@tekmidian/pai 0.9.15 → 0.10.0

package/dist/{indexer-backend-BrrCnasS.mjs.map → indexer-backend-DcYThufd.mjs.map}

@@ -1 +1 @@
-{"version":3,"file":"indexer-backend-BrrCnasS.mjs","names":[],"sources":["../src/memory/indexer/async.ts"],"sourcesContent":["/**\n * Backend-aware async indexer for PAI federation memory.\n *\n * Provides the same functionality as sync.ts but writes through the\n * StorageBackend interface instead of directly to better-sqlite3.\n * Used when the daemon is configured with the Postgres backend.\n *\n * The SQLite path still uses sync.ts directly (which is faster for SQLite\n * due to synchronous transactions).\n */\n\nimport { readFileSync, statSync, existsSync } from \"node:fs\";\nimport { join, relative, basename } from \"node:path\";\nimport type { Database } from \"better-sqlite3\";\nimport type { StorageBackend, ChunkRow } from \"../../storage/interface.js\";\nimport { chunkMarkdown } from \"../chunker.js\";\nimport {\n  sha256File,\n  chunkId,\n  detectTier,\n  walkMdFiles,\n  walkContentFiles,\n  isPathTooBroadForContentScan,\n  parseSessionTitleChunk,\n  yieldToEventLoop,\n  INDEX_YIELD_EVERY,\n} from \"./helpers.js\";\nimport type { IndexResult } from \"./types.js\";\n\nexport type { IndexResult };\n\n// ---------------------------------------------------------------------------\n// Single-file indexing via StorageBackend\n// ---------------------------------------------------------------------------\n\n/**\n * Index a single file through the StorageBackend interface.\n * Returns true if the file was re-indexed (changed or new), false if skipped.\n */\nexport async function indexFileWithBackend(\n  backend: StorageBackend,\n  projectId: number,\n  rootPath: string,\n  relativePath: string,\n  source: string,\n  tier: string,\n): Promise<boolean> {\n  const absPath = join(rootPath, relativePath);\n\n  let content: string;\n  let stat: ReturnType<typeof statSync>;\n  try {\n    content = readFileSync(absPath, \"utf8\");\n    stat = statSync(absPath);\n  } catch {\n    return false;\n  }\n\n  const hash = sha256File(content);\n  const mtime = Math.floor(stat.mtimeMs);\n  const size = stat.size;\n\n  // Change detection\n  const existingHash = await backend.getFileHash(projectId, relativePath);\n  if (existingHash === hash) return false;\n\n  // Delete old chunks\n  await backend.deleteChunksForFile(projectId, relativePath);\n\n  // Chunk the content\n  const rawChunks = chunkMarkdown(content);\n  const updatedAt = Date.now();\n\n  const chunks: ChunkRow[] = rawChunks.map((c, i) => ({\n    id: chunkId(projectId, relativePath, i, c.startLine, c.endLine),\n    projectId,\n    source,\n    tier,\n    path: relativePath,\n    startLine: c.startLine,\n    endLine: c.endLine,\n    hash: c.hash,\n    text: c.text,\n    updatedAt,\n    embedding: null,\n  }));\n\n  // Insert chunks + update file record\n  await backend.insertChunks(chunks);\n  await backend.upsertFile({ projectId, path: relativePath, source, tier, hash, mtime, size });\n\n  return true;\n}\n\n// ---------------------------------------------------------------------------\n// Project-level indexing via StorageBackend\n// ---------------------------------------------------------------------------\n\nexport async function indexProjectWithBackend(\n  backend: StorageBackend,\n  projectId: number,\n  rootPath: string,\n  claudeNotesDir?: string | null,\n): Promise<IndexResult> {\n  const result: IndexResult = { filesProcessed: 0, chunksCreated: 0, filesSkipped: 0 };\n\n  const filesToIndex: Array<{ absPath: string; rootBase: string; source: string; tier: string }> = [];\n\n  const rootMemoryMd = join(rootPath, \"MEMORY.md\");\n  if (existsSync(rootMemoryMd)) {\n    filesToIndex.push({ absPath: rootMemoryMd, rootBase: rootPath, source: \"memory\", tier: \"evergreen\" });\n  }\n\n  const memoryDir = join(rootPath, \"memory\");\n  for (const absPath of walkMdFiles(memoryDir)) {\n    const relPath = relative(rootPath, absPath);\n    const tier = detectTier(relPath);\n    filesToIndex.push({ absPath, rootBase: rootPath, source: \"memory\", tier });\n  }\n\n  const notesDir = join(rootPath, \"Notes\");\n  for (const absPath of walkMdFiles(notesDir)) {\n    filesToIndex.push({ absPath, rootBase: rootPath, source: \"notes\", tier: \"session\" });\n  }\n\n  // Synthetic session-title chunks for Notes files\n  {\n    const updatedAt = Date.now();\n    for (const absPath of walkMdFiles(notesDir)) {\n      const fileName = basename(absPath);\n      const text = parseSessionTitleChunk(fileName);\n      if (!text) continue;\n      const relPath = relative(rootPath, absPath);\n      const syntheticPath = `${relPath}::title`;\n      const id = chunkId(projectId, syntheticPath, 0, 0, 0);\n      const hash = sha256File(text);\n      const titleChunk: ChunkRow = {\n        id, projectId, source: \"notes\", tier: \"session\",\n        path: syntheticPath, startLine: 0, endLine: 0,\n        hash, text, updatedAt, embedding: null,\n      };\n      try {\n        await backend.insertChunks([titleChunk]);\n      } catch {\n        // Skip title chunks that cause backend errors\n      }\n    }\n  }\n\n  if (!isPathTooBroadForContentScan(rootPath)) {\n    for (const absPath of walkContentFiles(rootPath)) {\n      filesToIndex.push({ absPath, rootBase: rootPath, source: \"content\", tier: \"topic\" });\n    }\n  }\n\n  if (claudeNotesDir && claudeNotesDir !== notesDir) {\n    for (const absPath of walkMdFiles(claudeNotesDir)) {\n      filesToIndex.push({ absPath, rootBase: claudeNotesDir, source: \"notes\", tier: \"session\" });\n    }\n\n    // Synthetic title chunks for claude notes dir\n    {\n      const updatedAt = Date.now();\n      for (const absPath of walkMdFiles(claudeNotesDir)) {\n        const fileName = basename(absPath);\n        const text = parseSessionTitleChunk(fileName);\n        if (!text) continue;\n        const relPath = relative(claudeNotesDir, absPath);\n        const syntheticPath = `${relPath}::title`;\n        const id = chunkId(projectId, syntheticPath, 0, 0, 0);\n        const hash = sha256File(text);\n        const titleChunk: ChunkRow = {\n          id, projectId, source: \"notes\", tier: \"session\",\n          path: syntheticPath, startLine: 0, endLine: 0,\n          hash, text, updatedAt, embedding: null,\n        };\n        try {\n          await backend.insertChunks([titleChunk]);\n        } catch {\n          // Skip title chunks that cause backend errors\n        }\n      }\n    }\n\n    if (claudeNotesDir.endsWith(\"/Notes\")) {\n      