关于 qe-code-intelligence
qe-code-intelligence 是指导使用 v3 代码智能能力的技能,面向需要理解和分析代码库的开发者。它涵盖知识图谱、语义搜索、依赖映射及上下文精简。
主要功能
- 01构建代码知识图谱
- 02语义代码搜索
- 03映射代码依赖
- 04分析变更影响
- 05精简 AI 上下文
适用场景
- 01快速熟悉陌生代码库
- 02按含义查找代码
- 03评估文件变更影响
安装
复制下面的命令,在终端中运行。
$
npx skills add https:/ / github.com/ proffesor-for-testing/ agentic-qe --skill qe-code-intelligence安装后在 Claude Code 中输入/qe-code-intelligence确认已加载。
包含的文件
除 SKILL.md 外,安装时一起下载的参考资料和脚本。
evals/qe-code-intelligence.yamlschemas/output.jsonscripts/validate-config.json详细介绍
QE 代码智能
用途
指导如何使用 v3 的代码智能功能,包括知识图谱构建、语义代码搜索、依赖关系映射,以及大幅减少令牌用量的上下文感知代码理解。
启用场景
- 理解陌生代码时
- 按语义搜索代码时
- 分析依赖关系时
- 构建代码知识图谱时
- 减少 AI 操作的上下文用量时
快速入门
# Index codebase into knowledge graph
aqe code index src/ --incremental
# Semantic code search
aqe code search "authentication middleware"
# Analyze change impact
aqe code impact src/services/UserService.ts --depth 3
# Map dependencies
aqe code deps src/
# Analyze complexity and find hotspots
aqe code complexity src/
# Generate C4 architecture diagrams (Mermaid) with a confidence score
aqe code c4 .
Agent 工作流
// Build knowledge graph
Task("Index codebase", `
Build knowledge graph for the project:
- Parse all TypeScript files in src/
- Extract entities (classes, functions, types)
- Map relationships (imports, calls, inheritance)
- Generate embeddings for semantic search
Store in AgentDB vector database.
`, "qe-kg-builder")
// Semantic search
Task("Find relevant code", `
Search for code related to "user authentication flow":
- Use semantic similarity (not just keyword)
- Include related functions and types
- Rank by relevance score
- Return with minimal context (80% token reduction)
`, "qe-code-intelligence")
知识图谱操作
1. 代码库索引
await knowledgeGraph.index({
source: 'src/**/*.ts',
extraction: {
entities: ['class', 'function', 'interface', 'type', 'variable'],
relationships: ['imports', 'calls', 'extends', 'implements', 'uses'],
metadata: ['jsdoc', 'complexity', 'lines']
},
embeddings: {
model: 'code-embedding',
dimensions: 384,
normalize: true
},
incremental: true // Only index changed files
});
2. 语义搜索
await semanticSearcher.search({
query: 'payment processing with stripe',
options: {
similarity: 'cosine',
threshold: 0.7,
limit: 20,
includeContext: true
},
filters: {
fileTypes: ['.ts', '.tsx'],
excludePaths: ['node_modules', 'dist']
}
});
3. 依赖分析
await dependencyMapper.analyze({
entry: 'src/services/OrderService.ts',
depth: 3,
direction: 'both', // imports and importedBy
output: {
graph: true,
metrics: {
afferentCoupling: true,
efferentCoupling: true,
instability: true
}
}
});
令牌缩减策略
// Get context with 80% token reduction
const context = await codeIntelligence.getOptimizedContext({
query: 'implement user registration',
budget: 4000, // max tokens
strategy: {
relevanceRanking: true,
summarization: true,
codeCompression: true,
deduplication: true
},
include: {
signatures: true,
implementations: 'relevant-only',
comments: 'essential',
examples: 'top-3'
}
});
知识图谱架构
interface KnowledgeGraph {
entities: {
id: string;
type: 'class' | 'function' | 'interface' | 'type' | 'file';
name: string;
file: string;
line: number;
embedding: number[];
metadata: Record<string, any>;
}[];
relationships: {
source: string;
target: string;
type: 'imports' | 'calls' | 'extends' | 'implements' | 'uses';
weight: number;
}[];
indexes: {
byName: Map<string, string[]>;
byFile: Map<string, string[]>;
byType: Map<string, string[]>;
};
}
搜索结果
interface SearchResult {
entity: {
name: string;
type: string;
file: string;
line: number;
};
relevance: number;
snippet: string;
context: {
before: string[];
after: string[];
related: string[];
};
explanation: string;
}
CLI 示例
# Full reindex
aqe code index src/
# Incremental index (changed files only)
aqe code index src/ --incremental
# Index only files changed since a git ref
aqe code index . --git-since HEAD~5
# Semantic code search
aqe code search "database connection"
# Change impact analysis
aqe code impact src/services/UserService.ts
# Dependency mapping
aqe code deps src/ --depth 5
# Complexity metrics and hotspots
aqe code complexity src/ --format json
注意事项
- WARNING:code-intelligence 领域的成功率为 18%——对于简单查询,优先使用直接的 grep/glob,而非基于 Agent 的代码搜索
- 知识图谱构建在超过 50K LOC 的代码库上会失败——请限定到特定模块
- 如果没有特定领域的嵌入,语义搜索会返回不相关的结果——请始终手动验证搜索结果
- Agent 声称可“减少 80% 的令牌用量”,但可能会跳过关键上下文——请确认结果中包含关键文件
- 使用前必须初始化 Fleet:运行
aqe health进行诊断;如果遇到初始化错误,则运行aqe init重新初始化
协作
主要 Agent:qe-kg-builder、qe-dependency-mapper、qe-impact-analyzer、qe-code-complexity 协调 Agent:qe-code-intelligence 相关技能:qe-test-generation、qe-defect-intelligence
评论
支持 Markdown