@theihtisham/agent-shadow-brain 1.2.0 → 3.0.0

package/dist/brain/learning-engine.js

@@ -0,0 +1,855 @@
+// src/brain/learning-engine.ts — Self-Improvement Learning Engine for Shadow Brain
+// Learns from past analysis sessions and gets smarter over time.
+import * as fs from 'fs/promises';
+import * as path from 'path';
+import * as os from 'os';
+import * as crypto from 'crypto';
+// ── Learning Engine ─────────────────────────────────────────────────────────
+export class LearningEngine {
+    constructor(projectDir, llmClient) {
+        this.saveTimer = null;
+        this.projectDir = projectDir;
+        this.llmClient = llmClient ?? null;
+        this.storePath = path.join(os.homedir(), '.shadow-brain', 'learning.json');
+        this.store = {
+            version: 1,
+            lessons: [],
+            falsePositiveRates: {},
+            codePatterns: [],
+            agentPreferences: {},
+            projectKnowledge: {},
+            lastTraining: new Date(),
+        };
+    }
+    // ── Persistence ─────────────────────────────────────────────────────────
+    async load() {
+        try {
+            const dir = path.dirname(this.storePath);
+            await fs.mkdir(dir, { recursive: true });
+            const raw = await fs.readFile(this.storePath, 'utf-8');
+            const parsed = JSON.parse(raw);
+            // Hydrate date strings back to Date objects
+            if (parsed.lastTraining)
+                parsed.lastTraining = new Date(parsed.lastTraining);
+            if (parsed.lessons) {
+                parsed.lessons = parsed.lessons.map((l) => ({
+                    ...l,
+                    lastSeen: new Date(l.lastSeen),
+                }));
+            }
+            if (parsed.codePatterns) {
+                parsed.codePatterns = parsed.codePatterns.map((p) => ({
+                    ...p,
+                    lastSeen: new Date(p.lastSeen),
+                }));
+            }
+            if (parsed.projectKnowledge) {
+                for (const key of Object.keys(parsed.projectKnowledge)) {
+                    parsed.projectKnowledge[key].lastUpdated = new Date(parsed.projectKnowledge[key].lastUpdated);
+                }
+            }
+            this.store = {
+                version: parsed.version ?? 1,
+                lessons: parsed.lessons ?? [],
+                falsePositiveRates: parsed.falsePositiveRates ?? {},
+                codePatterns: parsed.codePatterns ?? [],
+                agentPreferences: parsed.agentPreferences ?? {},
+                projectKnowledge: parsed.projectKnowledge ?? {},
+                lastTraining: parsed.lastTraining ?? new Date(),
+            };
+        }
+        catch {
+            // File doesn't exist or is corrupt — start fresh
+            this.store = {
+                version: 1,
+                lessons: [],
+                falsePositiveRates: {},
+                codePatterns: [],
+                agentPreferences: {},
+                projectKnowledge: {},
+                lastTraining: new Date(),
+            };
+        }
+    }
+    async save() {
+        try {
+            const dir = path.dirname(this.storePath);
+            await fs.mkdir(dir, { recursive: true });
+            const data = JSON.stringify(this.store, null, 2);
+            await fs.writeFile(this.storePath, data, 'utf-8');
+        }
+        catch (err) {
+            console.error(`[LearningEngine] Failed to save store: ${err.message}`);
+        }
+    }
+    scheduleSave() {
+        if (this.saveTimer) {
+            clearTimeout(this.saveTimer);
+        }
+        // Debounce: wait 2 seconds before actually writing to disk
+        this.saveTimer = setTimeout(() => {
+            this.save().catch((err) => {
+                console.error(`[LearningEngine] Debounced save failed: ${err.message}`);
+            });
+        }, 2000);
+    }
+    // ── Learning Methods ────────────────────────────────────────────────────
+    learnFromInsights(insights, changes) {
+        try {
+            const now = new Date();
+            for (const insight of insights) {
+                // Track false positive rates per insight category
+                const category = insight.type;
+                if (!this.store.falsePositiveRates[category]) {
+                    this.store.falsePositiveRates[category] = { reported: 0, dismissed: 0 };
+                }