const claudeProjectDir = claudeNotesDir.slice(0, -\"/Notes\".length);\n      const claudeMemoryMd = join(claudeProjectDir, \"MEMORY.md\");\n      if (existsSync(claudeMemoryMd)) {\n        filesToIndex.push({ absPath: claudeMemoryMd, rootBase: claudeProjectDir, source: \"memory\", tier: \"evergreen\" });\n      }\n      const claudeMemoryDir = join(claudeProjectDir, \"memory\");\n      for (const absPath of walkMdFiles(claudeMemoryDir)) {\n        const relPath = relative(claudeProjectDir, absPath);\n        const tier = detectTier(relPath);\n        filesToIndex.push({ absPath, rootBase: claudeProjectDir, source: \"memory\", tier });\n      }\n    }\n  }\n\n  await yieldToEventLoop();\n\n  let filesSinceYield = 0;\n\n  for (const { absPath, rootBase, source, tier } of filesToIndex) {\n    if (filesSinceYield >= INDEX_YIELD_EVERY) {\n      await yieldToEventLoop();\n      filesSinceYield = 0;\n    }\n    filesSinceYield++;\n\n    const relPath = relative(rootBase, absPath);\n    try {\n      const changed = await indexFileWithBackend(backend, projectId, rootBase, relPath, source, tier);\n\n      if (changed) {\n        const ids = await backend.getChunkIds(projectId, relPath);\n        result.filesProcessed++;\n        result.chunksCreated += ids.length;\n      } else {\n        result.filesSkipped++;\n      }\n    } catch {\n      // Skip files that cause backend errors (e.g. null bytes in Postgres)\n      result.filesSkipped++;\n    }\n  }\n\n  // Prune stale paths\n  const livePaths = new Set<string>();\n  for (const { absPath, rootBase } of filesToIndex) {\n    livePaths.add(relative(rootBase, absPath));\n  }\n\n  const dbChunkPaths = await backend.getDistinctChunkPaths(projectId);\n\n  const stalePaths: string[] = [];\n  for (const p of dbChunkPaths) {\n    const basePath = p.endsWith(\"::title\") ? p.slice(0, -\"::title\".length) : p;\n    if (!livePaths.has(basePath)) {\n      stalePaths.push(p);\n    }\n  }\n\n  if (stalePaths.length > 0) {\n    await backend.deletePaths(projectId, stalePaths);\n  }\n\n  return result;\n}\n\n// ---------------------------------------------------------------------------\n// Embedding generation via StorageBackend\n// ---------------------------------------------------------------------------\n\nconst EMBED_BATCH_SIZE = 50;\nconst EMBED_YIELD_EVERY = 1;\n\n/**\n * Generate and store embeddings for all unembedded chunks via the StorageBackend.\n *\n * Processes chunks in batches of EMBED_BATCH_SIZE, yielding to the event loop\n * every EMBED_YIELD_EVERY chunks to avoid blocking IPC calls from MCP shims.\n *\n * The optional `shouldStop` callback is checked between every batch. When it\n * returns true the embed loop exits early so the caller (e.g. the daemon\n * shutdown handler) can close the pool without racing against active queries.\n *\n * Returns the number of newly embedded chunks.\n */\nexport async function embedChunksWithBackend(\n  backend: StorageBackend,\n  shouldStop?: () => boolean,\n  projectNames?: Map<number, string>,\n): Promise<number> {\n  const { generateEmbedding, serializeEmbedding } = await import(\"../embeddings.js\");\n\n  const rows = await backend.getUnembeddedChunkIds();\n  if (rows.length === 0) return 0;\n\n  const total = rows.length;\n  let embedded = 0;\n\n  // Build a summary of what needs embedding: count chunks per project_id\n  const projectChunkCounts = new Map<number, { count: number; samplePath: string }>();\n  for (const row of rows) {\n    const entry = projectChunkCounts.get(row.project_id);\n    if (entry) {\n      entry.count++;\n    } else {\n      projectChunkCounts.set(row.project_id, { count: 1, samplePath: row.path });\n    }\n  }\n  const pName = (pid: number) => projectNames?.get(pid) ?? `project ${pid}`;\n  const projectSummary = Array.from(projectChunkCounts.entries())\n    .map(([pid, { count, samplePath }]) => `  ${pName(pid)}: ${count} chunks (e.g. ${samplePath})`)\n    .join(\"\\n\");\n  process.stderr.write(\n    `[pai-daemon] Embed pass: ${total} unembedded chunks across ${projectChunkCounts.size} project(s)\\n${projectSummary}\\n`\n  );\n\n  // Track current project for transition logging\n  let currentProjectId = -1;\n  let projectEmbedded = 0;\n\n  for (let i = 0; i < rows.length; i += EMBED_BATCH_SIZE) {\n    // Check cancellation between every batch before touching the pool again\n    if (shouldStop?.()) {\n      process.stderr.write(\n        `[pai-daemon] Embed pass cancelled after ${embedded}/${total} chunks (shutdown requested)\\n`\n      );\n      break;\n    }\n\n    const batch = rows.slice(i, i + EMBED_BATCH_SIZE);\n\n    for (let j = 0; j < batch.length; j++) {\n      const { id, text, project_id, path } = batch[j];\n\n      // Log when switching to a new project\n      if (project_id !== currentProjectId) {\n        if (currentProjectId !== -1) {\n          process.stderr.write(\n            `[pai-daemon] Finished ${pName(currentProjectId)}: ${projectEmbedded} chunks embedded\\n`\n          );\n        }\n        const info = projectChunkCounts.get(project_id);\n        process.stderr.write(\n          `[pai-daemon] Embedding ${pName(project_id)} (${info?.count ?? \"?\"} chunks, starting at ${path})\\n`\n        );\n        currentProjectId = project_id;\n        projectEmbedded = 0;\n      }\n\n      // Yield to the event loop periodically to keep IPC responsive\n      if ((embedded + j) % EMBED_YIELD_EVERY === 0) {\n        await yieldToEventLoop();\n      }\n\n      const vec = await generateEmbedding(text);\n      const blob = serializeEmbedding(vec);\n      await backend.updateEmbedding(id, blob);\n      projectEmbedded++;\n    }\n\n    embedded += batch.length;\n\n    // Log progress with current file path for context\n    const lastChunk = batch[batch.length - 1];\n    process.stderr.write(\n      `[pai-daemon] Embedded ${embedded}/${total} chunks (${pName(lastChunk.project_id)}: ${lastChunk.path})\\n`\n    );\n  }\n\n  // Log final project completion\n  if (currentProjectId !