Skill Downloader
Provide a review-first workflow for discovering and installing skills from trusted sources.
Scope
Use this skill for:
- - searching for skills
- comparing candidates from one or more trusted sources
- inspecting ClawHub-hosted skills before installation
- installing or updating a skill after explicit user approval
Keep recommendations proportional to the information available. If important fields are missing, say so before making a recommendation.
Core workflow
- 1. Identify whether the user wants discovery only, inspection, or discovery plus installation.
- Search trusted sources and collect source-labeled results.
- Inspect the most relevant candidates when detailed metadata is available.
- Present the result set with clear completeness labeling:
-
Full: enough information to compare and recommend
-
Partial: some fields are unavailable
-
Minimal: only basic identification is available
- 5. Ask for explicit approval before any install or update step.
- Review source material before writing files.
- Prefer official registry workflows when available.
- Preserve user data when updating an existing skill.
Source policy
Use trusted sources first. Prefer ClawHub when the requested skill is hosted there. Use supplementary sources when needed to confirm availability, compare options, or recover from partial metadata.
If rate limits or source failures prevent a full inspection:
- - report the limitation clearly
- mark missing fields as INLINECODE0
- let the user choose whether to wait, proceed with partial data, or stop
Inspect workflow
When the user asks to inspect a specific result (for example "inspect 第一个" or "inspect 这个"):
- 1. Resolve the target skill name or slug from the previous result list.
- Run the official inspection workflow first.
- If more detail is needed, prefer reading the inspected skill's listed files or primary text files.
- Do not invent or guess unsupported CLI subcommands.
- If a field is unavailable, mark it as
unknown instead of inferring it. - Do not claim that a candidate is newer, better, equivalent to a built-in skill, or safer than another option unless the evidence explicitly supports that statement.
Inspect output template
For inspected skills, prefer this structure:
- - Name
- Slug
- Owner
- Version
- License
- Security status
- Summary
- Key capabilities
- Requirements / dependencies
- Notable files reviewed
- Unknown fields
- Recommendation status:
yes, conditional, or INLINECODE4 - Recommendation basis:
Full, Partial, or INLINECODE7
Only give a firm recommendation when the information level is Full.
Installation policy
Do not install or update anything without explicit user approval.
When installing or updating:
- - use the requested destination (
~/.openclaw/skills/ for global or {workspace}/skills/ for local by default) - prefer official registry workflows when available
- use a review-first fallback workflow only when needed
- avoid symlink-based installs
- preserve unrelated user data during updates
Safety policy
Treat this skill as review-first, not execution-first.
- - inspect source material before writing files
- prefer trusted registries and repository sources
- avoid silent assumptions when metadata is incomplete
- explain risks if a candidate appears suspicious
- avoid unsupported claims about parity, freshness, or superiority without direct evidence
References
For low-level source handling details, fallback mechanics, and implementation notes, read:
技能下载器
提供一种以审查为先的工作流程,用于从可信来源发现和安装技能。
适用范围
使用此技能进行:
- - 搜索技能
- 比较来自一个或多个可信来源的候选技能
- 在安装前检查托管于 ClawHub 的技能
- 在获得用户明确批准后安装或更新技能
建议应与可用信息相称。如果重要字段缺失,应在提出建议前说明。
核心工作流程
- 1. 确定用户仅需发现、检查,还是发现并安装。
- 搜索可信来源,并收集带有来源标签的结果。
- 在可获得详细元数据时,检查最相关的候选技能。
- 呈现结果集,并附带清晰的完整性标注:
-
完整:有足够信息进行比较和推荐
-
部分:某些字段不可用
-
最少:仅有基本标识信息
- 5. 在任何安装或更新步骤前,请求明确批准。
- 在写入文件前审查源材料。
- 优先使用官方注册表工作流程(如可用)。
- 更新现有技能时保留用户数据。
来源策略
优先使用可信来源。当请求的技能托管于 ClawHub 时,优先选择 ClawHub。在需要确认可用性、比较选项或从部分元数据中恢复时,使用补充来源。
如果速率限制或来源故障导致无法全面检查:
- - 清晰报告限制情况
- 将缺失字段标记为 unknown
- 让用户选择是等待、使用部分数据继续,还是停止
检查工作流程
当用户要求检查特定结果时(例如“检查第一个”或“检查这个”):
- 1. 从之前的结果列表中解析目标技能名称或别名。
- 首先运行官方检查工作流程。
- 如需更多细节,优先阅读被检查技能所列出的文件或主要文本文件。
- 不要编造或猜测不支持的 CLI 子命令。
- 如果某个字段不可用,将其标记为 unknown,而非推断。
- 除非有明确证据支持,否则不得声称某个候选技能比内置技能更新、更好、等效或更安全。
检查输出模板
对于被检查的技能,优先采用以下结构:
- - 名称
- 别名
- 所有者
- 版本
- 许可证
- 安全状态
- 摘要
- 关键能力
- 要求/依赖项
- 已审查的显著文件
- 未知字段
- 推荐状态:是、有条件 或 否
- 推荐依据:完整、部分 或 最少
仅在信息级别为完整时,才给出明确推荐。
安装策略
未经用户明确批准,不得安装或更新任何内容。
安装或更新时:
- - 使用请求的目标路径(默认为全局 ~/.openclaw/skills/ 或本地 {workspace}/skills/)
- 优先使用官方注册表工作流程(如可用)
- 仅在必要时使用以审查为先的回退工作流程
- 避免基于符号链接的安装
- 更新期间保留无关的用户数据
安全策略
将此技能视为审查为先,而非执行为先。
- - 在写入文件前审查源材料
- 优先使用可信注册表和仓库来源
- 元数据不完整时避免静默假设
- 如果候选技能看起来可疑,解释相关风险
- 在没有直接证据的情况下,避免对等同性、新鲜度或优越性做出无依据的声明
参考资料
有关底层源处理细节、回退机制和实现说明,请阅读:
- - references/advanced-workflows.md