agent-cli-creator

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Agent CLI Creator

Agent CLI 创建器

Guide an AI Coding Agent to build a website-automating CLI tool backed by the kimi-webbridge browser daemon.

指导AI编码Agent构建基于kimi-webbridge浏览器守护进程的网站自动化CLI工具。

Phase 1: Prerequisites

第一阶段：前置条件

bash

~/.kimi-webbridge/bin/kimi-webbridge status

Result	Action
`running: true` + `extension_connected: true`	Proceed
Command not found	Tell user to visit https://www.kimi.team/features/webbridge (中文: https://www.kimi.team/zh-cn/features/webbridge) to install
`running: false` or `extension_connected: false`	Invoke the `kimi-webbridge` skill → references/operations.md

bash

~/.kimi-webbridge/bin/kimi-webbridge status

结果	操作
`running: true` + `extension_connected: true`	继续执行
命令未找到	告知用户访问https://www.kimi.team/features/webbridge（中文: https://www.kimi.team/zh-cn/features/webbridge）进行安装
`running: false` 或 `extension_connected: false`	调用 `kimi-webbridge` 技能 → references/operations.md

Phase 2: Requirements Interview

第二阶段：需求沟通

Ask the user in one message, wait for reply:

Target website URL (required)
Programming language — Go (recommended if in this repo) / Python / Node.js / Other
Login required? — Yes / No / Unknown (can skip for now)
First 1–3 features — pick from common categories:
- Read: home feed, search, profile page, post/item detail
- Write: create post, like/unlike, comment, bookmark/save
- Account: login-status, user info

Explain iterative development to the user:

"We'll start with 1–3 features to validate the approach end-to-end before adding more. Site archaeology for features you don't need yet wastes time. You can always add features later by re-running from Phase 3."

用一条消息询问用户，等待回复：

目标网站URL（必填）
编程语言 — Go（若在此仓库中推荐使用）/ Python / Node.js / 其他
是否需要登录？ — 是 / 否 / 未知（可暂时跳过）
首批1–3个功能 — 从常见类别中选择：
- 读取：首页信息流、搜索、个人主页、帖子/商品详情
- 写入：创建帖子、点赞/取消点赞、评论、收藏/保存
- 账户：登录状态、用户信息

向用户解释迭代开发理念：

“我们将从1-3个功能开始，先端到端验证方案后再添加更多功能。为目前不需要的功能进行网站探查会浪费时间。之后你可以通过重新执行第三阶段来添加新功能。”

Phase 3: Site Archaeology (mandatory before writing any code)

第三阶段：网站探查（编写代码前必须完成）

This phase is non-negotiable. Do not write business logic until archaeology is complete.

For each planned feature, run the full protocol in

references/site-exploration.md

The protocol yields: API endpoint, required headers, response shape, and a verified

evaluate

call that proves the API works inside the browser session.

Only proceed to Phase 4 when you have a working
evaluate
call for every planned feature.

此阶段不可省略。在完成探查前请勿编写业务逻辑。

针对每个计划开发的功能，执行

references/site-exploration.md

中的完整流程。

该流程将产出：API端点、所需请求头、响应格式，以及经过验证的

evaluate

调用，用于证明该API在浏览器会话中可正常工作。

仅当所有计划功能都拥有可用的
evaluate
调用时，才可进入第四阶段。

Phase 4: Implement

第四阶段：开发实现

Order matters — do not skip ahead:

Project scaffold — see
```
references/go-layout.md
```
for Go; adapt conventions for other languages
login-status
command — if the site requires login; see
```
references/login-handling.md
```
Read commands — no side effects; implement and test first
Write commands — side effects (post, like, etc.); implement after reads work

After each command is implemented, immediately verify before moving on:

bash

{platform}-cli {command} --help          # --help must work
{platform}-cli {command} [args]          # must return {"ok": true, "data": ...}

For write commands, also verify the error path (e.g., wrong ID, missing flag) returns

{"ok": false, ...}

with non-zero exit.

If verification fails, fix before implementing the next command.

Universal CLI contract (all languages):

```
--help
```
/
```
-h
```
must work on every command

All output:

{"ok": true, "data": ...}

{"ok": false, "error": {"code": "...", "message": "..."}}

Non-zero exit code on error

Go: 用

go mod init {platform}-cli

初始化独立 module，按

references/go-layout.md

的结构搭建，包含自己的轻量 browser client 和 output helper。

顺序至关重要，请勿跳过步骤：

项目脚手架 — Go语言可参考
```
references/go-layout.md
```
；其他语言需适配相应规范
login-status
命令 — 若网站需要登录，请参考
```
references/login-handling.md
```
读取类命令 — 无副作用，优先实现并测试
写入类命令 — 有副作用（如发布帖子、点赞等），需在读取类命令正常工作后再实现

每个命令实现完成后，需立即验证通过再进行下一步：

bash

{platform}-cli {command} --help          # --help must work
{platform}-cli {command} [args]          # must return {"ok": true, "data": ...}

对于写入类命令，还需验证错误场景（如ID错误、缺少参数）会返回

{"ok": false, ...}

并以非零状态码退出。

若验证失败，需修复后再实现下一个命令。

通用CLI规范（所有语言）：

```
--help
```
/
```
-h
```
必须在每个命令中可用

所有输出格式必须为：

{"ok": true, "data": ...}

或

{"ok": false, "error": {"code": "...", "message": "..."}}

出错时返回非零状态码

Go语言： 使用

go mod init {platform}-cli

初始化独立module，按照

references/go-layout.md

的结构搭建项目，包含自定义的轻量browser client和output helper。

Phase 5: Write Companion Skill

第五阶段：编写配套技能

After the CLI works, create

~/.claude/skills/{platform}-cli/SKILL.md

using the template in

references/companion-skill-template.md

Purpose: tells a future AI agent how to use the CLI, not how to build it. The two skills serve different audiences at different times.

CLI工具正常工作后，使用

references/companion-skill-template.md

中的模板创建

~/.claude/skills/{platform}-cli/SKILL.md

文件。

目的： 告知未来的AI Agent如何使用该CLI工具，而非如何构建它。这两个技能在不同时期服务于不同的使用场景。