== -1) {\n    process.stderr.write(\n      `[pai-daemon] Finished ${pName(currentProjectId)}: ${projectEmbedded} chunks embedded\\n`\n    );\n  }\n\n  return embedded;\n}\n\n// ---------------------------------------------------------------------------\n// Global indexing via StorageBackend\n// ---------------------------------------------------------------------------\n\nexport async function indexAllWithBackend(\n  backend: StorageBackend,\n  registryDb: Database,\n): Promise<{ projects: number; result: IndexResult }> {\n  const projects = registryDb\n    .prepare(\"SELECT id, root_path, claude_notes_dir FROM projects WHERE status = 'active'\")\n    .all() as Array<{ id: number; root_path: string; claude_notes_dir: string | null }>;\n\n  const totals: IndexResult = { filesProcessed: 0, chunksCreated: 0, filesSkipped: 0 };\n\n  for (const project of projects) {\n    await yieldToEventLoop();\n    const r = await indexProjectWithBackend(backend, project.id, project.root_path, project.claude_notes_dir);\n    totals.filesProcessed += r.filesProcessed;\n    totals.chunksCreated += r.chunksCreated;\n    totals.filesSkipped += r.filesSkipped;\n  }\n\n  return { projects: projects.length, result: totals };\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;AAuCA,eAAsB,qBACpB,SACA,WACA,UACA,cACA,QACA,MACkB;CAClB,MAAM,UAAU,KAAK,UAAU,aAAa;CAE5C,IAAI;CACJ,IAAI;AACJ,KAAI;AACF,YAAU,aAAa,SAAS,OAAO;AACvC,SAAO,SAAS,QAAQ;SAClB;AACN,SAAO;;CAGT,MAAM,OAAO,WAAW,QAAQ;CAChC,MAAM,QAAQ,KAAK,MAAM,KAAK,QAAQ;CACtC,MAAM,OAAO,KAAK;AAIlB,KADqB,MAAM,QAAQ,YAAY,WAAW,aAAa,KAClD,KAAM,QAAO;AAGlC,OAAM,QAAQ,oBAAoB,WAAW,aAAa;CAG1D,MAAM,YAAY,cAAc,QAAQ;CACxC,MAAM,YAAY,KAAK,KAAK;CAE5B,MAAM,SAAqB,UAAU,KAAK,GAAG,OAAO;EAClD,IAAI,QAAQ,WAAW,cAAc,GAAG,EAAE,WAAW,EAAE,QAAQ;EAC/D;EACA;EACA;EACA,MAAM;EACN,WAAW,EAAE;EACb,SAAS,EAAE;EACX,MAAM,EAAE;EACR,MAAM,EAAE;EACR;EACA,WAAW;EACZ,EAAE;AAGH,OAAM,QAAQ,aAAa,OAAO;AAClC,OAAM,QAAQ,WAAW;EAAE;EAAW,MAAM;EAAc;EAAQ;EAAM;EAAM;EAAO;EAAM,CAAC;AAE5F,QAAO;;AAOT,eAAsB,wBACpB,SACA,WACA,UACA,gBACsB;CACtB,MAAM,SAAsB;EAAE,gBAAgB;EAAG,eAAe;EAAG,cAAc;EAAG;CAEpF,MAAM,eAA2F,EAAE;CAEnG,MAAM,eAAe,KAAK,UAAU,YAAY;AAChD,KAAI,WAAW,aAAa,CAC1B,cAAa,KAAK;EAAE,SAAS;EAAc,UAAU;EAAU,QAAQ;EAAU,MAAM;EAAa,CAAC;CAGvG,MAAM,YAAY,KAAK,UAAU,SAAS;AAC1C,MAAK,MAAM,WAAW,YAAY,UAAU,EAAE;EAE5C,MAAM,OAAO,WADG,SAAS,UAAU,QAAQ,CACX;AAChC,eAAa,KAAK;GAAE;GAAS,UAAU;GAAU,QAAQ;GAAU;GAAM,CAAC;;CAG5E,MAAM,WAAW,KAAK,UAAU,QAAQ;AACxC,MAAK,MAAM,WAAW,YAAY,SAAS,CACzC,cAAa,KAAK;EAAE;EAAS,UAAU;EAAU,QAAQ;EAAS,MAAM;EAAW,CAAC;CAItF;EACE,MAAM,YAAY,KAAK,KAAK;AAC5B,OAAK,MAAM,WAAW,YAAY,SAAS,EAAE;GAE3C,MAAM,OAAO,uBADI,SAAS,QAAQ,CACW;AAC7C,OAAI,CAAC,KAAM;GAEX,MAAM,gBAAgB,GADN,SAAS,UAAU,QAAQ,CACV;GAGjC,MAAM,aAAuB;IAC3B,IAHS,QAAQ,WAAW,eAAe,GAAG,GAAG,EAAE;IAG/C;IAAW,QAAQ;IAAS,MAAM;IACtC,MAAM;IAAe,WAAW;IAAG,SAAS;IAC5C,MAJW,WAAW,KAAK;IAIrB;IAAM;IAAW,WAAW;IACnC;AACD,OAAI;AACF,UAAM,QAAQ,aAAa,CAAC,WAAW,CAAC;WAClC;;;AAMZ,KAAI,CAAC,6BAA6B,SAAS,CACzC,MAAK,MAAM,WAAW,iBAAiB,SAAS,CAC9C,cAAa,KAAK;EAAE;EAAS,UAAU;EAAU,QAAQ;EAAW,MAAM;EAAS,CAAC;AAIxF,KAAI,kBAAkB,mBAAmB,UAAU;AACjD,OAAK,MAAM,WAAW,YAAY,eAAe,CAC/C,cAAa,KAAK;GAAE;GAAS,UAAU;GAAgB,QAAQ;GAAS,MAAM;GAAW,CAAC;EAI5F;GACE,MAAM,YAAY,KAAK,KAAK;AAC5B,QAAK,MAAM,WAAW,YAAY,eAAe,EAAE;IAEjD,MAAM,OAAO,uBADI,SAAS,QAAQ,CACW;AAC7C,QAAI,CAAC,KAAM;IAEX,MAAM,gBAAgB,GADN,SAAS,gBAAgB,QAAQ,CAChB;IAGjC,MAAM,aAAuB;KAC3B,IAHS,QAAQ,WAAW,eAAe,GAAG,GAAG,EAAE;KAG/C;KAAW,QAAQ;KAAS,MAAM;KACtC,MAAM;KAAe,WAAW;KAAG,SAAS;KAC5C,MAJW,WAAW,KAAK;KAIrB;KAAM;KAAW,WAAW;KACnC;AACD,QAAI;AACF,WAAM,QAAQ,aAAa,CAAC,WAAW,CAAC;YAClC;;;AAMZ,MAAI,eAAe,SAAS,SAAS,EAAE;GACrC,MAAM,mBAAmB,eAAe,MAAM,GAAG,GAAiB;GAClE,MAAM,iBAAiB,KAAK,kBAAkB,YAAY;AAC1D,OAAI,WAAW,eAAe,CAC5B,cAAa,KAAK;IAAE,SAAS;IAAgB,UAAU;IAAkB,QAAQ;IAAU,MAAM;IAAa,CAAC;GAEjH,MAAM,kBAAkB,KAAK,kBAAkB,SAAS;AACxD,QAAK,MAAM,WAAW,YAAY,gBAAgB,EAAE;IAElD,MAAM,OAAO,WADG,SAAS,kBAAkB,QAAQ,CACnB;AAChC,iBAAa,KAAK;KAAE;KAAS,UAAU;KAAkB,QAAQ;KAAU;KAAM,CAAC;;;;AAKxF,OAAM,kBAAkB;CAExB,IAAI,kBAAkB;AAEtB,MAAK,MAAM,EAAE,SAAS,UAAU,QAAQ,UAAU,cAAc;AAC9D,MAAI,mBAAmB,mBAAmB;AACxC,SAAM,kBAAkB;AACxB,qBAAkB;;AAEpB;EAEA,MAAM,UAAU,SAAS,UAAU,QAAQ;AAC3C,MAAI;AAGF,OAFgB,MAAM,qBAAqB,SAAS,WAAW,UAAU,SAAS,QAAQ,KAAK,EAElF;IACX,MAAM,MAAM,MAAM,QAAQ,YAAY,WAAW,QAAQ;AACzD,WAAO;AACP,WAAO,iBAAiB,IAAI;SAE5B,QAAO;UAEH;AAEN,UAAO;;;CAKX,MAAM,4BAAY,IAAI,KAAa;AACnC,MAAK,MAAM,EAAE,SAAS,cAAc,aAClC,WAAU,IAAI,SAAS,UAAU,QAAQ,CAAC;CAG5C,MAAM,eAAe,MAAM,QAAQ,sBAAsB,UAAU;CAEnE,MAAM,aAAuB,EAAE;AAC/B,MAAK,MAAM,KAAK,cAAc;EAC5B,MAAM,WAAW,EAAE,SAAS,UAAU,GAAG,EAAE,MAAM,GAAG,GAAkB,GAAG;AACzE,MAAI,CAAC,UAAU,IAAI,SAAS,CAC1B,YAAW,KAAK,EAAE;;AAItB,KAAI,WAAW,SAAS,EACtB,OAAM,QAAQ,YAAY,WAAW,WAAW;AAGlD,QAAO;;AAOT,MAAM,mBAAmB;AACzB,MAAM,oBAAoB;;;;;;;;;;;;;AAc1B,eAAsB,uBACpB,SACA,YACA,cACiB;CACjB,MAAM,EAAE,mBAAmB,uBAAuB,MAAM,OAAO;CAE/D,MAAM,OAAO,MAAM,QAAQ,uBAAuB;AAClD,KAAI,KAAK,WAAW,EAAG,QAAO;CAE9B,MAAM,QAAQ,KAAK;CACnB,IAAI,WAAW;CAGf,MAAM,qCAAqB,IAAI,KAAoD;AACnF,MAAK,MAAM,OAAO,MAAM;EACtB,MAAM,QAAQ,mBAAmB,IAAI,IAAI,WAAW;AACpD,MAAI,MACF,OAAM;MAEN,oBAAmB,IAAI,IAAI,YAAY;GAAE,OAAO;GAAG,YAAY,IAAI;GAAM,CAAC;;CAG9E,MAAM,SAAS,QAAgB,cAAc,IAAI,IAAI,IAAI,WAAW;CACpE,MAAM,iBAAiB,MAAM,KAAK,mBAAmB,SAAS,CAAC,CAC5D,KAAK,CAAC,KAAK,EAAE,OAAO,kBAAkB,KAAK,MAAM,IAAI,CAAC,IAAI,MAAM,gBAAgB,WAAW,GAAG,CAC9F,KAAK,KAAK;AACb,SAAQ,OAAO,MACb,4BAA4B,MAAM,4BAA4B,mBAAmB,KAAK,eAAe,eAAe,IACrH;CAGD,IAAI,mBAAmB;CACvB,IAAI,kBAAkB;AAEtB,MAAK,IAAI,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK,kBAAkB;AAEtD,MAAI,cAAc,EAAE;AAClB,WAAQ,OAAO,MACb,2CAA2C,SAAS,GAAG,MAAM,gCAC9D;AACD;;EAGF,MAAM,QAAQ,KAAK,MAAM,GAAG,IAAI,iBAAiB;AAEjD,OAAK,IAAI,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK;GACrC,MAAM,EAAE,IAAI,MAAM,YAAY,SAAS,MAAM;AAG7C,OAAI,eAAe,kBAAkB;AACnC,QAAI,qBAAqB,GACvB,SAAQ,OAAO,MACb,yBAAyB,MAAM,iBAAiB,CAAC,IAAI,gBAAgB,oBACtE;IAEH,MAAM,OAAO,mBAAmB,IAAI,WAAW;AAC/C,YAAQ,OAAO,MACb,0BAA0B,MAAM,WAAW,CAAC,IAAI,MAAM,SAAS,IAAI,uBAAuB,KAAK,KAChG;AACD,uBAAmB;AACnB,sBAAkB;;AAIpB,QAAK,WAAW,KAAK,sBAAsB,EACzC,OAAM,kBAAkB;GAI1B,MAAM,OAAO,mBADD,MAAM,kBAAkB,KAAK,CACL;AACpC,SAAM,QAAQ,gBAAgB,IAAI,KAAK;AACvC;;AAGF,cAAY,MAAM;EAGlB,MAAM,YAAY,MAAM,MAAM,SAAS;AACvC,UAAQ,OAAO,MACb,yBAAyB,SAAS,GAAG,MAAM,WAAW,MAAM,UAAU,WAAW,CAAC,IAAI,UAAU,KAAK,KACtG;;AAIH,KAAI,qBAAqB,GACvB,SAAQ,OAAO,MACb,yBAAyB,MAAM,iBAAiB,CAAC,IAAI,gBAAgB,oBACtE;AAGH,QAAO;;AAOT,eAAsB,oBACpB,SACA,YACoD;CACpD,MAAM,WAAW,WACd,QAAQ,+EAA+E,CACvF,KAAK;CAER,MAAM,SAAsB;EAAE,gBAAgB;EAAG,eAAe;EAAG,cAAc;EAAG;AAEpF,MAAK,MAAM,WAAW,UAAU;AAC9B,QAAM,kBAAkB;EACxB,MAAM,IAAI,MAAM,wBAAwB,SAAS,QAAQ,IAAI,QAAQ,WAAW,QAAQ,iBAAiB;AACzG,SAAO,kBAAkB,EAAE;AAC3B,SAAO,iBAAiB,EAAE;AAC1B,SAAO,gBAAgB,EAAE;;AAG3B,QAAO;EAAE,UAAU,SAAS;EAAQ,QAAQ;EAAQ"}