+                this.store.falsePositiveRates[category].reported++;
+                // Create or update lessons based on patterns
+                for (const change of changes) {
+                    const pattern = this.extractPattern(change);
+                    if (!pattern)
+                        continue;
+                    const existingLesson = this.store.lessons.find((l) => l.pattern === pattern && l.category === category);
+                    if (existingLesson) {
+                        existingLesson.occurrences++;
+                        existingLesson.lastSeen = now;
+                        // Slowly increase confidence when seen repeatedly
+                        existingLesson.confidence = Math.min(1.0, existingLesson.confidence + 0.02);
+                    }
+                    else {
+                        this.store.lessons.push({
+                            id: `lesson-${Date.now()}-${Math.random().toString(36).slice(2, 9)}`,
+                            category,
+                            pattern,
+                            lesson: insight.content.slice(0, 200),
+                            confidence: 0.3,
+                            occurrences: 1,
+                            lastSeen: now,
+                            source: 'rule',
+                        });
+                    }
+                }
+                // Track code patterns
+                for (const change of changes) {
+                    if (!change.path)
+                        continue;
+                    const ext = path.extname(change.path).replace('.', '');
+                    if (!ext)
+                        continue;
+                    const contentPattern = this.extractCodePattern(change.content ?? '');
+                    if (!contentPattern)
+                        continue;
+                    const existingPattern = this.store.codePatterns.find((p) => p.pattern === contentPattern && p.language === ext);
+                    if (existingPattern) {
+                        existingPattern.frequency++;
+                        existingPattern.lastSeen = now;
+                        if (!existingPattern.associatedInsights.includes(insight.type)) {
+                            existingPattern.associatedInsights.push(insight.type);
+                        }
+                    }
+                    else {
+                        this.store.codePatterns.push({
+                            id: `pattern-${Date.now()}-${Math.random().toString(36).slice(2, 9)}`,
+                            language: ext,
+                            pattern: contentPattern,
+                            description: `Observed in ${path.basename(change.path)}`,
+                            frequency: 1,
+                            lastSeen: now,
+                            associatedInsights: [insight.type],
+                        });
+                    }
+                }
+            }
+            this.store.lastTraining = now;
+            this.scheduleSave();
+        }
+        catch (err) {
+            console.error(`[LearningEngine] learnFromInsights error: ${err.message}`);
+        }
+    }
+    learnFromFeedback(insightId, accepted) {
+        try {
+            if (!accepted) {
+                // Find the lesson associated with this insight and reduce confidence
+                const lesson = this.store.lessons.find((l) => l.lesson.slice(0, 50) === insightId.slice(0, 50));
+                if (lesson) {
+                    lesson.confidence = Math.max(0, lesson.confidence - 0.15);
+                }
+                // Track in false positive rates
+                if (!this.store.falsePositiveRates[insightId]) {
+                    this.store.falsePositiveRates[insightId] = { reported: 0, dismissed: 0 };
+                }
+                this.store.falsePositiveRates[insightId].dismissed++;
+            }
+            this.scheduleSave();
+        }
+        catch (err) {
+            console.error(`[LearningEngine] learnFromFeedback error: ${err.message}`);
+        }
+    }
+    async learnProjectPatterns(context, files) {
+        if (!this.llmClient)
+            return;
+        try {
+            const fileSample = files.slice(0, 20).map((f) => path.basename(f)).join(', ');
+            const langList = context.language.join(', ');
+            const prompt = `Analyze this project and extract conventions, patterns, and architecture insights.
