CLI-Anything
Use this skill to work with the local CLI-Anything/ repository in the workspace and turn its methodology into something usable from OpenClaw.
What this skill is for
Use it for three common cases:
- 1. Assess feasibility — inspect a target app/repo and decide whether CLI-Anything is a good fit.
- Use an existing harness — work with one of the repo's prebuilt
agent-harness examples. - Wrap for OpenClaw — turn CLI-Anything guidance or an existing harness into an OpenClaw skill/workflow.
Local source of truth
The repository is expected at:
Read these first when needed:
- -
CLI-Anything/README.md — overall platform and examples - INLINECODE4 — generation methodology and quality bar
- INLINECODE5 — plugin behavior and expected output layout
For a fast survey of bundled examples, read references/bundled-harnesses.md.
Core workflow
1) Classify the user's request
Decide which path applies:
- - Methodology request: user wants to understand or apply CLI-Anything generally
- Existing harness request: user wants to use/demo one of the included examples like GIMP or LibreOffice
- Packaging request: user wants an OpenClaw skill or ClawHub package derived from CLI-Anything
2) Inspect prerequisites
Before promising execution, check:
- - Python 3.10+
- target software presence if using a real harness backend
- whether the request is about:
- building a harness
- running a generated CLI
- packaging/publishing
3) Prefer existing harnesses before generation
If the repo already includes a matching harness under <software>/agent-harness/, use that as the baseline instead of pretending generation has to happen from scratch.
4) For OpenClaw packaging, separate method from implementation
CLI-Anything itself is not a native OpenClaw skill package. When adapting it:
- - keep the OpenClaw
SKILL.md focused on when to use and how to navigate the local repo - move large methodology text into references
- do not claim that
/plugin or /cli-anything slash commands are directly available inside OpenClaw unless you have actually wired them up
Practical guidance
When the user wants to use an existing bundled harness
- 1. Locate INLINECODE11
- Read its
setup.py, package layout, and local README/SOP file - Identify hard dependencies on the real software backend
- Install or verify Python requirements only as needed
- Validate the CLI entry point and a minimal command
When the user wants to make a new app agent-native
Use CLI-Anything as methodology, not magic:
- 1. Analyze backend engine, data model, existing CLI/API hooks
- Define command groups and state model
- Create or refine INLINECODE13
- Add tests and a INLINECODE14
- Install the resulting package to PATH
- Verify real backend execution, not mock-only behavior
When the user wants an OpenClaw skill
There are two good outputs:
- - Method skill: teaches OpenClaw how to use CLI-Anything on local repos
- Harness skill: wraps one generated CLI or one concrete software target
Default to a method skill unless the user clearly wants a single app workflow.
Rules
- - Do not say CLI-Anything is already a native OpenClaw skill unless you created that wrapper.
- Do not promise a generated CLI works until you verify its entry point and real backend dependencies.
- Treat third-party generated code as reviewable output, not automatically trusted output.
- For publishing, require explicit user intent before pushing anything external.
- If packaging for ClawHub, keep the skill lean and point to local references instead of pasting huge docs into
SKILL.md.
Useful script
Use scripts/inspect_cli_anything.py to quickly inspect the local repo and enumerate bundled harnesses. It prints JSON with:
- - whether the repo exists
- discovered harnesses
- whether each harness has
setup.py, package directory, README, and E2E tests
Run:
CODEBLOCK0
Use scripts/recommend_harness.py to rank the bundled harnesses and suggest good first validation targets.
