specweave 1.0.592 → 2.0.0

package/dist/src/core/reflection/reflect-handler.js

@@ -1,742 +0,0 @@
-/**
- * Reflect Handler - Reflection system with semantic deduplication
- *
- * ARCHITECTURE (v4.0):
- * - PRIMARY: .specweave/skill-memories/{skill}.md (for KNOWN_SKILLS)
- * - OVERFLOW: CLAUDE.md Skill Memories section (for unknown skills only)
- * - LLM-based semantic deduplication (existing memories in extraction prompt)
- * - Auto-pruning of old learnings
- *
- * WHAT IT REMEMBERS:
- * - SpecWeave workflow preferences (how user uses SpecWeave)
- * - Skill-specific learnings (how to improve skill behavior)
- * - Project-specific context (tech stack preferences, conventions)
- *
- * WHAT IT DOES NOT REMEMBER:
- * - Generic coding patterns (use Zustand, prefer hooks, etc.)
- * - Implementation details unrelated to SpecWeave
- *
- * @module core/reflection/reflect-handler
- */
-import * as fs from 'fs';
-import * as path from 'path';
-import { runClaudeCli, parseJsonFromOutput } from '../../utils/claude-cli-runner.js';
-import { consoleLogger as logger } from '../../utils/logger.js';
-import { writeSkillMemories, readAllSkillMemories, formatMemoriesForPrompt, } from './skill-memories.js';
-/**
- * Default configuration
- */
-export const DEFAULT_REFLECT_CONFIG = {
-    enabled: true,
-    model: 'haiku',
-    maxLearningsPerSession: 3,
-    autoNudge: true,
-};
-/**
- * Known SpecWeave skills/plugins for validation
- */
-const KNOWN_SKILLS = [
-    'mobile',
-    'frontend',
-    'backend',
-    'testing',
-    'infrastructure',
-    'kubernetes',
-    'architect',
-    'qa-engineer',
-    'security',
-    'docs-writer',
-    'performance',
-    'tdd-cycle',
-    'pm',
-    'devops',
-    'payments',
-    'ml',
-    'kafka',
-    'confluent',
-    'github',
-    'jira',
-    'ado',
-    'release',
-    'diagrams',
-    'general', // fallback for general SpecWeave learnings
-];
-/**
- * Build the LLM prompt for extracting skill learnings
- *
- * v4.0: Includes existing memories for semantic deduplication.
- * The LLM will skip learnings that are semantically similar to existing ones.
- */
-function buildExtractionPrompt(transcript, existingMemoriesStr) {
-    const existingSection = existingMemoriesStr
-        ? `
-=== EXISTING LEARNINGS (DO NOT DUPLICATE - skip if semantically similar) ===
-${existingMemoriesStr}
-=== END EXISTING ===
-
-CRITICAL DEDUP RULE: If a potential learning is semantically similar to ANY existing learning above, DO NOT include it.
-Semantic similarity means: same core idea, even if worded differently.
-
-`
-        : '';
-    return `You are analyzing a Claude Code session transcript to extract ONLY high-value, NEW learnings.
-
-CRITICAL: Be VERY selective. Most sessions have 0-1 learnings worth saving. Only extract if:
-1. User EXPLICITLY corrected Claude's behavior ("No, don't X", "Wrong, do Y instead")
-2. User stated a PERMANENT preference ("Always X", "Never Y", "For this project...")
-3. Learning will CHANGE how a skill behaves in FUTURE sessions
-4. Learning is NOT semantically similar to any existing learning
-${existingSection}
-WHAT TO EXTRACT (only if HIGH confidence it's reusable AND truly new):
-1. **User corrections**: Claude did X, user said "No, do Y instead"
-   - Example: "pm: Enable interview process during increment creation" (user corrected workflow)
-   - Example: "architect: Skip ADR for hotfixes" (user said don't create ADRs for hotfixes)
-
-2. **Explicit project preferences**: User stated how THIS PROJECT differs
-   - Example: "testing: Run E2E on port 8081" (project-specific config)
-   - Example: "frontend: Use Tailwind, not styled-components" (tech choice for project)
-
-3. **Workflow corrections**: User corrected SpecWeave command usage
-   - Example: "general: Run sw:validate before sw:done" (workflow preference)
-
-WHAT NOT TO EXTRACT (be strict!):
-- Anything semantically similar to existing learnings (check above!)