+{"version":3,"file":"indexer-backend-DcYThufd.mjs","names":[],"sources":["../src/memory/indexer/async.ts"],"sourcesContent":["/**\n * Backend-aware async indexer for PAI federation memory.\n *\n * Provides the same functionality as sync.ts but writes through the\n * StorageBackend interface instead of directly to better-sqlite3.\n * Used when the daemon is configured with the Postgres backend.\n *\n * The SQLite path still uses sync.ts directly (which is faster for SQLite\n * due to synchronous transactions).\n */\n\nimport { readFileSync, statSync, existsSync } from \"node:fs\";\nimport { join, relative, basename } from \"node:path\";\nimport type { Database } from \"better-sqlite3\";\nimport type { StorageBackend, ChunkRow } from \"../../storage/interface.js\";\nimport { chunkMarkdown } from \"../chunker.js\";\nimport {\n  sha256File,\n  chunkId,\n  detectTier,\n  walkMdFiles,\n  walkContentFiles,\n  isPathTooBroadForContentScan,\n  parseSessionTitleChunk,\n  yieldToEventLoop,\n  INDEX_YIELD_EVERY,\n} from \"./helpers.js\";\nimport type { IndexResult } from \"./types.js\";\n\nexport type { IndexResult };\n\n// ---------------------------------------------------------------------------\n// Single-file indexing via StorageBackend\n// ---------------------------------------------------------------------------\n\n/**\n * Index a single file through the StorageBackend interface.\n * Returns true if the file was re-indexed (changed or new), false if skipped.\n */\nexport async function indexFileWithBackend(\n  backend: StorageBackend,\n  projectId: number,\n  rootPath: string,\n  relativePath: string,\n  source: string,\n  tier: string,\n): Promise<boolean> {\n  const absPath = join(rootPath, relativePath);\n\n  let content: string;\n  let stat: ReturnType<typeof statSync>;\n  try {\n    content = readFileSync(absPath, \"utf8\");\n    stat = statSync(absPath);\n  } catch {\n    return false;\n  }\n\n  const hash = sha256File(content);\n  const mtime = Math.floor(stat.mtimeMs);\n  const size = stat.size;\n\n  // Change detection\n  const existingHash = await backend.getFileHash(projectId, relativePath);\n  if (existingHash === hash) return false;\n\n  // Delete old chunks\n  await backend.deleteChunksForFile(projectId, relativePath);\n\n  // Chunk the content\n  const rawChunks = chunkMarkdown(content);\n  const updatedAt = Date.now();\n\n  const chunks: ChunkRow[] = rawChunks.map((c, i) => ({\n    id: chunkId(projectId, relativePath, i, c.startLine, c.endLine),\n    projectId,\n    source,\n    tier,\n    path: relativePath,\n    startLine: c.startLine,\n    endLine: c.endLine,\n    hash: c.hash,\n    text: c.text,\n    updatedAt,\n    embedding: null,\n  }));\n\n  // Insert chunks + update file record\n  await backend.insertChunks(chunks);\n  await backend.upsertFile({ projectId, path: relativePath, source, tier, hash, mtime, size });\n\n  return true;\n}\n\n// ---------------------------------------------------------------------------\n// Project-level indexing via StorageBackend\n// ---------------------------------------------------------------------------\n\nexport async function indexProjectWithBackend(\n  backend: StorageBackend,\n  projectId: number,\n  rootPath: string,\n  claudeNotesDir?: string | null,\n): Promise<IndexResult> {\n  const result: IndexResult = { filesProcessed: 0, chunksCreated: 0, filesSkipped: 0 };\n\n  const filesToIndex: Array<{ absPath: string; rootBase: string; source: string; tier: string }> = [];\n\n  const rootMemoryMd = join(rootPath, \"MEMORY.md\");\n  if (existsSync(rootMemoryMd)) {\n    filesToIndex.push({ absPath: rootMemoryMd, rootBase: rootPath, source: \"memory\", tier: \"evergreen\" });\n  }\n\n  const memoryDir = join(rootPath, \"memory\");\n  for (const absPath of walkMdFiles(memoryDir)) {\n    const relPath = relative(rootPath, absPath);\n    const tier = detectTier(relPath);\n    filesToIndex.push({ absPath, rootBase: rootPath, source: \"memory\", tier });\n  }\n\n  const notesDir = join(rootPath, \"Notes\");\n  for (const absPath of walkMdFiles(notesDir)) {\n    filesToIndex.push({ absPath, rootBase: rootPath, source: \"notes\", tier: \"session\" });\n  }\n\n  // Synthetic session-title chunks for Notes files\n  {\n    const updatedAt = Date.now();\n    for (const absPath of walkMdFiles(notesDir)) {\n      const fileName = basename(absPath);\n      const text = parseSessionTitleChunk(fileName);\n      if (!text) continue;\n      const relPath = relative(rootPath, absPath);\n      const syntheticPath = `${relPath}::title`;\n      const id = chunkId(projectId, syntheticPath, 0, 0, 0);\n      const hash = sha256File(text);\n      const titleChunk: ChunkRow = {\n        id, projectId, source: \"notes\", tier: \"session\",\n        path: syntheticPath, startLine: 0, endLine: 0,\n        hash, text, updatedAt, embedding: null,\n      };\n      try {\n        await backend.insertChunks([titleChunk]);\n      } catch {\n        // Skip title chunks that cause backend errors\n      }\n    }\n  }\n\n  if (!isPathTooBroadForContentScan(rootPath)) {\n    for (const absPath of walkContentFiles(rootPath)) {\n      filesToIndex.push({ absPath, rootBase: rootPath, source: \"content\", tier: \"topic\" });\n    }\n  }\n\n  if (claudeNotesDir && claudeNotesDir !== notesDir) {\n    for (const absPath of walkMdFiles(claudeNotesDir)) {\n      filesToIndex.push({ absPath, rootBase: claudeNotesDir, source: \"notes\", tier: \"session\" });\n    }\n\n    // Synthetic title chunks for claude notes dir\n    {\n      const updatedAt = Date.now();\n      for (const absPath of walkMdFiles(claudeNotesDir)) {\n        const fileName = basename(absPath);\n        const text = parseSessionTitleChunk(fileName);\n        if (!text) continue;\n        const relPath = relative(claudeNotesDir, absPath);\n        const syntheticPath = `${relPath}::title`;\n        const id = chunkId(projectId, syntheticPath, 0, 0, 0);\n        const hash = sha256File(text);\n        const titleChunk: ChunkRow = {\n          id, projectId, source: \"notes\", tier: \"session\",\n          path: syntheticPath, startLine: 0, endLine: 0,\n          hash, text, updatedAt, embedding: null,\n        };\n        try {\n          await backend.insertChunks([titleChunk]);\n        } catch {\n          // Skip title chunks that cause backend errors\n        }\n      }\n    }\n\n    if (claudeNotesDir.endsWith(\"/Notes\")) {\n      const claudeProjectDir = claudeNotesDir.slice(0, -\"/Notes\".length);\n      const claudeMemoryMd = join(claudeProjectDir, \"MEMORY.md\");\n      if (existsSync(claudeMemoryMd)) {\n        filesToIndex.push({ absPath: claudeMemoryMd, rootBase: claudeProjectDir, source: \"memory\", tier: \"evergreen\" });\n      }\n      const claudeMemoryDir = join(claudeProjectDir, \"memory\");\n      for (const absPath of walkMdFiles(claudeMemoryDir)) {\n        const relPath = relative(claudeProjectDir, absPath);\n        const tier = detectTier(relPath);\n        filesToIndex.push({ absPath, rootBase: claudeProjectDir, source: \"memory\", tier });\n      }\n    }\n  }\n\n  await yieldToEventLoop();\n\n  let filesSinceYield = 0;\n\n  for (const { absPath, rootBase, source, tier } of filesToIndex) {\n    if (filesSinceYield >= INDEX_YIELD_EVERY) {\n      await yieldToEventLoop();\n      filesSinceYield = 0;\n    }\n    filesSinceYield++;\n\n    const relPath = relative(rootBase, absPath);\n    try {\n      const changed = await indexFileWithBackend(backend, projectId, rootBase, relPath, source, tier);\n\n      if (changed) {\n        const ids = await backend.getChunkIds(projectId, relPath);\n        result.filesProcessed++;\n        result.chunksCreated += ids.length;\n      } else {\n        result.filesSkipped++;\n      }\n    } catch {\n      // Skip files that cause backend errors (e.g. null bytes in Postgres)\n      result.filesSkipped++;\n    }\n  }\n\n  // Prune stale paths\n  const livePaths = new Set<string>();\n  for (const { absPath, rootBase } of filesToIndex) {\n    livePaths.add(relative(rootBase, absPath));\n  }\n\n  const dbChunkPaths = await backend.getDistinctChunkPaths(projectId);\n\n  const stalePaths: string[] = [];\n  for (const p of dbChunkPaths) {\n    const basePath = p.endsWith(\"::title\") ? p.slice(0, -\"::title\".length) : p;\n    if (!livePaths.has(basePath)) {\n      stalePaths.push(p);\n    }\n  }\n\n  if (stalePaths.length > 0) {\n    await backend.deletePaths(projectId, stalePaths);\n  }\n\n  return result;\n}\n\n// ---------------------------------------------------------------------------\n// Embedding generation via StorageBackend\n// ---------------------------------------------------------------------------\n\nconst EMBED_BATCH_SIZE = 50;\nconst EMBED_YIELD_EVERY = 1;\n\n/**\n * Generate and store embeddings for all unembedded chunks via the StorageBackend.\n *\n * Processes chunks in batches of EMBED_BATCH_SIZE, yielding to the event loop\n * every EMBED_YIELD_EVERY chunks to avoid blocking IPC calls from MCP shims.\n *\n * The optional `shouldStop` callback is checked between every batch. When it\n * returns true the embed loop exits early so the caller (e.g. the daemon\n * shutdown handler) can close the pool without racing against active queries.\n *\n * Returns the number of newly embedded chunks.\n */\nexport async function embedChunksWithBackend(\n  backend: StorageBackend,\n  shouldStop?: () => boolean,\n  projectNames?: Map<number, string>,\n): Promise<number> {\n  const { generateEmbedding, serializeEmbedding } = await import(\"../embeddings.js\");\n\n  const rows = await backend.getUnembeddedChunkIds();\n  if (rows.length === 0) return 0;\n\n  const total = rows.length;\n  let embedded = 0;\n\n  // Build a summary of what needs embedding: count chunks per project_id\n  const projectChunkCounts = new Map<number, { count: number; samplePath: string }>();\n  for (const row of rows) {\n    const entry = projectChunkCounts.get(row.project_id);\n    if (entry) {\n      entry.count++;\n    } else {\n      projectChunkCounts.set(row.project_id, { count: 1, samplePath: row.path });\n    }\n  }\n  const pName = (pid: number) => projectNames?.get(pid) ?? `project ${pid}`;\n  const projectSummary = Array.from(projectChunkCounts.entries())\n    .map(([pid, { count, samplePath }]) => `  ${pName(pid)}: ${count} chunks (e.g. ${samplePath})`)\n    .join(\"\\n\");\n  process.stderr.write(\n    `[pai-daemon] Embed pass: ${total} unembedded chunks across ${projectChunkCounts.size} project(s)\\n${projectSummary}\\n`\n  );\n\n  // Track current project for transition logging\n  let currentProjectId = -1;\n  let projectEmbedded = 0;\n\n  for (let i = 0; i < rows.length; i += EMBED_BATCH_SIZE) {\n    // Check cancellation between every batch before touching the pool again\n    if (shouldStop?.()) {\n      process.stderr.write(\n        `[pai-daemon] Embed pass cancelled after ${embedded}/${total} chunks (shutdown requested)\\n`\n      );\n      break;\n    }\n\n    const batch = rows.slice(i, i + EMBED_BATCH_SIZE);\n\n    for (let j = 0; j < batch.length; j++) {\n      const { id, text, project_id, path } = batch[j];\n\n      // Log when switching to a new project\n      if (project_id !== currentProjectId) {\n        if (currentProjectId !== -1) {\n          process.stderr.write(\n            `[pai-daemon] Finished ${pName(currentProjectId)}: ${projectEmbedded} chunks embedded\\n`\n          );\n        }\n        const info = projectChunkCounts.get(project_id);\n        process.stderr.write(\n          `[pai-daemon] Embedding ${pName(project_id)} (${info?.count ?? \"?\"} chunks, starting at ${path})\\n`\n        );\n        currentProjectId = project_id;\n        projectEmbedded = 0;\n      }\n\n      // Yield to the event loop periodically to keep IPC responsive\n      if ((embedded + j) % EMBED_YIELD_EVERY === 0) {\n        await yieldToEventLoop();\n      }\n\n      const vec = await generateEmbedding(text);\n      const blob = serializeEmbedding(vec);\n      await backend.updateEmbedding(id, blob);\n      projectEmbedded++;\n    }\n\n    embedded += batch.length;\n\n    // Log progress with current file path for context\n    const lastChunk = batch[batch.length - 1];\n    process.stderr.write(\n      `[pai-daemon] Embedded ${embedded}/${total} chunks (${pName(lastChunk.project_id)}: ${lastChunk.path})\\n`\n    );\n  }\n\n  // Log final project completion\n  if (currentProjectId !== -1) {\n    process.stderr.write(\n      `[pai-daemon] Finished ${pName(currentProjectId)}: ${projectEmbedded} chunks embedded\\n`\n    );\n  }\n\n  return embedded;\n}\n\n// ---------------------------------------------------------------------------\n// Global indexing via StorageBackend\n// ---------------------------------------------------------------------------\n\nexport async function indexAllWithBackend(\n  backend: StorageBackend,\n  registryDb: Database,\n): Promise<{ projects: number; result: IndexResult }> {\n  const projects = registryDb\n    .prepare(\"SELECT id, root_path, claude_notes_dir FROM projects WHERE status = 'active'\")\n    .all() as Array<{ id: number; root_path: string; claude_notes_dir: string | null }>;\n\n  const totals: IndexResult = { filesProcessed: 0, chunksCreated: 0, filesSkipped: 0 };\n\n  for (const project of projects) {\n    await yieldToEventLoop();\n    const r = await indexProjectWithBackend(backend, project.id, project.root_path, project.claude_notes_dir);\n    totals.filesProcessed += r.filesProcessed;\n    totals.chunksCreated += r.chunksCreated;\n    totals.filesSkipped += r.filesSkipped;\n  }\n\n  