+
+Project: ${context.name}
+Languages: ${langList}
+Framework: ${context.framework ?? 'unknown'}
+Files sample: ${fileSample}
+Directory structure: ${context.structure.slice(0, 30).join('\n')}
+
+Return a JSON object with these fields:
+{
+  "conventions": ["list of coding conventions observed, e.g. 'uses arrow functions', 'snake_case file names'"],
+  "architecture": "brief description of the architecture pattern used",
+  "commonPatterns": ["common code patterns found in this project"],
+  "avoidPatterns": ["patterns or practices this project avoids"],
+  "dependencies": ["key dependencies this project relies on"]
+}`;
+            const systemPrompt = 'You are a codebase analysis expert. Extract patterns and conventions from project metadata. Return valid JSON only.';
+            let result;
+            try {
+                result = await this.llmClient.completeWithSchema(prompt, 
+                // Minimal inline schema since we can't use zod here without importing it
+                // The LLMClient.completeWithSchema needs a ZodSchema, so we'll parse manually
+                null, systemPrompt);
+            }
+            catch {
+                // Manual JSON parse fallback
+                const raw = await this.llmClient.complete(prompt, systemPrompt);
+                const cleaned = raw.replace(/```json\n?/g, '').replace(/```\n?/g, '').trim();
+                result = JSON.parse(cleaned);
+            }
+            const knowledge = {
+                name: context.name,
+                conventions: result.conventions ?? [],
+                architecture: result.architecture ?? '',
+                commonPatterns: result.commonPatterns ?? [],
+                avoidPatterns: result.avoidPatterns ?? [],
+                dependencies: result.dependencies ?? [],
+                lastUpdated: new Date(),
+            };
+            this.store.projectKnowledge[context.rootDir] = knowledge;
+            this.scheduleSave();
+        }
+        catch (err) {
+            console.error(`[LearningEngine] learnProjectPatterns error: ${err.message}`);
+        }
+    }
+    // ── Retrieval Methods ───────────────────────────────────────────────────
+    getRelevantLessons(changes) {
+        const now = new Date();
+        const relevant = [];
+        for (const change of changes) {
+            const pattern = this.extractPattern(change);
+            if (!pattern)
+                continue;
+            for (const lesson of this.store.lessons) {
+                // Match by pattern similarity or file extension
+                if (lesson.pattern === pattern ||
+                    change.path.endsWith(`.${lesson.pattern.split(':')[0]}`) ||
+                    lesson.pattern.includes(path.extname(change.path).replace('.', ''))) {
+                    // Boost confidence for recently seen lessons
+                    const daysSinceLastSeen = (now.getTime() - new Date(lesson.lastSeen).getTime()) / (1000 * 60 * 60 * 24);
+                    const recencyBoost = Math.max(0, 1 - daysSinceLastSeen / 30) * 0.1;
+                    relevant.push({
+                        ...lesson,
+                        confidence: Math.min(1.0, lesson.confidence + recencyBoost),
+                    });
+                }
+            }
+        }
+        // Sort by confidence descending, deduplicate
+        const seen = new Set();
+        return relevant
+            .sort((a, b) => b.confidence - a.confidence)
+            .filter((l) => {
+            if (seen.has(l.id))
+                return false;
+            seen.add(l.id);
+            return true;
+        })
+            .slice(0, 20);
+    }
+    getProjectKnowledge(projectDir) {
+        return this.store.projectKnowledge[projectDir] ?? null;
+    }
+    async getIntelligenceBoost(changes, context) {
+        const insights = [];
+        try {
+            // 1. Rule-based insights from learned lessons
+            const relevantLessons = this.getRelevantLessons(changes);
+            for (const lesson of relevantLessons) {
+                if (lesson.confidence < 0.5)
+                    continue;
+                insights.push({
+                    type: 'pattern',
+                    priority: lesson.confidence > 0.8 ? 'high' : lesson.confidence > 0.6 ? 'medium' : 'low',
+                    title: `Learned: ${lesson.category} pattern detected`,
+                    content: lesson.lesson,
+                    files: changes.map((c) => c.path),
+                    timestamp: new Date(),
+                });
+            }
+            // 2. Knowledge-based insights from project conventions
+            const knowledge = this.getProjectKnowledge(context.rootDir);
+            if (knowledge) {
+                for (const change of changes) {
+                    if (!change.content)
+                        continue;
+                    // Check if changes violate known avoid patterns
+                    for (const avoid of knowledge.avoidPatterns) {
+                        const patternLower = avoid.toLowerCase();
+                        const contentLower = change.content.toLowerCase();
+                        if (contentLower.includes(patternLower)) {
+                            insights.push({
+                                type: 'warning',
+                                priority: 'medium',
+                                title: `Project convention violation: ${avoid}`,
