skill refinement, quality tracking, and feedback-driven improvement.
- Collects user ratings and feedback automatically after each skill execution
- Aggregates usage data and analyzes it regularly to suggest improvements
- Provides health checks, best practices repository, and standard rating cards
- Offers scripts for health reporting, expectation setting, compliance checking, and skill polishing
- Includes structured storage for feedback, expectations, metrics, history, and knowledge base
元数据
常见问题
Skill Polisher 是什么？
当用户需要持续打磨和进化技能、追踪技能质量、收集反馈改进、检查技能规范合规性时使用。每次技能执行后自动收集用户评分，沉淀使用数据，定期分析并输出改进建议。包含技能健康度检查、最佳实践沉淀、标准化评分卡等功能。🔒 只读分析，不修改任何技能文件。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 312 次。
如何安装 Skill Polisher？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install skill-polisher」即可一键安装，无需额外配置。
Skill Polisher 是免费的吗？
是的，Skill Polisher 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Skill Polisher 支持哪些平台？
Skill Polisher 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 Skill Polisher？
由 Lucius.C（@luciuscao）开发并维护，当前版本 v1.0.3。
推荐 Skills