-- Generic coding advice (TypeScript patterns, React hooks, etc.)
-- One-time fixes that won't recur
-- Implementation details (specific file paths, variable names)
-- Things Claude did correctly (no correction = no learning)
-- Vague feedback ("good job", "that works") - these are NOT learnings
-- Standard best practices that any developer would know
-
-SKILL CATEGORIES (use these exact names):
-mobile, frontend, backend, testing, infrastructure, kubernetes, architect,
-qa-engineer, security, docs-writer, performance, tdd-cycle,
-pm, devops, payments, ml, kafka, confluent, github, jira, ado, release,
-diagrams, general
-
-IMPORTANCE TEST: Before including a learning, ask:
-1. "Will this CHANGE how Claude behaves in a FUTURE session with this skill?"
-2. "Is this semantically different from ALL existing learnings?"
-If either answer is NO → don't include it.
-
-Respond with ONLY valid JSON (no markdown, no explanation):
-{
-  "skillLearnings": [
-    { "skill": "pm", "learning": "Enable interview process during increment creation" }
-  ]
-}
-
-If NO high-value NEW learnings found (most sessions), return: {"skillLearnings": []}
-
-=== SESSION TRANSCRIPT ===
-${transcript.length > 8000 ? '... (earlier content truncated)\n' : ''}${transcript.slice(-8000)}
-=== END TRANSCRIPT ===`;
-}
-/**
- * Extract learnings from transcript using LLM
- *
- * @param transcript - Session transcript
- * @param model - Claude model to use
- * @param existingMemoriesStr - Formatted existing memories for dedup (optional)
- */
-async function extractLearningsViaLLM(transcript, model, existingMemoriesStr = '') {
-    const prompt = buildExtractionPrompt(transcript, existingMemoriesStr);
-    const result = runClaudeCli({
-        model,
-        prompt,
-        timeoutMs: 45000, // 45s timeout for extraction
-    });
-    if (!result.success) {
-        return {
-            success: false,
-            learnings: [],
-            error: result.error,
-            durationMs: result.durationMs,
-        };
-    }
-    const parsed = parseJsonFromOutput(result.stdout);
-    if (!parsed.success || !parsed.data) {
-        return {
-            success: false,
-            learnings: [],
-            error: parsed.error || 'Failed to parse LLM response',
-            durationMs: result.durationMs,
-        };
-    }
-    // Validate and filter learnings
-    // Keep ALL skills (known + unknown) - unknown skills go to CLAUDE.md overflow
-    const validLearnings = (parsed.data.skillLearnings || [])
-        .filter((l) => l.skill && l.learning)
-        .slice(0, 5); // Max 5 learnings per extraction
-    return {
-        success: true,
-        learnings: validLearnings,
-        durationMs: result.durationMs,
-    };
-}
-/**
- * Find CLAUDE.md in project
- */
-function findClaudeMd(projectRoot) {
-    const possiblePaths = [
-        path.join(projectRoot, 'CLAUDE.md'),
-        path.join(projectRoot, 'claude.md'),
-    ];
-    for (const p of possiblePaths) {
-        if (fs.existsSync(p)) {
-            return p;
-        }
-    }
-    return null;
-}
-/**
- * Parse existing Skill Memories section from CLAUDE.md
- */
-function parseSkillMemories(content) {
-    const memories = new Map();
-    // Find the Skill Memories section
-    const sectionMatch = content.match(/## Skill Memories\s*\n([\s\S]*?)(?=\n## |\n---|\n<!-- SW:|$)/);
-    if (!sectionMatch) {
-        return memories;
-    }
-    const sectionContent = sectionMatch[1];
-    // Parse each skill subsection
-    const skillMatches = sectionContent.matchAll(/### (\w+[-\w]*)\s*\n([\s\S]*?)(?=\n### |\n## |$)/g);
-    for (const match of skillMatches) {
-        const skillName = match[1].toLowerCase();
-        const skillContent = match[2];
-        // Extract bullet points
-        const bulletPoints = skillContent
-            .split('\n')
-            .filter((line) => line.trim().startsWith('- '))
-            .map((line) => line.trim().replace(/^- \*\*\d{4}-\d{2}-\d{2}\*\*:\s*/, '').trim());
-        if (bulletPoints.length > 0) {
-            memories.set(skillName, bulletPoints);