+                                content: `This project avoids "${avoid}". Consider refactoring in ${path.basename(change.path)}.`,
+                                files: [change.path],
+                                timestamp: new Date(),
+                            });
+                        }
+                    }
+                }
+                // Check for missing conventions
+                for (const convention of knowledge.conventions) {
+                    const filesWithoutConvention = changes.filter((c) => {
+                        if (!c.content)
+                            return false;
+                        return !this.contentFollowsConvention(c.content, convention);
+                    });
+                    if (filesWithoutConvention.length > 0) {
+                        insights.push({
+                            type: 'suggestion',
+                            priority: 'low',
+                            title: `Convention check: ${convention}`,
+                            content: `Some files may not follow project convention "${convention}": ${filesWithoutConvention.map((f) => path.basename(f.path)).join(', ')}`,
+                            files: filesWithoutConvention.map((f) => f.path),
+                            timestamp: new Date(),
+                        });
+                    }
+                }
+            }
+            // 3. LLM-powered super insights (if client available)
+            if (this.llmClient && changes.length > 0) {
+                const llmInsights = await this.generateLLMInsights(changes, context, knowledge);
+                insights.push(...llmInsights);
+            }
+        }
+        catch (err) {
+            console.error(`[LearningEngine] getIntelligenceBoost error: ${err.message}`);
+        }
+        return insights;
+    }
+    getStats() {
+        const lessons = this.store.lessons;
+        const totalConfidence = lessons.reduce((sum, l) => sum + l.confidence, 0);
+        const avgConfidence = lessons.length > 0 ? totalConfidence / lessons.length : 0;
+        return {
+            totalLessons: lessons.length,
+            totalPatterns: this.store.codePatterns.length,
+            avgConfidence: Math.round(avgConfidence * 100) / 100,
+            projectCount: Object.keys(this.store.projectKnowledge).length,
+        };
+    }
+    // ── Orchestrator-Facing Methods ────────────────────────────────────────────
+    /**
+     * learnFromProject — Scan the project for patterns and extract lessons.
+     * Called by the orchestrator during a learning cycle.
+     */
+    async learnFromProject() {
+        try {
+            const srcDir = path.join(this.projectDir, 'src');
+            const scanDir = await this.dirExists(srcDir) ? srcDir : this.projectDir;
+            const sourceFiles = await this.collectSourceFiles(scanDir, 100);
+            if (sourceFiles.length === 0)
+                return;
+            const now = new Date();
+            for (const filePath of sourceFiles) {
+                try {
+                    const content = await fs.readFile(filePath, 'utf-8');
+                    const ext = path.extname(filePath).replace('.', '');
+                    // Extract code patterns from each file
+                    const patternKey = this.extractCodePattern(content);
+                    if (!patternKey)
+                        continue;
+                    // Record or update each individual pattern
+                    const individualPatterns = patternKey.split('+');
+                    for (const pat of individualPatterns) {
+                        const existing = this.store.codePatterns.find((p) => p.pattern === pat && p.language === ext);
+                        if (existing) {
+                            existing.frequency++;
+                            existing.lastSeen = now;
+                        }
+                        else {
+                            this.store.codePatterns.push({
+                                id: `pattern-${Date.now()}-${crypto.randomUUID().slice(0, 8)}`,
+                                language: ext,
+                                pattern: pat,
+                                description: `Observed in ${path.relative(this.projectDir, filePath)}`,
+                                frequency: 1,
+                                lastSeen: now,
+                                associatedInsights: [],
+                            });
+                        }
+                    }
+                    // Create lessons for notable patterns
+                    this.createLessonFromContent(content, filePath, ext, now);
+                }
+                catch {
+                    // Skip unreadable files
+                }
+            }
+            // If LLM is available, enrich lessons with AI analysis
+            if (this.llmClient && this.store.codePatterns.length > 0) {
+                try {
+                    await this.enrichLessonsWithLLM();
+                }
+                catch (err) {
+                    console.error(`[LearningEngine] LLM enrichment failed: ${err.message}`);
+                }
+            }
+            this.store.lastTraining = now;
+            this.scheduleSave();
+        }
+        catch (err) {
+            console.error(`[LearningEngine] learnFromProject error: ${err.message}`);
+        }
+    }
+    /**
+     * getLessons — Return all learned lessons in the format the orchestrator expects.
