AI Code Toolkit

/** * ReviewCodeChange Hook for Claude Code * * DESIGN PATTERNS: * - Class-based hook pattern: Encapsulates lifecycle hooks in a single class * - Fail-open pattern: Errors don't block, just provide warnings * - Single responsibility: Each hook handles specific lifecycle stage * * CODING STANDARDS: * - Export a class with preToolUse, postToolUse methods * - Handle all errors gracefully with fail-open behavior * - Format messages clearly for LLM consumption * * AVOID: * - Blocking operations on errors * - Complex business logic (delegate to tools/services) * - Mutating context object */ import type { ClaudeCodeHookInput, HookResponse } from '@agiflowai/hooks-adapter'; import { ExecutionLogService, DECISION_SKIP, DECISION_DENY, DECISION_ALLOW, } from '@agiflowai/hooks-adapter'; import { CLAUDE_CODE, isValidLlmTool, type LlmToolId } from '@agiflowai/coding-agent-bridge'; import { CodeReviewService } from '../../services/CodeReview'; import { TemplateFinder } from '../../services/TemplateFinder'; import { ArchitectParser } from '../../services/ArchitectParser'; import { PatternMatcher } from '../../services/PatternMatcher'; /** * ReviewCodeChange Hook class for Claude Code * * Provides lifecycle hooks for tool execution: * - preToolUse: Not applicable for this hook (returns skip) * - postToolUse: Reviews code after file edit/write operations */ export class ReviewCodeChangeHook { /** * PreToolUse hook - not applicable for reviewCodeChange * This tool is only called after file operations * * @param _context - Claude Code hook input (unused) * @returns Hook response with skip decision */ async preToolUse(_context: ClaudeCodeHookInput): Promise<HookResponse> { return { decision: DECISION_SKIP, message: 'PreToolUse not applicable for reviewCodeChange', }; } /** * PostToolUse hook for Claude Code * Reviews code after file edit/write operations and provides feedback * * @param context - Claude Code hook input * @returns Hook response with code review feedback or skip */ async postToolUse(context: ClaudeCodeHookInput): Promise<HookResponse> { // Extract file path from tool input const filePath = context.tool_input?.file_path; // Only process file operations if (!filePath) { return { decision: DECISION_SKIP, message: 'Not a file operation', }; } try { // Create execution log service for this session const executionLog = new ExecutionLogService(context.session_id); // Check if file was recently reviewed (debounce within 3 seconds) const wasRecent = await executionLog.wasRecentlyReviewed( filePath, 3000, // 3 seconds debounce ); if (wasRecent) { return { decision: DECISION_SKIP, message: 'File was recently reviewed (within 3 seconds), skipping to avoid noise', }; } // Get matched file patterns for logging const templateFinder = new TemplateFinder(); const architectParser = new ArchitectParser(); const patternMatcher = new PatternMatcher(); const templateMapping = await templateFinder.findTemplateForFile(filePath); const templateConfig = templateMapping ? await architectParser.parseArchitectFile(templateMapping.templatePath) : null; const globalConfig = await architectParser.parseGlobalArchitectFile(); const filePatterns = patternMatcher.getMatchedFilePatterns( filePath, templateConfig, globalConfig, templateMapping?.projectPath, ); // Get current file metadata for change detection const fileMetadata = await executionLog.getFileMetadata(filePath); // Derive operation from tool name const operation = extractOperation(context.tool_name); // Check if file has changed since last review (skip if unchanged) const fileChanged = await executionLog.hasFileChanged( filePath, DECISION_ALLOW, // Check against last successful review ); if (!fileChanged) { return { decision: DECISION_SKIP, message: 'File unchanged since last review', }; } // Execute: Review the code change using service directly // Validate and use context.llm_tool if provided, otherwise fallback to CLAUDE_CODE constant let llmTool: LlmToolId = CLAUDE_CODE; if (context.llm_tool && isValidLlmTool(context.llm_tool)) { llmTool = context.llm_tool; } const service = new CodeReviewService({ llmTool }); const data = await service.reviewCodeChange(filePath); // If fixes are required (must_do or must_not_do violations), block with full response if (data.fix_required) { await executionLog.logExecution({ filePath: filePath, operation: operation, decision: DECISION_DENY, filePattern: filePatterns, fileMtime: fileMetadata?.mtime, fileChecksum: fileMetadata?.checksum, }); return { decision: DECISION_DENY, // Will map to 'block' in PostToolUse output message: JSON.stringify(data, null, 2), // Full AI response }; } // Otherwise (no fix required), provide feedback and issues without blocking // decision: 'allow' means additionalContext is used, not blocking await executionLog.logExecution({ filePath: filePath, operation: operation, decision: DECISION_ALLOW, filePattern: filePatterns, fileMtime: fileMetadata?.mtime, fileChecksum: fileMetadata?.checksum, }); return { decision: DECISION_ALLOW, message: JSON.stringify( { feedback: data.feedback, identified_issues: data.identified_issues, }, null, 2, ), }; } catch (error) { // Fail open: skip hook and let Claude continue return { decision: DECISION_SKIP, message: `⚠️ Hook error: ${error instanceof Error ? error.message : String(error)}`, }; } } } /** * Extract operation type from tool name */ function extractOperation(toolName: string): string { if (toolName === 'Edit' || toolName === 'Update') return 'edit'; if (toolName === 'Write') return 'write'; if (toolName === 'Read') return 'read'; return 'unknown'; }