-        }
-    }
-    return memories;
-}
-/**
- * Check if a learning is a duplicate
- */
-function isDuplicate(existing, newLearning) {
-    const normalizedNew = newLearning.toLowerCase().trim();
-    for (const e of existing) {
-        const normalizedExisting = e.toLowerCase().trim();
-        // Exact match
-        if (normalizedExisting === normalizedNew)
-            return true;
-        // Substring match (one contains the other)
-        if (normalizedExisting.includes(normalizedNew) || normalizedNew.includes(normalizedExisting)) {
-            return true;
-        }
-    }
-    return false;
-}
-/**
- * Generate the Skill Memories section content
- */
-function generateSkillMemoriesSection(memories) {
-    if (memories.size === 0) {
-        return '';
-    }
-    const lines = [];
-    lines.push('## Skill Memories');
-    lines.push('');
-    lines.push('<!-- Auto-captured by SpecWeave reflect. Edit or delete as needed. -->');
-    lines.push('');
-    // Sort skills alphabetically, but put 'general' last
-    const sortedSkills = [...memories.keys()].sort((a, b) => {
-        if (a === 'general')
-            return 1;
-        if (b === 'general')
-            return -1;
-        return a.localeCompare(b);
-    });
-    for (const skill of sortedSkills) {
-        const learnings = memories.get(skill) || [];
-        if (learnings.length === 0)
-            continue;
-        const skillTitle = skill
-            .split('-')
-            .map((w) => w.charAt(0).toUpperCase() + w.slice(1))
-            .join(' ');
-        lines.push(`### ${skillTitle}`);
-        for (const learning of learnings) {
-            // Add date prefix if not present
-            if (!learning.match(/^\*\*\d{4}-\d{2}-\d{2}\*\*/)) {
-                const today = new Date().toISOString().split('T')[0];
-                lines.push(`- **${today}**: ${learning}`);
-            }
-            else {
-                lines.push(`- ${learning}`);
-            }
-        }
-        lines.push('');
-    }
-    return lines.join('\n');
-}
-/**
- * Find the USER's Skill Memories section (NOT the template section)
- *
- * Template sections are inside <!-- SW:SECTION:* --> markers and get overwritten
- * by instruction updates. We need to find/create a section OUTSIDE these markers.
- */
-function findUserSkillMemoriesSection(content) {
-    // Find all ## Skill Memories occurrences
-    const regex = /## Skill Memories/g;
-    let match;
-    const positions = [];
-    while ((match = regex.exec(content)) !== null) {
-        positions.push(match.index);
-    }
-    // For each position, check if it's inside a template section
-    for (const pos of positions) {
-        // Look backwards for <!-- SW:SECTION: marker
-        const beforeContent = content.slice(0, pos);
-        const lastSectionStart = beforeContent.lastIndexOf('<!-- SW:SECTION:');
-        const lastSectionEnd = beforeContent.lastIndexOf('<!-- SW:END:');
-        // If there's a section start AFTER the last section end, we're inside a template
-        if (lastSectionStart > lastSectionEnd) {
-            // This is inside a template section, skip it
-            continue;
-        }
-        // This is a user section (not inside template markers)
-        // Find where this section ends
-        const afterContent = content.slice(pos);
-        const sectionEndMatch = afterContent.match(/\n(?=## |\n---|\n<!-- SW:SECTION:|$)/);
-        const endOffset = sectionEndMatch ? sectionEndMatch.index : afterContent.length;
-        return {
-            found: true,
-            startIndex: pos,
-            endIndex: pos + endOffset,
-            sectionContent: content.slice(pos, pos + endOffset),
-        };
-    }
-    return { found: false, startIndex: -1, endIndex: -1, sectionContent: '' };
-}
-/**
- * Update CLAUDE.md with new learnings
- *
- * IMPORTANT: Writes to the USER's section, not the template section.
- * Template sections (inside <!-- SW:SECTION: --> markers) are overwritten
- * by instruction updates, so we must write elsewhere.