+     */
+    async getLessons() {
+        return this.store.lessons
+            .sort((a, b) => b.confidence - a.confidence)
+            .map((l) => ({
+            category: l.category,
+            pattern: l.pattern,
+            lesson: l.lesson,
+            confidence: l.confidence,
+        }));
+    }
+    /**
+     * recordInsight — Record a BrainInsight as a learned lesson.
+     * Called by the orchestrator when critical/high-priority insights are generated.
+     */
+    async recordInsight(insight) {
+        try {
+            const now = new Date();
+            // Create a lesson from the insight
+            const pattern = insight.files && insight.files.length > 0
+                ? `${path.extname(insight.files[0]).replace('.', '') || 'general'}:${insight.type}`
+                : `general:${insight.type}`;
+            const existing = this.store.lessons.find((l) => l.pattern === pattern && l.lesson === insight.content.slice(0, 300));
+            if (existing) {
+                existing.occurrences++;
+                existing.lastSeen = now;
+                // Boost confidence slightly for repeated observations
+                existing.confidence = Math.min(1.0, existing.confidence + 0.05);
+            }
+            else {
+                this.store.lessons.push({
+                    id: `lesson-${Date.now()}-${crypto.randomUUID().slice(0, 8)}`,
+                    category: insight.type,
+                    pattern,
+                    lesson: insight.content.slice(0, 300),
+                    confidence: insight.priority === 'critical' ? 0.7 : insight.priority === 'high' ? 0.5 : 0.3,
+                    occurrences: 1,
+                    lastSeen: now,
+                    source: 'rule',
+                });
+            }
+            // Update false positive tracking
+            if (!this.store.falsePositiveRates[insight.type]) {
+                this.store.falsePositiveRates[insight.type] = { reported: 0, dismissed: 0 };
+            }
+            this.store.falsePositiveRates[insight.type].reported++;
+            this.store.lastTraining = now;
+            this.scheduleSave();
+        }
+        catch (err) {
+            console.error(`[LearningEngine] recordInsight error: ${err.message}`);
+        }
+    }
+    // ── Cleanup ─────────────────────────────────────────────────────────────
+    cleanup() {
+        try {
+            const now = new Date();
+            const sixtyDaysAgo = new Date(now.getTime() - 60 * 24 * 60 * 60 * 1000);
+            // Remove low-confidence lessons
+            this.store.lessons = this.store.lessons.filter((l) => l.confidence >= 0.1);
+            // Remove stale code patterns (older than 60 days)
+            this.store.codePatterns = this.store.codePatterns.filter((p) => new Date(p.lastSeen) > sixtyDaysAgo);
+            this.scheduleSave();
+        }
+        catch (err) {
+            console.error(`[LearningEngine] cleanup error: ${err.message}`);
+        }
+    }
+    // ── Private Helpers ─────────────────────────────────────────────────────
+    async dirExists(dirPath) {
+        try {
+            const stat = await fs.stat(dirPath);
+            return stat.isDirectory();
+        }
+        catch {
+            return false;
+        }
+    }
+    async collectSourceFiles(dir, limit) {
+        const results = [];
+        const skipDirs = new Set([
+            'node_modules', '.git', 'dist', 'build', 'coverage',
+            '.next', '.nuxt', '.cache', '.turbo', '__pycache__',
+            'target', 'vendor', '.shadow-brain',
+        ]);
+        const sourceExtensions = new Set([
+            '.ts', '.tsx', '.js', '.jsx', '.mjs', '.cjs',
+            '.py', '.go', '.rs', '.java', '.rb',
+        ]);
+        const queue = [dir];
+        while (queue.length > 0 && results.length < limit) {
+            const current = queue.shift();
+            let entries;
+            try {
+                const dirents = await fs.readdir(current, { withFileTypes: true });
+                entries = dirents.map((e) => ({
+                    name: e.name,
+                    fullPath: path.join(current, e.name),
+                    isDir: e.isDirectory(),
+                }));
+            }
+            catch {
+                continue;
+            }
+            for (const entry of entries) {
+                if (results.length >= limit)
+                    break;
+                if (entry.isDir) {
+                    if (!entry.name.startsWith('.') && !skipDirs.has(entry.name)) {