return { projects: projects.length, result: totals };\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;AAuCA,eAAsB,qBACpB,SACA,WACA,UACA,cACA,QACA,MACkB;CAClB,MAAM,UAAU,KAAK,UAAU,aAAa;CAE5C,IAAI;CACJ,IAAI;AACJ,KAAI;AACF,YAAU,aAAa,SAAS,OAAO;AACvC,SAAO,SAAS,QAAQ;SAClB;AACN,SAAO;;CAGT,MAAM,OAAO,WAAW,QAAQ;CAChC,MAAM,QAAQ,KAAK,MAAM,KAAK,QAAQ;CACtC,MAAM,OAAO,KAAK;AAIlB,KADqB,MAAM,QAAQ,YAAY,WAAW,aAAa,KAClD,KAAM,QAAO;AAGlC,OAAM,QAAQ,oBAAoB,WAAW,aAAa;CAG1D,MAAM,YAAY,cAAc,QAAQ;CACxC,MAAM,YAAY,KAAK,KAAK;CAE5B,MAAM,SAAqB,UAAU,KAAK,GAAG,OAAO;EAClD,IAAI,QAAQ,WAAW,cAAc,GAAG,EAAE,WAAW,EAAE,QAAQ;EAC/D;EACA;EACA;EACA,MAAM;EACN,WAAW,EAAE;EACb,SAAS,EAAE;EACX,MAAM,EAAE;EACR,MAAM,EAAE;EACR;EACA,WAAW;EACZ,EAAE;AAGH,OAAM,QAAQ,aAAa,OAAO;AAClC,OAAM,QAAQ,WAAW;EAAE;EAAW,MAAM;EAAc;EAAQ;EAAM;EAAM;EAAO;EAAM,CAAC;AAE5F,QAAO;;AAOT,eAAsB,wBACpB,SACA,WACA,UACA,gBACsB;CACtB,MAAM,SAAsB;EAAE,gBAAgB;EAAG,eAAe;EAAG,cAAc;EAAG;CAEpF,MAAM,eAA2F,EAAE;CAEnG,MAAM,eAAe,KAAK,UAAU,YAAY;AAChD,KAAI,WAAW,aAAa,CAC1B,cAAa,KAAK;EAAE,SAAS;EAAc,UAAU;EAAU,QAAQ;EAAU,MAAM;EAAa,CAAC;CAGvG,MAAM,YAAY,KAAK,UAAU,SAAS;AAC1C,MAAK,MAAM,WAAW,YAAY,UAAU,EAAE;EAE5C,MAAM,OAAO,WADG,SAAS,UAAU,QAAQ,CACX;AAChC,eAAa,KAAK;GAAE;GAAS,UAAU;GAAU,QAAQ;GAAU;GAAM,CAAC;;CAG5E,MAAM,WAAW,KAAK,UAAU,QAAQ;AACxC,MAAK,MAAM,WAAW,YAAY,SAAS,CACzC,cAAa,KAAK;EAAE;EAAS,UAAU;EAAU,QAAQ;EAAS,MAAM;EAAW,CAAC;CAItF;EACE,MAAM,YAAY,KAAK,KAAK;AAC5B,OAAK,MAAM,WAAW,YAAY,SAAS,EAAE;GAE3C,MAAM,OAAO,uBADI,SAAS,QAAQ,CACW;AAC7C,OAAI,CAAC,KAAM;GAEX,MAAM,gBAAgB,GADN,SAAS,UAAU,QAAQ,CACV;GAGjC,MAAM,aAAuB;IAC3B,IAHS,QAAQ,WAAW,eAAe,GAAG,GAAG,EAAE;IAG/C;IAAW,QAAQ;IAAS,MAAM;IACtC,MAAM;IAAe,WAAW;IAAG,SAAS;IAC5C,MAJW,WAAW,KAAK;IAIrB;IAAM;IAAW,WAAW;IACnC;AACD,OAAI;AACF,UAAM,QAAQ,aAAa,CAAC,WAAW,CAAC;WAClC;;;AAMZ,KAAI,CAAC,6BAA6B,SAAS,CACzC,MAAK,MAAM,WAAW,iBAAiB,SAAS,CAC9C,cAAa,KAAK;EAAE;EAAS,UAAU;EAAU,QAAQ;EAAW,MAAM;EAAS,CAAC;AAIxF,KAAI,kBAAkB,mBAAmB,UAAU;AACjD,OAAK,MAAM,WAAW,YAAY,eAAe,CAC/C,cAAa,KAAK;GAAE;GAAS,UAAU;GAAgB,QAAQ;GAAS,MAAM;GAAW,CAAC;EAI5F;GACE,MAAM,YAAY,KAAK,KAAK;AAC5B,QAAK,MAAM,WAAW,YAAY,eAAe,EAAE;IAEjD,MAAM,OAAO,uBADI,SAAS,QAAQ,CACW;AAC7C,QAAI,CAAC,KAAM;IAEX,MAAM,gBAAgB,GADN,SAAS,gBAAgB,QAAQ,CAChB;IAGjC,MAAM,aAAuB;KAC3B,IAHS,QAAQ,WAAW,eAAe,GAAG,GAAG,EAAE;KAG/C;KAAW,QAAQ;KAAS,MAAM;KACtC,MAAM;KAAe,WAAW;KAAG,SAAS;KAC5C,MAJW,WAAW,KAAK;KAIrB;KAAM;KAAW,WAAW;KACnC;AACD,QAAI;AACF,WAAM,QAAQ,aAAa,CAAC,WAAW,CAAC;YAClC;;;AAMZ,MAAI,eAAe,SAAS,SAAS,EAAE;GACrC,MAAM,mBAAmB,eAAe,MAAM,GAAG,GAAiB;GAClE,MAAM,iBAAiB,KAAK,kBAAkB,YAAY;AAC1D,OAAI,WAAW,eAAe,CAC5B,cAAa,KAAK;IAAE,SAAS;IAAgB,UAAU;IAAkB,QAAQ;IAAU,MAAM;IAAa,CAAC;GAEjH,MAAM,kBAAkB,KAAK,kBAAkB,SAAS;AACxD,QAAK,MAAM,WAAW,YAAY,gBAAgB,EAAE;IAElD,MAAM,OAAO,WADG,SAAS,kBAAkB,QAAQ,CACnB;AAChC,iBAAa,KAAK;KAAE;KAAS,UAAU;KAAkB,QAAQ;KAAU;KAAM,CAAC;;;;AAKxF,OAAM,kBAAkB;CAExB,IAAI,kBAAkB;AAEtB,MAAK,MAAM,EAAE,SAAS,UAAU,QAAQ,UAAU,cAAc;AAC9D,MAAI,mBAAmB,mBAAmB;AACxC,SAAM,kBAAkB;AACxB,qBAAkB;;AAEpB;EAEA,MAAM,UAAU,SAAS,UAAU,QAAQ;AAC3C,MAAI;AAGF,OAFgB,MAAM,qBAAqB,SAAS,WAAW,UAAU,SAAS,QAAQ,KAAK,EAElF;IACX,MAAM,MAAM,MAAM,QAAQ,YAAY,WAAW,QAAQ;AACzD,WAAO;AACP,WAAO,iBAAiB,IAAI;SAE5B,QAAO;UAEH;AAEN,UAAO;;;CAKX,MAAM,4BAAY,IAAI,KAAa;AACnC,MAAK,MAAM,EAAE,SAAS,cAAc,aAClC,WAAU,IAAI,SAAS,UAAU,QAAQ,CAAC;CAG5C,MAAM,eAAe,MAAM,QAAQ,sBAAsB,UAAU;CAEnE,MAAM,aAAuB,EAAE;AAC/B,MAAK,MAAM,KAAK,cAAc;EAC5B,MAAM,WAAW,EAAE,SAAS,UAAU,GAAG,EAAE,MAAM,GAAG,GAAkB,GAAG;AACzE,MAAI,CAAC,UAAU,IAAI,SAAS,CAC1B,YAAW,KAAK,EAAE;;AAItB,KAAI,WAAW,SAAS,EACtB,OAAM,QAAQ,YAAY,WAAW,WAAW;AAGlD,QAAO;;AAOT,MAAM,mBAAmB;AACzB,MAAM,oBAAoB;;;;;;;;;;;;;AAc1B,eAAsB,uBACpB,SACA,YACA,cACiB;CACjB,MAAM,EAAE,mBAAmB,uBAAuB,MAAM,OAAO;CAE/D,MAAM,OAAO,MAAM,QAAQ,uBAAuB;AAClD,KAAI,KAAK,WAAW,EAAG,QAAO;CAE9B,MAAM,QAAQ,KAAK;CACnB,IAAI,WAAW;CAGf,MAAM,qCAAqB,IAAI,KAAoD;AACnF,MAAK,MAAM,OAAO,MAAM;EACtB,MAAM,QAAQ,mBAAmB,IAAI,IAAI,WAAW;AACpD,MAAI,MACF,OAAM;MAEN,oBAAmB,IAAI,IAAI,YAAY;GAAE,OAAO;GAAG,YAAY,IAAI;GAAM,CAAC;;CAG9E,MAAM,SAAS,QAAgB,cAAc,IAAI,IAAI,IAAI,WAAW;CACpE,MAAM,iBAAiB,MAAM,KAAK,mBAAmB,SAAS,CAAC,CAC5D,KAAK,CAAC,KAAK,EAAE,OAAO,kBAAkB,KAAK,MAAM,IAAI,CAAC,IAAI,MAAM,gBAAgB,WAAW,GAAG,CAC9F,KAAK,KAAK;AACb,SAAQ,OAAO,MACb,4BAA4B,MAAM,4BAA4B,mBAAmB,KAAK,eAAe,eAAe,IACrH;CAGD,IAAI,mBAAmB;CACvB,IAAI,kBAAkB;AAEtB,MAAK,IAAI,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK,kBAAkB;AAEtD,MAAI,cAAc,EAAE;AAClB,WAAQ,OAAO,MACb,2CAA2C,SAAS,GAAG,MAAM,gCAC9D;AACD;;EAGF,MAAM,QAAQ,KAAK,MAAM,GAAG,IAAI,iBAAiB;AAEjD,OAAK,IAAI,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK;GACrC,MAAM,EAAE,IAAI,MAAM,YAAY,SAAS,MAAM;AAG7C,OAAI,eAAe,kBAAkB;AACnC,QAAI,qBAAqB,GACvB,SAAQ,OAAO,MACb,yBAAyB,MAAM,iBAAiB,CAAC,IAAI,gBAAgB,oBACtE;IAEH,MAAM,OAAO,mBAAmB,IAAI,WAAW;AAC/C,YAAQ,OAAO,MACb,0BAA0B,MAAM,WAAW,CAAC,IAAI,MAAM,SAAS,IAAI,uBAAuB,KAAK,KAChG;AACD,uBAAmB;AACnB,sBAAkB;;AAIpB,QAAK,WAAW,KAAK,sBAAsB,EACzC,OAAM,kBAAkB;GAI1B,MAAM,OAAO,mBADD,MAAM,kBAAkB,KAAK,CACL;AACpC,SAAM,QAAQ,gBAAgB,IAAI,KAAK;AACvC;;AAGF,cAAY,MAAM;EAGlB,MAAM,YAAY,MAAM,MAAM,SAAS;AACvC,UAAQ,OAAO,MACb,yBAAyB,SAAS,GAAG,MAAM,WAAW,MAAM,UAAU,WAAW,CAAC,IAAI,UAAU,KAAK,KACtG;;AAIH,KAAI,qBAAqB,GACvB,SAAQ,OAAO,MACb,yBAAyB,MAAM,iBAAiB,CAAC,IAAI,gBAAgB,oBACtE;AAGH,QAAO;;AAOT,eAAsB,oBACpB,SACA,YACoD;CACpD,MAAM,WAAW,WACd,QAAQ,+EAA+E,CACvF,KAAK;CAER,MAAM,SAAsB;EAAE,gBAAgB;EAAG,eAAe;EAAG,cAAc;EAAG;AAEpF,MAAK,MAAM,WAAW,UAAU;AAC9B,QAAM,kBAAkB;EACxB,MAAM,IAAI,MAAM,wBAAwB,SAAS,QAAQ,IAAI,QAAQ,WAAW,QAAQ,iBAAiB;AACzG,SAAO,kBAAkB,EAAE;AAC3B,SAAO,iBAAiB,EAAE;AAC1B,SAAO,gBAAgB,EAAE;;AAG3B,QAAO;EAAE,UAAU,SAAS;EAAQ,QAAQ;EAAQ"}

package/dist/{kg-extraction-r5KkTmWY.mjs → kg-extraction-iR1BKKWK.mjs}

@@ -2,7 +2,7 @@ import { _ as warn, c as ok, i as err, n as dim, r as encodeDir, s as now, t as
 import { a as slugify, i as parseSessionFilename, n as decodeEncodedDir, t as buildEncodedDirMap } from "./migrate-fLD6rAdO.mjs";
 import { n as ensurePaiMarker, t as discoverPaiMarkers } from "./pai-marker-CXQPX2P6.mjs";
 import { i as kgAdd, o as kgInvalidate, r as upsertKgEntity, s as kgQuery } from "./kg-entity-DVzTsy6G.mjs";
-import { existsSync, mkdirSync, readFileSync, readdirSync, statSync, writeFileSync } from "node:fs";
+import { existsSync, mkdirSync, readFileSync, readdirSync, realpathSync, statSync, writeFileSync } from "node:fs";
 import { homedir } from "node:os";
 import { basename, join, resolve } from "node:path";
 
@@ -71,6 +71,35 @@ function upsertSession(db, projectId, number, date, slug, title, filename) {
 //#endregion
 //#region src/cli/commands/registry/scan.ts
 /** Registry scan command: walk ~/.claude/projects/ and populate the registry. */
+/**
+* Build a reverse map from encoded-dir → real path using the clc session
+* registry (~/.claude/session.json).