- */
-function updateClaudeMd(claudeMdPath, newLearnings) {
-    let content = fs.readFileSync(claudeMdPath, 'utf-8');
-    // Find user's section (not template)
-    const userSection = findUserSkillMemoriesSection(content);
-    // Parse existing memories from user section only
-    const existingMemories = userSection.found
-        ? parseSkillMemories(userSection.sectionContent)
-        : new Map();
-    let added = 0;
-    let skipped = 0;
-    // Add new learnings
-    for (const learning of newLearnings) {
-        const skill = learning.skill.toLowerCase();
-        const existing = existingMemories.get(skill) || [];
-        if (isDuplicate(existing, learning.learning)) {
-            skipped++;
-            continue;
-        }
-        existing.push(learning.learning);
-        existingMemories.set(skill, existing);
-        added++;
-    }
-    if (added === 0) {
-        return { added: 0, skipped };
-    }
-    // Generate new section content
-    const newSectionContent = generateSkillMemoriesSection(existingMemories);
-    if (userSection.found) {
-        // Replace user's existing section
-        content =
-            content.slice(0, userSection.startIndex) +
-                newSectionContent.trimEnd() +
-                content.slice(userSection.endIndex);
-    }
-    else {
-        // No user section found - create one after the template section
-        // Look for <!-- SW:END:reflect --> marker
-        const reflectEndMatch = content.match(/<!-- SW:END:reflect -->\s*\n/);
-        if (reflectEndMatch) {
-            const insertPos = reflectEndMatch.index + reflectEndMatch[0].length;
-            content =
-                content.slice(0, insertPos) +
-                    '\n' +
-                    newSectionContent +
-                    '\n' +
-                    content.slice(insertPos);
-        }
-        else {
-            // Fallback: insert before "original" section or append at end
-            const insertMatch = content.match(/\n---\s*\n<!-- ↓ ORIGINAL ↓ -->/);
-            if (insertMatch) {
-                content = content.replace(insertMatch[0], `\n${newSectionContent}\n${insertMatch[0]}`);
-            }
-            else {
-                content = content.trimEnd() + '\n\n' + newSectionContent;
-            }
-        }
-    }
-    fs.writeFileSync(claudeMdPath, content);
-    return { added, skipped };
-}
-/**
- * Read reflect configuration from project config
- */
-export function readReflectConfig(projectRoot) {
-    const configPath = path.join(projectRoot, '.specweave', 'config.json');
-    const config = { ...DEFAULT_REFLECT_CONFIG };
-    if (!fs.existsSync(configPath)) {
-        return config;
-    }
-    try {
-        const configContent = JSON.parse(fs.readFileSync(configPath, 'utf-8'));
-        if (configContent.reflect) {
-            if (typeof configContent.reflect.enabled === 'boolean') {
-                config.enabled = configContent.reflect.enabled;
-            }
-            if (configContent.reflect.model && ['haiku', 'sonnet', 'opus'].includes(configContent.reflect.model)) {
-                config.model = configContent.reflect.model;
-            }
-            if (typeof configContent.reflect.maxLearningsPerSession === 'number') {
-                config.maxLearningsPerSession = configContent.reflect.maxLearningsPerSession;
-            }
-            if (typeof configContent.reflect.autoNudge === 'boolean') {
-                config.autoNudge = configContent.reflect.autoNudge;
-            }
-        }
-    }
-    catch {
-        // Ignore config read errors
-    }
-    return config;
-}
-/**
- * Main reflect handler - called at session end
- *
- * ARCHITECTURE (v4.0):
- * 1. Check config (enabled?)
