autosnippet 3.0.3 → 3.0.7

package/lib/core/enhancement/langchain-enhancement.js

@@ -0,0 +1,233 @@
+/**
+ * LangChain / Agent Enhancement Pack
+ * 条件: { languages: ['python'], frameworks: ['langchain'] }
+ *
+ * 覆盖 LLM Agent 开发生态:
+ *   - LangChain Chain / Agent / Tool
+ *   - RAG Pipeline (Retriever → Splitter → Embedding → VectorStore)
+ *   - Prompt Template / Output Parser
+ *   - LlamaIndex Query Engine
+ *   - 多轮对话 Memory
+ *   - Streaming / Callback
+ */
+
+import { EnhancementPack } from './EnhancementPack.js';
+
+class LangChainEnhancement extends EnhancementPack {
+  get id() {
+    return 'python-langchain';
+  }
+  get displayName() {
+    return 'LangChain / Agent Enhancement';
+  }
+  get conditions() {
+    return { languages: ['python'], frameworks: ['langchain'] };
+  }
+
+  getExtraDimensions() {
+    return [
+      {
+        id: 'langchain-chain-scan',
+        label: 'Chain / Agent 分析',
+        guide:
+          'LangChain Chain 拓扑分析 — LCEL (RunnableSequence | RunnableParallel) 链路、Agent 工具注册、AgentExecutor 配置、工具函数实现 (@tool decorator)',
+        knowledgeTypes: ['architecture', 'code-pattern'],
+        skillWorthy: true,
+        dualOutput: true,
+        skillMeta: {
+          name: 'project-langchain-chains',
+          description:
+            'LangChain chain topology, LCEL pipelines and agent tool registrations (auto-generated by enhancement)',
+        },
+      },
+      {
+        id: 'langchain-rag-scan',
+        label: 'RAG Pipeline 分析',
+        guide:
+          'RAG 检索增强生成管道分析 — Document Loader / Text Splitter / Embedding Model / VectorStore (Chroma/FAISS/Pinecone) 选型、Retriever 配置 (search_type/k/score_threshold)、Reranking 策略',
+        knowledgeTypes: ['architecture', 'code-pattern'],
+        skillWorthy: true,
+        dualOutput: true,
+        skillMeta: {
+          name: 'project-langchain-rag',
+          description:
+            'RAG pipeline — document loading, splitting, embedding and vector retrieval (auto-generated by enhancement)',
+        },
+      },
+      {
+        id: 'langchain-prompt-scan',
+        label: 'Prompt / Output 模式分析',
+        guide:
+          'Prompt 工程分析 — PromptTemplate / ChatPromptTemplate 结构、Few-shot 示例管理、Output Parser (JSON/Pydantic/Structured)、System Message 设计模式',
+        knowledgeTypes: ['code-pattern'],
+        skillWorthy: true,
+        dualOutput: true,
+        skillMeta: {
+          name: 'project-langchain-prompts',
+          description:
+            'Prompt templates, few-shot examples and output parser configurations (auto-generated by enhancement)',
+        },
+      },
+    ];
+  }
+
+  getGuardRules() {
+    return [
+      {
+        ruleId: 'langchain-prompt-injection',
+        category: 'safety',
+        dimension: 'file',
+        severity: 'warning',
+        languages: ['python'],
+        pattern: /(?:PromptTemplate|ChatPromptTemplate)[\s\S]*?(?:f['"]|\.format\s*\()/,
+        message: 'Prompt 中直接拼接用户输入可能导致 Prompt Injection — 使用 input_variables 参数化',
+      },
+      {
+        ruleId: 'langchain-no-bare-invoke',
+        category: 'correctness',
+        dimension: 'file',
+        severity: 'info',
+        languages: ['python'],
+        pattern: /\.invoke\s*\([^)]*\)\s*(?![\s\S]*?(?:try|except|catch))/,
+        message: 'Chain/Agent invoke 应包含错误处理 — LLM 调用可能超时/限流/返回异常格式',
+      },
+      {
+        ruleId: 'langchain-token-budget',
+        category: 'performance',
+        dimension: 'file',
+        severity: 'info',
+        languages: ['python'],
+        pattern: /max_tokens\s*=\s*(?:None|0)/,
+        message: '建议设置合理的 max_tokens 限制 — 防止意外高额 API 费用',
+      },
+      {
+        ruleId: 'langchain-hardcoded-api-key',
+        category: 'safety',
+        dimension: 'file',
+        severity: 'error',
+        languages: ['python'],
+        pattern: /(?:api_key|openai_api_key|anthropic_api_key)\s*=\s*['"][^'"]+['"]/i,
+        message: 'API Key 不应硬编码在代码中 — 使用环境变量 (os.environ) 或 .env 文件',