+                        queue.push(entry.fullPath);
+                    }
+                }
+                else {
+                    const ext = path.extname(entry.name);
+                    if (sourceExtensions.has(ext)) {
+                        results.push(entry.fullPath);
+                    }
+                }
+            }
+        }
+        return results;
+    }
+    createLessonFromContent(content, filePath, ext, now) {
+        const relPath = path.relative(this.projectDir, filePath);
+        const lessons = [];
+        // Security anti-patterns
+        if (/\beval\s*\(/.test(content)) {
+            lessons.push({
+                category: 'security',
+                lesson: `Avoid eval() in ${relPath} — it introduces code injection risks`,
+                pattern: `eval:${ext}`,
+            });
+        }
+        if (/innerHTML\s*=/.test(content) && !/sanitize|escape|DOMPurify/.test(content)) {
+            lessons.push({
+                category: 'security',
+                lesson: `Direct innerHTML assignment in ${relPath} without sanitization — XSS risk`,
+                pattern: `innerHTML:${ext}`,
+            });
+        }
+        // TypeScript quality
+        if (ext === 'ts' || ext === 'tsx') {
+            const anyMatches = content.match(/:\s*any\b/g);
+            if (anyMatches && anyMatches.length > 3) {
+                lessons.push({
+                    category: 'quality',
+                    lesson: `${relPath} uses 'any' type ${anyMatches.length} times — prefer specific types`,
+                    pattern: `excessive-any:${ext}`,
+                });
+            }
+            if (/\bas\s+any/.test(content)) {
+                lessons.push({
+                    category: 'quality',
+                    lesson: `Type assertion 'as any' found in ${relPath} — defeats type safety`,
+                    pattern: `as-any:${ext}`,
+                });
+            }
+        }
+        // Error handling
+        if (/catch\s*\(\s*\w*\s*\)\s*\{\s*\}/.test(content)) {
+            lessons.push({
+                category: 'quality',
+                lesson: `Empty catch block in ${relPath} — silently swallowing errors`,
+                pattern: `empty-catch:${ext}`,
+            });
+        }
+        // Performance
+        if (/\.forEach\s*\(/.test(content) && /await/.test(content)) {
+            lessons.push({
+                category: 'performance',
+                lesson: `Async operations inside forEach in ${relPath} — use for...of or Promise.all instead`,
+                pattern: `async-foreach:${ext}`,
+            });
+        }
+        // Maintainability
+        const lines = content.split('\n').length;
+        if (lines > 500) {
+            lessons.push({
+                category: 'maintainability',
+                lesson: `${relPath} is ${lines} lines long — consider splitting into smaller modules`,
+                pattern: `large-file:${ext}`,
+            });
+        }
+        // Add lessons to the store
+        for (const { category, lesson, pattern } of lessons) {
+            const existing = this.store.lessons.find((l) => l.pattern === pattern && l.category === category);
+            if (existing) {
+                existing.occurrences++;
+                existing.lastSeen = now;
+                existing.confidence = Math.min(1.0, existing.confidence + 0.03);
+            }
+            else {
+                this.store.lessons.push({
+                    id: `lesson-${Date.now()}-${crypto.randomUUID().slice(0, 8)}`,
+                    category,
+                    pattern,
+                    lesson,
+                    confidence: 0.4,
+                    occurrences: 1,
+                    lastSeen: now,
+                    source: 'rule',
+                });
+            }
+        }
+    }
+    async enrichLessonsWithLLM() {
+        if (!this.llmClient)
+            return;
+        // Pick the top patterns to analyze
+        const topPatterns = this.store.codePatterns
+            .sort((a, b) => b.frequency - a.frequency)
+            .slice(0, 10);
+        const patternSummary = topPatterns
+            .map((p) => `- [${p.language}] ${p.pattern} (frequency: ${p.frequency})`)
+            .join('\n');
+        const lessonSummary = this.store.lessons
+            .slice(0, 10)
+            .map((l) => `- [${l.category}] ${l.lesson}`)
+            .join('\n');
+        const prompt = `Given these code patterns and existing lessons from a project, suggest additional lessons that a senior developer would know.