+*
+* The clc registry stores { sessions: [{ directory, ... }, ...] }. For each
+* entry with a resolvable directory, we encode the realpath and add it to
+* the map. This lets the scanner recover project paths that are not in
+* Claude's session-registry.json but were registered by clc.
+*/
+function buildClcDirMap() {
+	const map = /* @__PURE__ */ new Map();
+	const sessionFile = join(homedir(), ".claude", "session.json");
+	if (!existsSync(sessionFile)) return map;
+	try {
+		const raw = readFileSync(sessionFile, "utf8");
+		const parsed = JSON.parse(raw);
+		for (const entry of parsed.sessions ?? []) {
+			const dir = entry.directory;
+			if (!dir) continue;
+			try {
+				const real = realpathSync(dir);
+				if (!existsSync(real)) continue;
+				const encoded = encodeDir(real);
+				map.set(encoded, real);
+			} catch {}
+		}
+	} catch {}
+	return map;
+}
 const CLAUDE_PROJECTS_DIR = join(homedir(), ".claude", "projects");
 const PAI_CONFIG_DIR = join(homedir(), ".pai");
 const PAI_CONFIG_FILE = join(PAI_CONFIG_DIR, "config.json");
@@ -121,8 +150,13 @@ function performScan(db) {
 		return statSync(join(CLAUDE_PROJECTS_DIR, name)).isDirectory();
 	});
 	const lookupMap = buildEncodedDirMap();
+	const clcMap = buildClcDirMap();
 	for (const encodedDir of entries) {
-		const rootPath = decodeEncodedDir(encodedDir, lookupMap);
+		let rootPath = decodeEncodedDir(encodedDir, lookupMap);
+		if (!existsSync(rootPath) && clcMap.has(encodedDir)) {
+			const clcPath = clcMap.get(encodedDir);
+			if (existsSync(clcPath)) rootPath = clcPath;
+		}
 		if (!existsSync(rootPath)) {
 			result.skipped.push(`${encodedDir} (decoded: ${rootPath} — path not found on disk)`);
 			result.projectsScanned++;
@@ -521,4 +555,4 @@ async function extractAndStoreTriples(pool, params) {
 
 //#endregion
 export { resolveHome as a, upsertSession as c, performScan as i, cmdScan as n, saveScanConfig as o, loadScanConfig as r, upsertProject as s, extractAndStoreTriples as t };
-//# sourceMappingURL=kg-extraction-r5KkTmWY.mjs.map
+//# sourceMappingURL=kg-extraction-iR1BKKWK.mjs.map

package/dist/kg-extraction-iR1BKKWK.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"kg-extraction-iR1BKKWK.mjs","names":[],"sources":["../src/cli/commands/registry/utils.ts","../src/cli/commands/registry/scan.ts","../src/daemon/templates/triple-extraction-prompt.ts","../src/memory/kg-extraction.ts"],"sourcesContent":["/** Shared database helpers for registry command operations. */\n\nimport type { Database } from \"better-sqlite3\";\nimport { now } from \"../../utils.js\";\n\n/**\n * Upsert a project row. Returns { id, isNew }.\n *\n * Matching priority:\n *  1. root_path  — most reliable; handles slug collisions\n *  2. encoded_dir — Claude project dirs are canonical\n *  3. Insert with suffix-deduplication on slug collision\n */\nexport function upsertProject(\n  db: Database,\n  slug: string,\n  rootPath: string,\n  encodedDir: string\n): { id: number; isNew: boolean } {\n  const ts = now();\n\n  const byPath = db\n    .prepare(\"SELECT id FROM projects WHERE root_path = ?\")\n    .get(rootPath) as { id: number } | undefined;\n\n  if (byPath) {\n    const encodedOwner = db\n      .prepare(\"SELECT id FROM projects WHERE encoded_dir = ?\")\n      .get(encodedDir) as { id: number } | undefined;\n\n    if (!encodedOwner || encodedOwner.id === byPath.id) {\n      db.prepare(\n        \"UPDATE projects SET encoded_dir = ?, updated_at = ? WHERE id = ?\"\n      ).run(encodedDir, ts, byPath.id);\n    }\n    return { id: byPath.id, isNew: false };\n  }\n\n  const byEncoded = db\n    .prepare(\"SELECT id FROM projects WHERE encoded_dir = ?\")\n    .get(encodedDir) as { id: number } | undefined;\n\n  if (byEncoded) {\n    const pathOwner = db\n      .prepare(\"SELECT id FROM projects WHERE root_path = ?\")\n      .get(rootPath) as { id: number } | undefined;\n\n    if (!pathOwner || pathOwner.id === byEncoded.id) {\n      db.prepare(\n        \"UPDATE projects SET root_path = ?, updated_at = ? WHERE id = ?\"\n      ).run(rootPath, ts, byEncoded.id);\n    }\n    return { id: byEncoded.id, isNew: false };\n  }\n\n  // Insert — deduplicate slug with numeric suffix if needed.\n  let finalSlug = slug;\n  let attempt = 0;\n  while (true) {\n    const conflict = db\n      .prepare(\"SELECT id FROM projects WHERE slug = ?\")\n      .get(finalSlug) as { id: number } | undefined;\n    if (!conflict) break;\n    attempt++;\n    finalSlug = `${slug}-${attempt}`;\n  }\n\n  const result = db\n    .prepare(\n      `INSERT OR IGNORE INTO projects\n         (slug, display_name, root_path, encoded_dir, type, status, created_at, updated_at)\n       VALUES (?, ?, ?, ?, 'local', 'active', ?, ?)`\n    )\n    .run(finalSlug, finalSlug, rootPath, encodedDir, ts, ts);\n\n  if (result.changes === 0) {\n    const fallback =\n      (db.prepare(\"SELECT id FROM projects WHERE encoded_dir = ?\").get(encodedDir) as { id: number } | undefined) ??\n      (db.prepare(\"SELECT id FROM projects WHERE root_path = ?\").get(rootPath) as { id: number } | undefined);\n\n    if (fallback) {\n      return { id: fallback.id, isNew: false };\n    }\n\n    throw new Error(\n      `upsertProject: INSERT OR IGNORE was suppressed but no matching row found ` +\n      `for root_path=${rootPath} encoded_dir=${encodedDir}`\n    );\n  }\n\n  return { id: result.lastInsertRowid as number, isNew: true };\n}\n\n/** Upsert a session note. Returns true if newly inserted. */\nexport function upsertSession(\n  db: Database,\n  projectId: number,\n  number: number,\n  date: string,\n  slug: string,\n  title: string,\n  filename: string\n): boolean {\n  const existing = db\n    .prepare(\"SELECT id FROM sessions WHERE project_id = ? AND number = ?\")\n    .get(projectId, number);\n\n  if (existing) return false;\n\n  const ts = now();\n  db.prepare(\n    `INSERT INTO sessions\n       (project_id, number, date, slug, title, filename, status, created_at)\n     VALUES (?, ?, ?, ?, ?, ?, 'completed', ?)`\n  ).run(projectId, number, date, slug, title, filename, ts);\n\n  return true;\n}\n","/** Registry scan command: walk ~/.claude/projects/ and populate the registry. */\n\nimport { existsSync, readdirSync, statSync, readFileSync, writeFileSync, mkdirSync } from \"node:fs\";\nimport { realpathSync } from \"node:fs\";\nimport { join, basename, resolve } from \"node:path\";\nimport { homedir } from \"node:os\";\nimport { ok, warn, err, dim, bold } from \"../../utils.js\";\nimport { encodeDir } from \"../../utils.js\";\nimport { decodeEncodedDir, slugify, parseSessionFilename, buildEncodedDirMap } from \"../../../registry/migrate.js\";\nimport { ensurePaiMarker, discoverPaiMarkers } from \"../../../registry/pai-marker.js\";\nimport { upsertProject, upsertSession } from \"./utils.js\";\nimport type { Database } from \"better-sqlite3\";\n\n// ---------------------------------------------------------------------------\n// clc session.json fallback map\n// ---------------------------------------------------------------------------\n\n/**\n * Build a reverse map from encoded-dir → real path using the clc session\n * registry (~/.claude/session.json).\n *\n * The clc registry stores { sessions: [{ directory, ... }, ...] }. For each\n * entry with a resolvable directory, we encode the realpath and add it to\n * the map. This lets the scanner recover project paths that are not in\n * Claude's session-registry.json but were registered by clc.