autosnippet
Version:
Extract code patterns into a knowledge base for AI coding assistants
164 lines (163 loc) • 5.72 kB
JavaScript
/**
* TopicClassifier — 按主题分组 KnowledgeEntry
*
* 将 kind='pattern' 的知识条目按主题分组，用于 Channel B（Smart Rules）。
* 每个主题对应一个 .mdc 文件，设 alwaysApply: false + 丰富的 description。
*/
/**
* 主题定义
* - dimensions: 关联的 bootstrap 维度
* - keywords: 用于匹配 entry 分类的关键词
* - descriptionKeywords: 用于 .mdc description 字段的关键词（Agent 关联性判断依据）
*/
const TOPIC_MAP = {
networking: {
dimensions: ['event-and-data-flow'],
descriptionKeywords: 'network, HTTP, API, request, response, URL, fetch, socket, REST, download, upload, error handling, retry, timeout',
},
ui: {
dimensions: ['code-pattern'],
descriptionKeywords: 'view, controller, UI, layout, animation, cell, table, collection, button, scroll, navigation, auto layout, gesture, storyboard',
},
data: {
dimensions: ['code-pattern', 'architecture'],
descriptionKeywords: 'model, storage, database, cache, CoreData, Realm, SQLite, keychain, UserDefaults, JSON, parsing, serialization, persistence',
},
architecture: {
dimensions: ['architecture', 'best-practice'],
descriptionKeywords: 'singleton, delegate, factory, observer, protocol, manager, service, dependency injection, module, MVVM, MVC, coordinator, router, design pattern',
},
conventions: {
dimensions: ['code-standard'],
descriptionKeywords: 'naming, format, style, import, header, prefix, convention, documentation, file organization, constants, enum, typedef, pragma mark',
},
};
export class TopicClassifier {
projectName;
/** @param projectName 项目名称 */
constructor(projectName = 'Project') {
this.projectName = projectName;
}
/**
* 将 patterns 按主题分组
* @param entries KnowledgeEntry 数组 (kind='pattern')
* @returns { topic: [entries] }
*/
group(entries) {
const grouped = {};
const unmatched = [];
for (const entry of entries) {
const topic = this._classifyEntry(entry);
if (topic) {
if (!grouped[topic]) {
grouped[topic] = [];
}
grouped[topic].push(entry);
}
else {
unmatched.push(entry);
}
}
// 未匹配的归入 'general' — 但只在有内容时
if (unmatched.length > 0) {
grouped.general = unmatched;
}
return grouped;
}
/**
* 为主题构建 description — Agent 判断关联性的唯一依据
*
* v3: 动态丰富化 — 除硬编码 baseKeywords 外，还从 entries 的
* tags, whenClause, trigger 中提取高价值关键词，让 Cursor 的
* description 匹配更精准。
*/
buildDescription(topic, entries) {
const topicDef = TOPIC_MAP[topic];
const baseKeywords = topicDef
? topicDef.descriptionKeywords
: entries
.map((e) => e.title || '')
.filter(Boolean)
.join(', ');
// 从 entries 提取动态关键词（tags + whenClause + title/description）
const entryKeywords = entries
.flatMap((e) => this._extractKeywords(e))
.filter(Boolean);
// 从 tags 直接提取（tags 本身就是高质量关键词）
const tagKeywords = entries
.flatMap((e) => e.tags || [])
.map((t) => t.toLowerCase())
.filter((t) => t.length >= 2 && !STOP_WORDS.has(t));
// 从 whenClause 提取（包含触发场景信息）
const whenKeywords = entries
.flatMap((e) => (e.whenClause || '').match(/[a-zA-Z]{3,}/g) || [])
.map((w) => w.toLowerCase())
.filter((w) => !STOP_WORDS.has(w));
const allExtra = [...new Set([...tagKeywords, ...whenKeywords, ...entryKeywords])].slice(0, 15);
const extra = allExtra.length > 0 ? `, ${allExtra.join(', ')}` : '';
return `${this._topicLabel(topic)} patterns for ${this.projectName} — ${baseKeywords}${extra}. Use when writing or reviewing ${this._topicLabel(topic).toLowerCase()}-related code.`;
}
/**
* 分类单个 entry 到主题 — 直读 AI 预计算的 topicHint
*/
_classifyEntry(entry) {
return entry.topicHint || null; // AI 没给 → null → 归入 general
}
/**
* 从 entry 提取关键词
*/
_extractKeywords(entry) {
const text = `${entry.title || ''} ${entry.description || ''}`;
// 提取英文关键词（3+ 字母）
const words = text.match(/[a-zA-Z]{3,}/g) || [];
const filtered = words.map((w) => w.toLowerCase()).filter((w) => !STOP_WORDS.has(w));
return [...new Set(filtered)].slice(0, 5);
}
_topicLabel(topic) {
const labels = {
networking: 'Networking',
ui: 'UI',
data: 'Data',
architecture: 'Architecture',
conventions: 'Conventions',
general: 'General',
};
return (labels[topic] || topic.charAt(0).toUpperCase() + topic.slice(1));
}
}
const STOP_WORDS = new Set([
'the',
'and',
'for',
'this',
'that',
'with',
'from',
'use',
'using',
'when',
'not',
'all',
'are',
'has',
'have',
'been',
'will',
'can',
'should',
'must',
'may',
'each',
'which',
'their',
'your',
'its',
'project',
'code',
'file',
'class',
'method',
'function',
'bootstrap',
]);
export default TopicClassifier;