+
+Observed patterns:
+${patternSummary}
+
+Existing lessons:
+${lessonSummary || 'None yet.'}
+
+Return a JSON array of lesson objects:
+[{
+  "category": "security" | "performance" | "quality" | "architecture" | "maintainability",
+  "pattern": "short pattern key",
+  "lesson": "actionable lesson text",
+  "confidence": 0.5
+}]
+
+Provide 3-5 non-obvious lessons. Do NOT repeat existing ones.`;
+        const systemPrompt = 'You are an expert code reviewer. Suggest non-obvious lessons from code patterns. Return valid JSON array only.';
+        try {
+            const response = await this.llmClient.complete(prompt, systemPrompt);
+            const cleaned = response.replace(/```json\n?/g, '').replace(/```\n?/g, '').trim();
+            let parsed;
+            try {
+                parsed = JSON.parse(cleaned);
+            }
+            catch {
+                const match = cleaned.match(/\[[\s\S]*\]/);
+                if (match) {
+                    parsed = JSON.parse(match[0]);
+                }
+                else {
+                    return;
+                }
+            }
+            if (!Array.isArray(parsed))
+                return;
+            const now = new Date();
+            for (const item of parsed) {
+                if (!item.category || !item.pattern || !item.lesson)
+                    continue;
+                const existing = this.store.lessons.find((l) => l.pattern === item.pattern && l.category === item.category);
+                if (existing)
+                    continue; // Don't overwrite
+                this.store.lessons.push({
+                    id: `lesson-llm-${Date.now()}-${crypto.randomUUID().slice(0, 8)}`,
+                    category: item.category,
+                    pattern: item.pattern,
+                    lesson: item.lesson,
+                    confidence: item.confidence ?? 0.5,
+                    occurrences: 1,
+                    lastSeen: now,
+                    source: 'llm',
+                });
+            }
+        }
+        catch {
+            // LLM enrichment is best-effort
+        }
+    }
+    extractPattern(change) {
+        if (!change.path)
+            return null;
+        const ext = path.extname(change.path).replace('.', '');
+        const type = change.type;
+        // Create a pattern key from file extension + change type
+        if (ext) {
+            return `${ext}:${type}`;
+        }
+        // Fallback: use directory-based pattern
+        const dir = path.dirname(change.path).split(path.sep).slice(-2).join('/');
+        return `dir:${dir}:${type}`;
+    }
+    extractCodePattern(content) {
+        if (!content || content.length < 10)
+            return null;
+        // Detect common patterns in code content
+        const patterns = [];
+        // Arrow functions
+        if (/=>\s*\{/.test(content))
+            patterns.push('arrow-fn');
+        // async/await
+        if (/async\s+/.test(content) && /await\s+/.test(content))
+            patterns.push('async-await');
+        // try/catch
+        if (/try\s*\{/.test(content) && /catch\s*\(/.test(content))
+            patterns.push('try-catch');
+        // console.log
+        if (/console\.log/.test(content))
+            patterns.push('console-log');
+        // TODO/FIXME
+        if (/TODO|FIXME|HACK|XXX/.test(content))
+            patterns.push('todo-comment');
+        // class-based
+        if (/class\s+\w+/.test(content))
+            patterns.push('class-based');
+        // export default
+        if (/export\s+default/.test(content))
+            patterns.push('export-default');
+        // named exports
+        if (/export\s+(const|function|class|interface|type)\s/.test(content))
+            patterns.push('named-export');
+        // type assertions
+        if (/as\s+\w+/.test(content))
+            patterns.push('type-assertion');
+        // null checks
+        if (/[!=]==?\s*null|[!=]==?\s*undefined|\?\.|!\./.test(content))
+            patterns.push('null-check');
+        // Promises
+        if (/new\s+Promise|\.then\(|\.catch\(/.test(content))
+            patterns.push('promise');
+        // Error throwing
