# Recipe: Small Skill Contribution English | [中文](#recipe小型-skill-贡献) MobileClaw skills should be narrow, inspectable, and easy to disable. A good skill does one thing well and makes its data flow obvious. ## A Good Skill - Has one clear purpose. - Accepts explicit inputs. - Returns structured, understandable output. - Avoids hidden network requests. - Avoids broad file, shell, or device privileges unless the user explicitly asks. - Fails with a useful error message. - Can remain on-demand instead of always injected. ## Contribution Checklist - Skill name: - Problem solved: - Inputs: - Output: - External network requests: - File access: - Device permissions: - Model calls: - Failure behavior: - Example prompt: - Example result: - Why this should be built in, listed in a market, or kept as a local custom skill: ## Review Questions - What data can the skill read? - What data can it send out? - Can the user understand what it will do before running it? - What happens on timeout, malformed input, or missing permissions? - Does it need to be promoted, or can it stay on-demand? ## Example Prompt ```text Create a skill that fetches a JSON endpoint and extracts the top-level keys. ``` Expected review notes: - The skill should only access the URL provided by the user. - It should use a timeout. - It should not log the full response if it may contain secrets. - It should return a concise list of keys and a clear error on invalid JSON. --- # Recipe:小型 Skill 贡献 [English](#recipe-small-skill-contribution) | 中文 MobileClaw Skill 应该保持窄范围、可检查、容易禁用。好的 Skill 只做好一件事,并让数据流清楚可见。 ## 好的 Skill - 有一个明确目的。 - 接收显式输入。 - 返回结构化、可理解的输出。 - 避免隐藏网络请求。 - 除非用户明确需要,否则避免宽泛的文件、Shell 或设备权限。 - 失败时返回有用错误。 - 可以保持按需加载,而不是默认常驻注入。 ## 贡献清单 - Skill 名称: - 解决的问题: - 输入: - 输出: - 外部网络请求: - 文件访问: - 设备权限: - 模型调用: - 失败行为: - 示例提示词: - 示例结果: - 为什么它应该内置、进入市场,或只作为本地自定义 Skill: ## 审查问题 - 这个 Skill 可以读取什么数据? - 它可以向外发送什么数据? - 用户运行前能否理解它会做什么? - 超时、输入格式错误或缺少权限时会发生什么? - 它需要提升为常驻能力,还是可以保持按需加载? ## 示例提示词 ```text Create a skill that fetches a JSON endpoint and extracts the top-level keys. ``` 预期审查要点: - Skill 只应访问用户提供的 URL。 - 应设置超时。 - 如果响应可能包含密钥,不应记录完整响应。 - 应返回简洁 key 列表,并在 JSON 无效时给出清楚错误。