Run:
CODEBLOCK1
Helpful local paths
- - Repo root: INLINECODE19
- Plugin docs: INLINECODE20
- Example harnesses: INLINECODE21
Reference files
- -
references/bundled-harnesses.md — quick map of included examples - INLINECODE23 — how to package CLI-Anything ideas into OpenClaw skills
- INLINECODE24 — first locally verified runnable example
CLI-Anything
使用此技能处理工作区中的本地 CLI-Anything/ 仓库,并将其方法论转化为可在 OpenClaw 中使用的工具。
此技能的用途
适用于三种常见场景:
- 1. 评估可行性 — 检查目标应用/仓库,判断 CLI-Anything 是否适用。
- 使用现有工具集 — 使用仓库中预构建的 agent-harness 示例。
- 封装为 OpenClaw 技能 — 将 CLI-Anything 指南或现有工具集转化为 OpenClaw 技能/工作流。
本地权威来源
仓库预期位于:
- - /root/.openclaw/workspace/CLI-Anything
需要时优先阅读以下内容:
- - CLI-Anything/README.md — 整体平台和示例
- CLI-Anything/cli-anything-plugin/HARNESS.md — 生成方法论和质量标准
- CLI-Anything/cli-anything-plugin/README.md — 插件行为和预期输出布局
如需快速浏览捆绑示例,请阅读 references/bundled-harnesses.md。
核心工作流
1) 分类用户请求
判断适用路径:
- - 方法论请求:用户希望理解或应用 CLI-Anything 的通用方法
- 现有工具集请求:用户希望使用/演示包含的示例(如 GIMP 或 LibreOffice)
- 封装请求:用户希望从 CLI-Anything 派生出 OpenClaw 技能或 ClawHub 包
2) 检查先决条件
在承诺执行前,检查:
- - Python 3.10+
- 如果使用真实工具集后端,需确认目标软件存在
- 请求是否涉及:
- 构建工具集
- 运行生成的 CLI
- 打包/发布
3) 优先使用现有工具集而非重新生成
如果仓库中已有 /agent-harness/ 下的匹配工具集,将其作为基准,而非假装需要从头开始生成。
4) 针对 OpenClaw 打包,分离方法与实现
CLI-Anything 本身不是原生的 OpenClaw 技能包。在适配时:
- - 保持 OpenClaw SKILL.md 聚焦于何时使用和如何导航本地仓库
- 将大量方法论文本移至参考资料
- 除非实际配置了 /plugin 或 /cli-anything 斜杠命令,否则不要声称这些命令在 OpenClaw 中可直接使用
实用指南
当用户希望使用现有捆绑工具集时
- 1. 定位 /agent-harness/
- 阅读其 setup.py、包布局以及本地 README/SOP 文件
- 识别对真实软件后端的硬依赖
- 仅在需要时安装或验证 Python 依赖
- 验证 CLI 入口点和最小命令
当用户希望使新应用支持代理原生操作时
将 CLI-Anything 作为方法论而非魔法使用:
- 1. 分析后端引擎、数据模型、现有 CLI/API 钩子
- 定义命令组和状态模型
- 创建或完善 agent-harness/
- 添加测试和 TEST.md
- 将生成的包安装到 PATH
- 验证真实后端执行,而非仅模拟行为
当用户希望获得 OpenClaw 技能时
有两种好的输出形式:
- - 方法技能:教会 OpenClaw 如何在本地仓库上使用 CLI-Anything
- 工具集技能:封装一个生成的 CLI 或一个具体的软件目标
除非用户明确希望单一应用工作流,否则默认使用方法技能。
规则
- - 除非你创建了封装器,否则不要声称 CLI-Anything 已是原生 OpenClaw 技能。
- 在验证生成的 CLI 入口点和真实后端依赖之前,不要承诺其可用性。
- 将第三方生成的代码视为可审查的输出,而非自动可信的输出。
- 发布前,需有明确的用户意图才能推送任何外部内容。
- 如果为 ClawHub 打包,保持技能精简,指向本地参考资料而非将大段文档粘贴到 SKILL.md 中。
实用脚本
使用 scripts/inspectclianything.py 快速检查本地仓库并枚举捆绑的工具集。输出 JSON 格式信息,包括:
- - 仓库是否存在
- 发现的工具集
- 每个工具集是否包含 setup.py、包目录、README 和端到端测试
运行:
bash
python3 /root/.openclaw/workspace/skills/cli-anything/scripts/inspectclianything.py
使用 scripts/recommend_harness.py 对捆绑工具集进行排序,并推荐良好的首选验证目标。
运行:
bash
python3 /root/.openclaw/workspace/skills/cli-anything/scripts/recommend_harness.py
有用的本地路径
- - 仓库根目录:/root/.openclaw/workspace/CLI-Anything
- 插件文档:/root/.openclaw/workspace/CLI-Anything/cli-anything-plugin
- 示例工具集:/root/.openclaw/workspace/CLI-Anything/*/agent-harness
参考文件
- - references/bundled-harnesses.md — 包含示例的快速概览
- references/openclaw-adaptation-notes.md — 如何将 CLI-Anything 思路打包为 OpenClaw 技能
- references/validated-example-gimp.md — 首个本地验证的可运行示例