- * 2. Read existing memories for LLM context (semantic dedup)
- * 3. Read transcript
- * 4. Extract learnings via LLM (with existing memories for dedup)
- * 5. Write KNOWN skills to .specweave/skill-memories/ (primary storage)
- * 6. Write UNKNOWN skills to CLAUDE.md (overflow only)
- */
-export async function handleReflectStop(transcriptPath, projectRoot) {
-    const startTime = performance.now();
-    // Initialize result
-    const result = {
-        ran: false,
-        inputSummary: { transcriptLines: 0 },
-        extracted: { skillLearnings: [] },
-        written: { learningsAdded: 0, learningsSkippedDuplicate: 0 },
-        durationMs: 0,
-    };
-    // Read config
-    const config = readReflectConfig(projectRoot);
-    if (!config.enabled) {
-        result.reason = 'Reflection disabled in config';
-        result.durationMs = performance.now() - startTime;
-        return result;
-    }
-    // Read transcript
-    if (!fs.existsSync(transcriptPath)) {
-        result.reason = `Transcript not found: ${transcriptPath}`;
-        result.durationMs = performance.now() - startTime;
-        return result;
-    }
-    const transcript = fs.readFileSync(transcriptPath, 'utf-8');
-    result.inputSummary.transcriptLines = transcript.split('\n').length;
-    if (transcript.trim().length < 100) {
-        result.reason = 'Transcript too short (< 100 chars)';
-        result.durationMs = performance.now() - startTime;
-        return result;
-    }
-    // Read existing memories for LLM context (semantic deduplication)
-    const existingMemories = readAllSkillMemories(projectRoot);
-    const existingMemoriesStr = formatMemoriesForPrompt(existingMemories);
-    logger.debug(`[reflect] Found ${existingMemories.length} existing memories for dedup context`);
-    // Extract learnings via LLM (with existing memories for semantic dedup)
-    result.model = config.model;
-    result.ran = true;
-    const extraction = await extractLearningsViaLLM(transcript, config.model, existingMemoriesStr);
-    if (!extraction.success) {
-        result.reason = `LLM extraction failed: ${extraction.error}`;
-        result.durationMs = performance.now() - startTime;
-        return result;
-    }
-    result.extracted.skillLearnings = extraction.learnings;
-    // Limit to max per session
-    const learningsToWrite = extraction.learnings.slice(0, config.maxLearningsPerSession);
-    if (learningsToWrite.length === 0) {
-        result.reason = 'No SpecWeave-specific learnings found';
-        result.durationMs = performance.now() - startTime;
-        return result;
-    }
-    // Separate learnings into known skills (skill-memories) vs unknown (CLAUDE.md overflow)
-    const knownSkillLearnings = learningsToWrite.filter((l) => KNOWN_SKILLS.includes(l.skill.toLowerCase()));
-    const unknownSkillLearnings = learningsToWrite.filter((l) => !KNOWN_SKILLS.includes(l.skill.toLowerCase()));
-    // Write KNOWN skills to skill-memories (primary storage)
-    try {
-        const skillMemoryResult = writeSkillMemories(projectRoot, knownSkillLearnings);
-        result.written.learningsAdded += skillMemoryResult.written;
-        result.written.learningsSkippedDuplicate += skillMemoryResult.skipped;
-        logger.debug(`[reflect] Wrote ${skillMemoryResult.written} skill memory files, skipped ${skillMemoryResult.skipped} duplicates`);
-    }
-    catch (err) {
-        // Non-fatal - continue even if skill memory write fails
-        logger.warn(`[reflect] Failed to write skill memories: ${err}`);
-    }
-    // Write UNKNOWN skills to CLAUDE.md (overflow storage only)
-    // This is for skills that don't have dedicated memory files
-    if (unknownSkillLearnings.length > 0) {
-        const claudeMdPath = findClaudeMd(projectRoot);
-        if (claudeMdPath) {
-            try {
-                const writeResult = updateClaudeMd(claudeMdPath, unknownSkillLearnings);
-                result.written.learningsAdded += writeResult.added;
-                result.written.learningsSkippedDuplicate += writeResult.skipped;
-                result.written.claudeMdPath = claudeMdPath;
-                logger.debug(`[reflect] Wrote ${writeResult.added} unknown skill learnings to CLAUDE.md overflow`);
-            }
-            catch (err) {
-                logger.warn(`[reflect] Failed to write to CLAUDE.md: ${err}`);
-            }