+      },
+    ];
+  }
+
+  detectPatterns(astSummary) {
+    const patterns = [];
+
+    // ── Chain / Runnable classes ──
+    for (const cls of astSummary.classes || []) {
+      if (cls.superclass && /Runnable|Chain|BaseTool|BaseRetriever/.test(cls.superclass)) {
+        patterns.push({
+          type: 'langchain-chain',
+          className: cls.name,
+          line: cls.line,
+          confidence: 0.9,
+        });
+      }
+    }
+
+    // ── @tool decorated functions ──
+    for (const m of astSummary.methods || []) {
+      if (m.decorators?.some((d) => /@tool/.test(d))) {
+        patterns.push({
+          type: 'langchain-tool',
+          methodName: m.name,
+          line: m.line,
+          confidence: 0.95,
+        });
+      }
+    }
+
+    // ── Agent/RAG setup functions ──
+    for (const m of astSummary.methods || []) {
+      if (!m.className && m.name) {
+        const nameLower = m.name.toLowerCase();
+        if (
+          nameLower.includes('create_agent') ||
+          nameLower.includes('build_chain') ||
+          nameLower.includes('setup_rag') ||
+          nameLower.includes('create_retriever') ||
+          nameLower.includes('get_llm') ||
+          nameLower.includes('create_chain') ||
+          nameLower.includes('build_graph')
+        ) {
+          patterns.push({
+            type: 'langchain-setup',
+            methodName: m.name,
+            line: m.line,
+            confidence: 0.8,
+          });
+        }
+      }
+    }
+
+    // ── Callback / Handler classes ──
+    for (const cls of astSummary.classes || []) {
+      if (cls.superclass && /CallbackHandler|BaseCallbackHandler/.test(cls.superclass)) {
+        patterns.push({
+          type: 'langchain-callback',
+          className: cls.name,
+          line: cls.line,
+          confidence: 0.9,
+        });
+      }
+    }
+
+    // ── Output parser classes ──
+    for (const cls of astSummary.classes || []) {
+      if (cls.superclass && /OutputParser|BaseOutputParser/.test(cls.superclass)) {
+        patterns.push({
+          type: 'langchain-output-parser',
+          className: cls.name,
+          line: cls.line,
+          confidence: 0.9,
+        });
+      }
+    }
+
+    // ── LangGraph nodes/edges ──
+    for (const m of astSummary.methods || []) {
+      if (!m.className && m.name) {
+        const nameLower = m.name.toLowerCase();
+        if (
+          nameLower.includes('node') ||
+          nameLower.includes('should_continue') ||
+          nameLower.includes('route_')
+        ) {
+          patterns.push({
+            type: 'langgraph-node',
+            methodName: m.name,
+            line: m.line,
+            confidence: 0.5,
+          });
+        }
+      }
+    }
+
+    // ── LangChain ecosystem imports ──
+    const lcImports = (astSummary.imports || []).filter(
+      (imp) =>
+        imp.includes('langchain') ||
+        imp.includes('langgraph') ||
+        imp.includes('langsmith') ||
+        imp.includes('llama_index') ||
+        imp.includes('chromadb') ||
+        imp.includes('pinecone') ||
+        imp.includes('faiss') ||
+        imp.includes('openai') ||
+        imp.includes('anthropic')
+    );
+    if (lcImports.length > 0) {
+      patterns.push({
+        type: 'langchain-ecosystem-usage',
+        importCount: lcImports.length,
+        confidence: 0.85,
+      });
+    }
+
+    return patterns;
+  }
+}
+
+export const pack = new LangChainEnhancement();

package/lib/core/enhancement/ml-enhancement.js

@@ -0,0 +1,265 @@
+/**
+ * AI/ML Enhancement Pack
+ * 条件: { languages: ['python'], frameworks: ['ml'] }
+ *
+ * 覆盖 PyTorch / TensorFlow / HuggingFace 机器学习生态:
+ *   - nn.Module 模型架构
+ *   - Training Loop 模式 (optimizer.zero_grad → loss.backward → optimizer.step)
+ *   - DataLoader / Dataset
+ *   - HuggingFace Trainer / Pipeline
+ *   - 模型保存/加载 (state_dict / safetensors)
+ *   - 分布式训练 (DDP / FSDP)
+ */
+
+import { EnhancementPack } from './EnhancementPack.js';
