refactor: skill cli

[hustcc]

 chart-visualization-skills/
  ├── src/
  │   ├── index.ts              # CLI 入口
  │   ├── commands/
  │   │   ├── retrieve.ts
  │   │   └── list.ts
  │   ├── core/
  │   │   ├── bm25.ts
  │   │   ├── retriever.ts
  │   │   └── types.ts
  │   └── scripts/
  │       └── build.ts
  ├── index/                    # JSON 索引文件
  ├── dist/                     # 编译产物
  ├── package.json              # 合并后，依赖以 cli 为准
  ├── tsconfig.json
  └── skills/                   # 原有 skill 文档

package.json 变更：

• 删除了旧的 @anthropic-ai/sdk、openai、express、cors、dotenv、uuid、ws 等依赖
• 保留 cli 的依赖：commander（运行时）、typescript、@types/node、vitest（开发时）
• build 脚本改为 tsc，新增 build:index 用于构建索引

编译和所有命令均验证通过。

[gemini-code-assist]

Code Review

This pull request undertakes a significant refactoring of the AntV Skills project, migrating the entire codebase from JavaScript to TypeScript. The changes introduce a new modular structure, a command-line interface built with commander.js, and a programmatic API for skill retrieval. Core functionalities like BM25 search, skill retrieval, and index building have been re-implemented in TypeScript, and project dependencies have been updated accordingly. Feedback highlights a high-severity bug in the custom YAML parser in src/scripts/build.ts that incorrectly handles multi-line strings, a logical issue with the englishPattern regex in src/core/bm25.ts due to lowercasing, and a recommendation to relocate build artifacts from src/index to dist/index for improved project organization.

[gemini-code-assist]

The manual YAML parser fails to correctly handle multi-line strings (block scalars). After the first line of a multi-line string is processed, meta[currentKey] is reassigned from an array to a string. On subsequent lines, the Array.isArray(meta[currentKey]) check will fail, causing all remaining lines of the multi-line string to be ignored. Consider using a robust YAML parsing library like js-yaml or fixing the accumulation logic to keep the lines in an array until the entire block is parsed.

[gemini-code-assist]

The englishPattern regex includes a character class for uppercase letters [A-Z], but it is applied to normalized, which has already been converted to lowercase. This makes the camelCase matching logic ineffective and the regex redundant. If camelCase splitting is intended, the regex should be applied to the original text before lowercasing, or the regex should be simplified to avoid confusion.

Suggested change

	const normalized = text.toLowerCase();
	const tokens: string[] = [];
	const englishPattern = /[a-z][a-z0-9]*(?:[A-Z][a-z0-9]*)*/gi;
	const normalized = text.toLowerCase();
	const tokens: string[] = [];
	const englishPattern = /[a-z][a-z0-9]*/gi;

[gemini-code-assist]

Generating build artifacts into the src/ directory is generally discouraged as it pollutes the source tree. It is better to generate the index files directly into the dist/ directory or a dedicated build/ directory. This would also allow simplifying the build script in package.json by removing the cp command.

Suggested change

	const INDEX_DIR = path.join(PKG_ROOT, 'src', 'index');
	const INDEX_DIR = path.join(PKG_ROOT, 'dist', 'index');