-        }
-    }
-    result.durationMs = performance.now() - startTime;
-    return result;
-}
-/**
- * Format reflect result for console output
- */
-export function formatReflectResult(result) {
-    const lines = [];
-    lines.push('════════════════════════════════════════════════════════════════');
-    lines.push('🔍 REFLECT STOP HOOK');
-    lines.push('════════════════════════════════════════════════════════════════');
-    if (!result.ran) {
-        lines.push(`⏭️  Skipped: ${result.reason}`);
-        lines.push(`⏱️  Duration: ${result.durationMs.toFixed(0)}ms`);
-        lines.push('════════════════════════════════════════════════════════════════');
-        return lines.join('\n');
-    }
-    lines.push(`📄 Transcript: ${result.inputSummary.transcriptLines} lines`);
-    lines.push(`🤖 Model: ${result.model}`);
-    lines.push('');
-    if (result.extracted.skillLearnings.length > 0) {
-        lines.push('────────────────────────────────────────────────────────────────');
-        lines.push('📝 EXTRACTED LEARNINGS');
-        lines.push('────────────────────────────────────────────────────────────────');
-        for (const learning of result.extracted.skillLearnings) {
-            lines.push(`   [${learning.skill}] ${learning.learning}`);
-        }
-        lines.push('');
-    }
-    lines.push('────────────────────────────────────────────────────────────────');
-    lines.push('💾 PERSISTENCE');
-    lines.push('────────────────────────────────────────────────────────────────');
-    if (result.written.learningsAdded > 0) {
-        lines.push(`✅ Added: ${result.written.learningsAdded} learning(s)`);
-        if (result.written.claudeMdPath) {
-            lines.push(`   → ${result.written.claudeMdPath}`);
-        }
-        else {
-            lines.push(`   → .specweave/skill-memories/`);
-        }
-    }
-    else {
-        lines.push('⏭️  Nothing new to add');
-    }
-    if (result.written.learningsSkippedDuplicate > 0) {
-        lines.push(`⏭️  Skipped: ${result.written.learningsSkippedDuplicate} duplicate(s)`);
-    }
-    lines.push('');
-    lines.push(`⏱️  Duration: ${result.durationMs.toFixed(0)}ms`);
-    lines.push('════════════════════════════════════════════════════════════════');
-    return lines.join('\n');
-}
-/**
- * Extract learnings from old memory file content (supports multiple formats)
- *
- * Supported formats:
- * 1. LRN format: `**Learning**: content`
- * 2. Arrow format: `- → content` or `- -> content`
- * 3. Simple bullet: `- content` (only if it looks like a learning, not a header)
- */
-function extractLearningsFromOldFormat(content) {
-    const learnings = [];
-    // Format 1: **Learning**: content (LRN-XXXXXXXX format)
-    const lrnMatches = content.matchAll(/\*\*Learning\*\*:\s*(.+)/g);
-    for (const match of lrnMatches) {
-        const learning = match[1].trim();
-        if (learning.length > 5) {
-            learnings.push(learning);
-        }
-    }
-    // Format 2: - → content or - -> content (arrow format)
-    const arrowMatches = content.matchAll(/^-\s*(?:→|->)+\s*(.+)$/gm);
-    for (const match of arrowMatches) {
-        const learning = match[1].trim();
-        if (learning.length > 5) {
-            learnings.push(learning);
-        }
-    }
-    // Format 3: Simple bullets (only if content looks like a learning rule)
-    // Skip headers, metadata lines, and very short content
-    const bulletMatches = content.matchAll(/^-\s+(?!→|->)(.+)$/gm);
-    for (const match of bulletMatches) {
-        const learning = match[1].trim();
-        // Only include if it looks like an actual learning (not metadata)
-        if (learning.length > 10 &&
-            !learning.startsWith('#') &&
-            !learning.startsWith('>') &&
-            !learning.toLowerCase().includes('project-specific') &&
-            !learning.toLowerCase().includes('max ') &&
-            !learning.toLowerCase().includes('auto-deduplicated')) {
-            learnings.push(learning);
-        }
-    }
-    return learnings;
-}
-/**
- * Migrate old memory files to CLAUDE.md
- *
- * This migrates content from:
- * - .specweave/memory/*.md
- * - ~/.specweave/memory/*.md
- *
- * Into the CLAUDE.md Skill Memories section.
- *
- * IMPORTANT: Always cleans up the deprecated memory directories,
- * even if no learnings were migrated (best-effort migration).