+
+class MLEnhancement extends EnhancementPack {
+  get id() {
+    return 'python-ml';
+  }
+  get displayName() {
+    return 'Python AI/ML (PyTorch/HuggingFace) Enhancement';
+  }
+  get conditions() {
+    return { languages: ['python'], frameworks: ['ml'] };
+  }
+
+  getExtraDimensions() {
+    return [
+      {
+        id: 'ml-model-architecture-scan',
+        label: '模型架构分析',
+        guide:
+          'nn.Module 子类分析 — 层级结构 (forward 方法调用链)、自定义 Layer、残差连接/Attention 模式、参数量估算、模型注册表',
+        knowledgeTypes: ['architecture', 'code-pattern'],
+        skillWorthy: true,
+        dualOutput: true,
+        skillMeta: {
+          name: 'project-ml-models',
+          description:
+            'PyTorch model architectures, custom layers and forward pass patterns (auto-generated by enhancement)',
+        },
+      },
+      {
+        id: 'ml-training-pipeline-scan',
+        label: 'Training Pipeline 分析',
+        guide:
+          'Training 流程分析 — Training Loop 结构 (epoch → batch → forward → loss → backward → step)、learning rate scheduler、gradient clipping/accumulation、早停策略、checkpoint 保存/恢复、HuggingFace Trainer 配置',
+        knowledgeTypes: ['code-pattern'],
+        skillWorthy: true,
+        dualOutput: true,
+        skillMeta: {
+          name: 'project-ml-training',
+          description:
+            'ML training pipeline — training loops, schedulers, checkpointing and HF Trainer (auto-generated by enhancement)',
+        },
+      },
+      {
+        id: 'ml-data-pipeline-scan',
+        label: '数据管道分析',
+        guide:
+          '数据处理管道分析 — Dataset/DataLoader 实现、数据增强 (transforms)、tokenizer 配置、特征工程、数据拆分策略 (train/val/test)',
+        knowledgeTypes: ['code-pattern', 'architecture'],
+        skillWorthy: true,
+        dualOutput: true,
+        skillMeta: {
+          name: 'project-ml-data',
+          description:
+            'ML data pipelines — Dataset/DataLoader, transforms and tokenization (auto-generated by enhancement)',
+        },
+      },
+    ];
+  }
+
+  getGuardRules() {
+    return [
+      {
+        ruleId: 'ml-no-eval-in-train',
+        category: 'correctness',
+        dimension: 'file',
+        severity: 'warning',
+        languages: ['python'],
+        pattern: /model\.train\(\)[\s\S]*?model\.eval\(\)[\s\S]*?loss\.backward/,
+        message: '训练循环中意外切换到 eval 模式后执行 backward — 确保 model.train() 在训练阶段',
+      },
+      {
+        ruleId: 'ml-device-mismatch',
+        category: 'correctness',
+        dimension: 'file',
+        severity: 'warning',
+        languages: ['python'],
+        pattern: /\.to\s*\(\s*['"](?:cuda|cpu)['"]\s*\)/,
+        message: '硬编码 device 字符串 — 建议使用 torch.device() 变量统一管理，支持 MPS/多 GPU',
+      },
+      {
+        ruleId: 'ml-missing-no-grad',
+        category: 'performance',
+        dimension: 'file',
+        severity: 'warning',
+        languages: ['python'],
+        pattern: /def\s+(?:evaluate|validate|test|predict|inference)\s*\([^)]*\)[\s\S]*?(?!torch\.no_grad|@torch\.no_grad)model\s*\(/,
+        message: '推理/评估函数应使用 @torch.no_grad() 或 with torch.no_grad() — 减少内存消耗并加速',
+      },
+      {
+        ruleId: 'ml-gradient-accumulation-zero',
+        category: 'correctness',
+        dimension: 'file',
+        severity: 'info',
+        languages: ['python'],
+        pattern: /loss\.backward\(\)[\s\S]*?optimizer\.step\(\)[\s\S]*?(?!optimizer\.zero_grad)/,
+        message: 'optimizer.step() 后应调用 optimizer.zero_grad() — 否则梯度会累积',
+      },
+      {
+        ruleId: 'ml-random-seed',
+        category: 'correctness',
+        dimension: 'file',
+        severity: 'info',
+        languages: ['python'],
+        pattern: /torch\.manual_seed|random\.seed|np\.random\.seed/,
+        message: '设置随机种子时建议同时设置 torch.manual_seed / torch.cuda.manual_seed_all / np.random.seed / random.seed 保证完全可复现',
+      },
+    ];
+  }
+
+  detectPatterns(astSummary) {
+    const patterns = [];
+
+    // ── nn.Module subclasses ──
