autosnippet
Version:
Extract code patterns into a knowledge base for AI coding assistants
54 lines (53 loc) • 2.04 kB
TypeScript
/**
* RoleRefiner — 四重信号融合角色精化
*
* 将 TargetClassifier 的正则推断 (~65% 准确率) 提升到 ≥90%，
* 通过融合 AST 结构、CallGraph 行为、DataFlow 数据流、EntityGraph 拓扑四重信号。
*
* 信号权重:
* AST 结构 0.30 继承链/协议/import/后缀
* CallGraph 行为 0.30 被调用分析/扇入扇出比/调用类型
* DataFlow 数据流 0.15 源汇分析/转换检测
* EntityGraph 拓扑 0.10 入度分析/模式检测
* 正则基线 0.15 TargetClassifier 结果
*
* @module RoleRefiner
*/
import type { BootstrapRepositoryImpl } from '../../repository/bootstrap/BootstrapRepository.js';
import type { CodeEntityRepositoryImpl } from '../../repository/code/CodeEntityRepository.js';
import type { KnowledgeEdgeRepositoryImpl } from '../../repository/knowledge/KnowledgeEdgeRepository.js';
import type { ModuleRole } from './PanoramaTypes.js';
export type { ModuleRole } from './PanoramaTypes.js';
export interface RoleSignal {
role: ModuleRole;
confidence: number;
weight: number;
source: string;
}
export type RoleResolution = 'clear' | 'uncertain' | 'fallback';
export interface RefinedRole {
refinedRole: ModuleRole;
confidence: number;
resolution: RoleResolution;
alternatives?: Array<[string, number]>;
signals: RoleSignal[];
}
export interface ModuleCandidate {
name: string;
inferredRole: ModuleRole;
files: string[];
/** 来自配置文件的层级名称（如 Boxfile 中的 layer 声明） */
configLayer?: string;
}
export declare class RoleRefiner {
#private;
constructor(bootstrapRepo: BootstrapRepositoryImpl, entityRepo: CodeEntityRepositoryImpl, edgeRepo: KnowledgeEdgeRepositoryImpl, projectRoot: string);
/**
* 精化单个模块的角色
*/
refineRole(module: ModuleCandidate): Promise<RefinedRole>;
/**
* 批量精化所有模块
*/
refineAll(modules: ModuleCandidate[]): Promise<Map<string, RefinedRole>>;
}