- */
-export function migrateOldMemoryFiles(projectRoot) {
-    const result = { migrated: 0, deleted: [] };
-    // Directories to check (always clean these up)
-    const memoryDirs = [
-        path.join(projectRoot, '.specweave', 'memory'),
-    ];
-    // Add global memory if it exists
-    const homeDir = process.env.HOME || process.env.USERPROFILE || '';
-    if (homeDir) {
-        memoryDirs.push(path.join(homeDir, '.specweave', 'memory'));
-    }
-    // Find CLAUDE.md (optional - we still clean up even if not found)
-    const claudeMdPath = findClaudeMd(projectRoot);
-    const learningsToMigrate = [];
-    for (const memoryDir of memoryDirs) {
-        if (!fs.existsSync(memoryDir))
-            continue;
-        const files = fs.readdirSync(memoryDir).filter((f) => f.endsWith('.md'));
-        for (const file of files) {
-            const filePath = path.join(memoryDir, file);
-            const skillName = path.basename(file, '.md');
-            try {
-                const content = fs.readFileSync(filePath, 'utf-8');
-                // Extract learnings using multi-format parser
-                const extractedLearnings = extractLearningsFromOldFormat(content);
-                for (const learning of extractedLearnings) {
-                    learningsToMigrate.push({
-                        skill: skillName,
-                        learning,
-                    });
-                }
-                // Always mark for deletion (deprecated directory)
-                result.deleted.push(filePath);
-            }
-            catch {
-                // Still mark for deletion even if read fails
-                result.deleted.push(filePath);
-            }
-        }
-    }
-    // Migrate learnings to CLAUDE.md if available
-    if (learningsToMigrate.length > 0 && claudeMdPath) {
-        const writeResult = updateClaudeMd(claudeMdPath, learningsToMigrate);
-        result.migrated = writeResult.added;
-    }
-    // Delete old files (always, regardless of migration success)
-    for (const filePath of result.deleted) {
-        try {
-            fs.unlinkSync(filePath);
-            logger.debug(`[reflect] Deleted old memory file: ${filePath}`);
-        }
-        catch {
-            // Ignore delete errors
-        }
-    }
-    // Remove directories (even if not empty - force cleanup of deprecated location)
-    for (const memoryDir of memoryDirs) {
-        try {
-            if (fs.existsSync(memoryDir)) {
-                // Remove any remaining files first
-                const remaining = fs.readdirSync(memoryDir);
-                for (const file of remaining) {
-                    try {
-                        fs.unlinkSync(path.join(memoryDir, file));
-                    }
-                    catch {
-                        // Ignore individual file delete errors
-                    }
-                }
-                // Then remove the directory
-                fs.rmdirSync(memoryDir);
-                logger.debug(`[reflect] Removed deprecated memory directory: ${memoryDir}`);
-            }
-        }
-        catch {
-            // Ignore directory removal errors
-        }
-    }
-    return result;
-}
-/**
- * Force cleanup of deprecated .specweave/memory/ directory
- *
- * This is a simpler, more aggressive cleanup that doesn't attempt migration.
- * Use this when you just want to ensure the deprecated directory is gone.
- *
- * @param projectRoot - Project root path
- * @returns Number of files/directories removed
- */
-export function cleanupDeprecatedMemoryDirectory(projectRoot) {
-    let cleaned = 0;
-    const memoryDirs = [
-        path.join(projectRoot, '.specweave', 'memory'),
-    ];
-    // Add global memory if it exists
-    const homeDir = process.env.HOME || process.env.USERPROFILE || '';
-    if (homeDir) {
-        memoryDirs.push(path.join(homeDir, '.specweave', 'memory'));
-    }
-    for (const memoryDir of memoryDirs) {
-        if (!fs.existsSync(memoryDir))
-            continue;
-        try {
-            // Remove all files in the directory
-            const files = fs.readdirSync(memoryDir);
-            for (const file of files) {
-                try {
-                    fs.unlinkSync(path.join(memoryDir, file));
-                    cleaned++;
-                }
-                catch {
-                    // Ignore individual file errors
-                }
-            }
-            // Remove the directory itself
-            fs.rmdirSync(memoryDir);
-            cleaned++;
-            logger.debug(`[reflect] Removed deprecated memory directory: ${memoryDir}`);
-        }
-        catch {
-            // Ignore directory errors
-        }
-    }
-    return cleaned;
-}
-//# sourceMappingURL=reflect-handler.js.map