+    for (const cls of astSummary.classes || []) {
+      if (cls.superclass && /Module$|nn\.Module/.test(cls.superclass)) {
+        patterns.push({
+          type: 'pytorch-model',
+          className: cls.name,
+          line: cls.line,
+          confidence: 0.95,
+        });
+      }
+    }
+
+    // ── Dataset subclasses ──
+    for (const cls of astSummary.classes || []) {
+      if (cls.superclass && /Dataset$|IterableDataset/.test(cls.superclass)) {
+        patterns.push({
+          type: 'pytorch-dataset',
+          className: cls.name,
+          line: cls.line,
+          confidence: 0.9,
+        });
+      }
+    }
+
+    // ── HuggingFace model/tokenizer/trainer ──
+    for (const cls of astSummary.classes || []) {
+      if (cls.superclass && /PreTrainedModel|PretrainedConfig|Trainer/.test(cls.superclass)) {
+        patterns.push({
+          type: 'huggingface-model',
+          className: cls.name,
+          line: cls.line,
+          confidence: 0.9,
+        });
+      }
+    }
+
+    // ── Training functions (train/train_one_epoch/train_step) ──
+    for (const m of astSummary.methods || []) {
+      const nameLower = m.name?.toLowerCase() || '';
+      if (
+        nameLower === 'train' ||
+        nameLower === 'train_one_epoch' ||
+        nameLower === 'train_step' ||
+        nameLower === 'training_step' ||
+        nameLower === 'train_loop'
+      ) {
+        patterns.push({
+          type: 'ml-training-function',
+          methodName: m.name,
+          line: m.line,
+          confidence: 0.85,
+        });
+      }
+    }
+
+    // ── Evaluation functions ──
+    for (const m of astSummary.methods || []) {
+      const nameLower = m.name?.toLowerCase() || '';
+      if (
+        nameLower === 'evaluate' ||
+        nameLower === 'validate' ||
+        nameLower === 'eval_step' ||
+        nameLower === 'validation_step' ||
+        nameLower === 'test_step'
+      ) {
+        patterns.push({
+          type: 'ml-evaluation-function',
+          methodName: m.name,
+          line: m.line,
+          confidence: 0.85,
+        });
+      }
+    }
+
+    // ── forward method (nn.Module) ──
+    for (const m of astSummary.methods || []) {
+      if (m.name === 'forward' && m.className) {
+        patterns.push({
+          type: 'pytorch-forward',
+          className: m.className,
+          methodName: m.name,
+          line: m.line,
+          confidence: 0.9,
+        });
+      }
+    }
+
+    // ── Lightning modules ──
+    for (const cls of astSummary.classes || []) {
+      if (cls.superclass && /LightningModule|LightningDataModule/.test(cls.superclass)) {
+        patterns.push({
+          type: 'pytorch-lightning-module',
+          className: cls.name,
+          line: cls.line,
+          confidence: 0.9,
+        });
+      }
+    }
+
+    // ── Loss function classes ──
+    for (const cls of astSummary.classes || []) {
+      const nameLower = cls.name?.toLowerCase() || '';
+      if (nameLower.includes('loss') && cls.superclass && /Module/.test(cls.superclass)) {
+        patterns.push({
+          type: 'ml-custom-loss',
+          className: cls.name,
+          line: cls.line,
+          confidence: 0.85,
+        });
+      }
+    }
+
+    // ── ML ecosystem imports ──
+    const mlImports = (astSummary.imports || []).filter(
+      (imp) =>
+        imp.includes('torch') ||
+        imp.includes('tensorflow') ||
+        imp.includes('transformers') ||
+        imp.includes('datasets') ||
+        imp.includes('accelerate') ||
+        imp.includes('lightning') ||
+        imp.includes('sklearn') ||
+        imp.includes('numpy') ||
+        imp.includes('wandb') ||
+        imp.includes('tensorboard')
+    );
+    if (mlImports.length > 0) {
+      patterns.push({
+        type: 'ml-ecosystem-usage',
+        importCount: mlImports.length,
+        confidence: 0.85,
+      });
+    }
+
+    return patterns;
+  }
+}
+
+export const pack = new MLEnhancement();