安全扫描
OpenClaw
可疑
medium confidence该技能的能力与其描述大致匹配(技能管理),但包含强大的文件和设置修改操作(hook生成、settings.json编辑、触发器注入),并未声明一些运行时依赖和潜在副作用——安装前请审查。
评估建议
该工具包似乎实现了其宣称的功能,但执行了高影响操作(移动技能目录、写入hook脚本到~/.claude/hooks、编辑~/.claude/settings.json)。在使用或运行 `/skill-kit trigger compile` 或升级之前:1) 检查您维护的 SKILL.md 文件中的任何 'triggers' 条目;2) 确保必需的 CLI 工具(特别是 jq、grep、sed、bash)存在;jq 用于 JSON 编辑但未被声明;3) 首先以 dry-run 模式运行编译命令以预览生成的脚本和设置更改;4) 备份 ~/.claude/settings.json 和您的 ~/.claude/skills/ 目录;5) 对任何具有 action: inject 或 block 的触发器付予特别注意;6) 如果计划允许升级自动修改 SKILL.md(依赖自动检测),确认工具将在进行编辑之前询问用户;注意文档中关于所需更改与用户确认更改的矛盾语言——更好地手动批准修改。如果您不愿授予工具注册 hook 和编辑 settings.json 的能力,请在审查生成的输出之前不要运行...详细分析 ▾
✓ 用途与能力
The name/description (skill management, writer/lint/merge/trigger/upgrade/dedup/etc.) align with the actions described in the files: scanning ~/.claude/skills, creating SKILL.md, merging topics, and compiling triggers. The presence of a trigger compiler and utilities for converting agents into skills is coherent with the stated purpose.
⚠ 指令范围
Runtime instructions and helper scripts read and modify many user files and system config: they read ~/.claude/agents, ~/.claude/skills, plugin paths, move directories to .bak, create files under ~/.claude/hooks and ~/.claude/data, and overwrite ~/.claude/settings.json entries. That behavior is powerful and outside a small helper's sandbox; while it matches the declared purpose, it can change system-wide settings and register hook scripts that can block commands or inject systemMessage JSON—these effects deserve careful review and explicit user confirmation before running.
ℹ 安装机制
This is instruction-only with one included script (scripts/trigger-compile.sh) — no install spec. That's lower risk than arbitrary remote downloads. However, the script depends on common CLI utilities (jq, grep, sed, jq-based JSON operations) which are not documented in the skill metadata. Missing declared runtime dependencies is a usability/security concern.
✓ 凭证需求
The skill requests no environment variables, credentials, or config paths in metadata. The documented operations act only on local files under the user's home and project directories (e.g., ~/.claude/*, .claude/*). This is proportionate to a skill-management tool.
⚠ 持久化与权限
The trigger compiler writes executable hook scripts into ~/.claude/hooks and edits ~/.claude/settings.json to register hooks. Those hooks can run commands, output injection JSON, or block tool use (inject/block/suggest). That is high privilege and can affect all agent sessions; while it's part of the skill's purpose, users should be aware and verify generated hooks. The skill is not always:true, and model invocation is allowed (normal), but the combination of settings modification + trigger injection capability increases blast radius if misused.
安全有层次,运行前请审查代码。
运行时依赖
无特殊依赖
版本
latestv0.1.12026/4/4
添加主题路由功能用于主题放置推荐,改进升级程序
● 可疑
安装命令 点击复制
官方npx clawhub@latest install skill-kit
镜像加速npx clawhub@latest install skill-kit --registry https://cn.clawhub-mirror.com
技能文档
用于创建、管理和维护 Claude 代码技能的综合工具包。包括以下命令:
| 命令 | 描述 | 链接 |
|---|---|---|
| architecture | 多主题技能结构和主题 | architecture.md |
| convert | 将代理或脚本转换为技能 | convert.md |
| dedup | 确定和合并重复技能 | dedup.md |
| lint | 验证和修复 SKILL.md 前置物 | lint.md |
| merge | 合并相关技能 | merge.md |
| route | 推荐技能中的主题放置 | route.md |
| trigger | 注册触发器和生成 hook | trigger.md |
| upgrade | 提升现有技能或添加主题 | upgrade.md |
| writer | 交互式技能创建向导 | writer.md |
核心工作流
创建 (skill-writer)
始终使用writer 确保正确的前置物和结构。
/skill-kit writer # 启动向导
维护 (upgrade/lint)
使用upgrade 添加新功能或主题到现有技能。
/skill-kit upgrade skill-name # 交互式升级
/skill-kit lint skill-name # 仅验证
改进类型:
- 添加主题:为新子功能添加文档
- 添加脚本:添加逻辑到
scripts/并在 SKILL.md 中引用 - 修复前置物:更正
triggers、depends-on或description
触发器 (自动生成 Hook)
/skill-kit trigger compile # 扫描技能 -> 生成调度程序 -> 注册在 settings.json
/skill-kit trigger list # 列出注册的触发器
/skill-kit trigger dry-run # 仅预览
在 SKILL.md 中声明 triggers -> 自动生成 hook 脚本 -> 自动注册在 settings.json。详细指南成功案例
场景 (2026-03-09):- 找到 3 个 openclaw 相关函数
- 提出 3 个合并选项
- 结果:实现成功,用户满意
关键因素:
- 识别 3 个函数
- "合并?" AskUserQuestion
- 使用 skill-writer 合并技能(多主题)
Ralph 模式 (AskUserQuestion 跳过)
如果存在.ralph/ 目录,则操作在 Ralph 模式。
工作流变化:
| 步骤 | 用户交互 | 工作流 |
|---|---|---|
| Step 1: 自动检测 | AskUserQuestion (multiSelect) | 向 .ralph/improvements.md 写入摘要信息 |
| Step 1.5: 合并逻辑 / 结构 | - | 记录 improvements.md |
| Step 2: 要求 | AskUserQuestion | 向 improvements.md 写入触发器/范围推荐 |
| Step 3: 类型推荐 | 推荐仅 | 记录 improvements.md |
| Step 4: 实现 | 直接操作 | 禁止 - 使用 [NEEDS_REVIEW] 标签 |
| Step 5: 验证 | 验证 | 自动验证 (更改完成后) |
自我改进
更改完成后,基于对话自我改进:- 识别失败和工作-around 模式
- 如果找到候选项,运行
/skill-kit upgrade skill-kit
数据来源:ClawHub ↗ · 中文优化:龙虾技能库
OpenClaw 技能定制 / 插件定制 / 私有工作流定制
免费技能或插件可能存在安全风险,如需更匹配、更安全的方案,建议联系付费定制