+        if (/throw\s+new\s+/.test(content))
+            patterns.push('throw-error');
+        return patterns.length > 0 ? patterns.join('+') : null;
+    }
+    contentFollowsConvention(content, convention) {
+        const lower = convention.toLowerCase();
+        if (lower.includes('arrow function') && !/=>/.test(content))
+            return false;
+        if (lower.includes('camelcase') && /_\w/.test(content) && !/^[A-Z]/.test(content))
+            return true;
+        if (lower.includes('snake_case') && /[a-z][A-Z]/.test(content))
+            return false;
+        if (lower.includes('semicolon') && /[^;{}]\s*$/m.test(content))
+            return false;
+        // Default: assume it follows
+        return true;
+    }
+    async generateLLMInsights(changes, context, knowledge) {
+        if (!this.llmClient)
+            return [];
+        try {
+            const changeSummaries = changes
+                .slice(0, 10)
+                .map((c) => `- ${c.type} ${c.path}${c.content ? ` (${c.content.split('\n').length} lines)` : ''}`)
+                .join('\n');
+            const knowledgeContext = knowledge
+                ? `Known conventions: ${knowledge.conventions.join(', ')}
+Architecture: ${knowledge.architecture}
+Avoid: ${knowledge.avoidPatterns.join(', ')}`
+                : 'No prior knowledge about this project.';
+            const lessonsContext = this.store.lessons
+                .filter((l) => l.confidence >= 0.6)
+                .slice(0, 10)
+                .map((l) => `- [${l.category}] ${l.lesson} (confidence: ${l.confidence.toFixed(2)})`)
+                .join('\n');
+            const prompt = `You are a senior code reviewer AI that has been learning from past code reviews.
+
+Analyze these file changes and provide insights that a standard rule-based linter would MISS:
+
+Project: ${context.name}
+Languages: ${context.language.join(', ')}
+Framework: ${context.framework ?? 'unknown'}
+
+Changes:
+${changeSummaries}
+
+${knowledgeContext}
+
+Previously learned lessons:
+${lessonsContext || 'No lessons learned yet.'}
+
+Return a JSON array of insights. Each insight has:
+{
+  "type": "review" | "suggestion" | "warning",
+  "priority": "critical" | "high" | "medium" | "low",
+  "title": "short title",
+  "content": "detailed explanation of the insight with actionable advice",
+  "files": ["affected file paths"]
+}
+
+Focus on: architectural concerns, maintainability, hidden bugs, performance pitfalls, security gotchas that require human-like understanding.
+Do NOT report obvious linting issues — only things that require deep understanding.`;
+            const systemPrompt = 'You are an expert code reviewer AI. Provide deep, non-obvious insights about code changes. Return a valid JSON array. No markdown fences.';
+            const response = await this.llmClient.complete(prompt, systemPrompt);
+            const cleaned = response.replace(/```json\n?/g, '').replace(/```\n?/g, '').trim();
+            let parsed;
+            try {
+                parsed = JSON.parse(cleaned);
+            }
+            catch {
+                // Try to extract JSON array from response
+                const match = cleaned.match(/\[[\s\S]*\]/);
+                if (match) {
+                    parsed = JSON.parse(match[0]);
+                }
+                else {
+                    return [];
+                }
+            }
+            if (!Array.isArray(parsed))
+                return [];
+            return parsed
+                .filter((item) => item.title && item.content)
+                .map((item) => ({
+                type: item.type ?? 'suggestion',
+                priority: item.priority ?? 'medium',
+                title: item.title,
+                content: item.content,
+                files: Array.isArray(item.files) ? item.files : [],
+                timestamp: new Date(),
+            }))
+                .slice(0, 10);
+        }
+        catch (err) {
+            console.error(`[LearningEngine] LLM insight generation failed: ${err.message}`);
+            return [];
+        }
+    }
+}
+//# sourceMappingURL=learning-engine.js.map