\n */\nfunction buildClcDirMap(): Map<string, string> {\n  const map = new Map<string, string>();\n  const sessionFile = join(homedir(), \".claude\", \"session.json\");\n  if (!existsSync(sessionFile)) return map;\n\n  try {\n    const raw = readFileSync(sessionFile, \"utf8\");\n    const parsed = JSON.parse(raw) as { sessions?: Array<{ directory?: string }> };\n    for (const entry of parsed.sessions ?? []) {\n      const dir = entry.directory;\n      if (!dir) continue;\n      try {\n        const real = realpathSync(dir);\n        if (!existsSync(real)) continue;\n        const encoded = encodeDir(real);\n        map.set(encoded, real);\n      } catch {\n        // Unresolvable path — skip\n      }\n    }\n  } catch {\n    // Unparseable — return empty map\n  }\n  return map;\n}\n\n// ---------------------------------------------------------------------------\n// Config helpers\n// ---------------------------------------------------------------------------\n\nconst CLAUDE_PROJECTS_DIR = join(homedir(), \".claude\", \"projects\");\nconst PAI_CONFIG_DIR = join(homedir(), \".pai\");\nconst PAI_CONFIG_FILE = join(PAI_CONFIG_DIR, \"config.json\");\n\ninterface PaiConfig {\n  scan_dirs: string[];\n}\n\nexport function loadScanConfig(): PaiConfig {\n  if (!existsSync(PAI_CONFIG_FILE)) return { scan_dirs: [] };\n  try {\n    return JSON.parse(readFileSync(PAI_CONFIG_FILE, \"utf8\")) as PaiConfig;\n  } catch {\n    return { scan_dirs: [] };\n  }\n}\n\nexport function saveScanConfig(config: PaiConfig): void {\n  mkdirSync(PAI_CONFIG_DIR, { recursive: true });\n  writeFileSync(PAI_CONFIG_FILE, JSON.stringify(config, null, 2) + \"\\n\", \"utf8\");\n}\n\nexport function resolveHome(p: string): string {\n  if (p.startsWith(\"~/\")) return join(homedir(), p.slice(2));\n  return resolve(p);\n}\n\n// ---------------------------------------------------------------------------\n// File discovery\n// ---------------------------------------------------------------------------\n\n/**\n * Recursively find all .md files in a directory, including YYYY/MM subdirectories.\n * Returns filenames (basename only).\n */\nexport function findNoteFiles(dir: string): string[] {\n  const results: string[] = [];\n  if (!existsSync(dir)) return results;\n\n  for (const entry of readdirSync(dir, { withFileTypes: true })) {\n    if (entry.isFile() && entry.name.endsWith(\".md\")) {\n      results.push(entry.name);\n    } else if (entry.isDirectory() && /^\\d{4}$/.test(entry.name)) {\n      const yearDir = join(dir, entry.name);\n      for (const monthEntry of readdirSync(yearDir, { withFileTypes: true })) {\n        if (monthEntry.isDirectory() && /^\\d{2}$/.test(monthEntry.name)) {\n          const monthDir = join(yearDir, monthEntry.name);\n          for (const noteEntry of readdirSync(monthDir, { withFileTypes: true })) {\n            if (noteEntry.isFile() && noteEntry.name.endsWith(\".md\")) {\n              results.push(noteEntry.name);\n            }\n          }\n        }\n      }\n    }\n  }\n  return results;\n}\n\n// ---------------------------------------------------------------------------\n// Scan result type\n// ---------------------------------------------------------------------------\n\nexport interface ScanResult {\n  projectsScanned: number;\n  projectsNew: number;\n  projectsUpdated: number;\n  sessionsScanned: number;\n  sessionsNew: number;\n  skipped: string[];\n}\n\n// ---------------------------------------------------------------------------\n// Core scan logic\n// ---------------------------------------------------------------------------\n\nexport function performScan(db: Database): ScanResult {\n  const result: ScanResult = {\n    projectsScanned: 0,\n    projectsNew: 0,\n    projectsUpdated: 0,\n    sessionsScanned: 0,\n    sessionsNew: 0,\n    skipped: [],\n  };\n\n  if (!existsSync(CLAUDE_PROJECTS_DIR)) {\n    throw new Error(`Claude projects directory not found: ${CLAUDE_PROJECTS_DIR}`);\n  }\n\n  const entries = readdirSync(CLAUDE_PROJECTS_DIR).filter((name) => {\n    const full = join(CLAUDE_PROJECTS_DIR, name);\n    return statSync(full).isDirectory();\n  });\n\n  const lookupMap = buildEncodedDirMap();\n  const clcMap = buildClcDirMap();\n\n  for (const encodedDir of entries) {\n    let rootPath = decodeEncodedDir(encodedDir, lookupMap);\n\n    // If the decoded path doesn't exist, try the clc session.json fallback map.\n    // The clc map is keyed by encodeDir(realpathSync(directory)), so paths that\n    // were registered by clc but not in session-registry.json can be recovered.\n    if (!existsSync(rootPath) && clcMap.has(encodedDir)) {\n      const clcPath = clcMap.get(encodedDir)!;\n      if (existsSync(clcPath)) {\n        rootPath = clcPath;\n      }\n    }\n\n    if (!existsSync(rootPath)) {\n      result.skipped.push(`${encodedDir} (decoded: ${rootPath} — path not found on disk)`);\n      result.projectsScanned++;\n      continue;\n    }\n\n    const slug = slugify(basename(rootPath) || encodedDir);\n    const { id, isNew } = upsertProject(db, slug, rootPath, encodedDir);\n\n    result.projectsScanned++;\n    if (isNew) result.projectsNew++;\n    else result.projectsUpdated++;\n\n    try {\n      ensurePaiMarker(rootPath, slug);\n    } catch {\n      // Non-fatal\n    }\n\n    const claudeNotesDir = join(CLAUDE_PROJECTS_DIR, encodedDir, \"Notes\");\n\n    if (existsSync(claudeNotesDir)) {\n      const rootNotesDir = join(rootPath, \"Notes\");\n      if (claudeNotesDir !== rootNotesDir) {\n        db.prepare(\n          \"UPDATE projects SET claude_notes_dir = ?, updated_at = ? WHERE id = ?\"\n        ).run(claudeNotesDir, Date.now(), id);\n      }\n    }\n\n    if (!existsSync(claudeNotesDir)) continue;\n\n    const noteFiles = findNoteFiles(claudeNotesDir);\n\n    for (const filename of noteFiles) {\n      const parsed = parseSessionFilename(filename);\n      if (!parsed) continue;\n\n      result.sessionsScanned++;\n      const isNewSession = upsertSession(db, id, parsed.number, parsed.date, parsed.slug, parsed.title, parsed.filename);\n      if (isNewSession) result.sessionsNew++;\n    }\n  }\n\n  // Phase 2: Scan project-root Notes/ for all registered active projects\n  {\n    const activeProjects = db\n      .prepare(\"SELECT id, slug, root_path FROM projects WHERE status = 'active'\")\n      .all() as { id: number; slug: string; root_path: string }[];\n\n    for (const project of activeProjects) {\n      const notesDir = join(project.root_path, \"Notes\");\n      if (!existsSync(notesDir)) continue;\n\n      let files: string[];\n      try {\n        files = findNoteFiles(notesDir);\n      } catch {\n        continue;\n      }\n\n      for (const filename of files) {\n        const parsed = parseSessionFilename(filename);\n        if (!parsed) continue;\n\n        result.sessionsScanned++;\n        const isNewSession = upsertSession(db, project.id, parsed.number, parsed.date, parsed.slug, parsed.title, parsed.filename);\n        if (isNewSession) result.sessionsNew++;\n      }\n    }\n  }\n\n  // Phase 3: Scan extra directories from config\n  const config = loadScanConfig();\n  if (config.scan_dirs.length) {\n    for (const rawDir of config.scan_dirs) {\n      const scanDir = resolveHome(rawDir);\n      if (!existsSync(scanDir)) {\n        result.skipped.push(`${rawDir} (configured scan_dir not found)`);\n        continue;\n      }\n\n      const children = readdirSync(scanDir).filter((name) => {\n        if (name.startsWith(\".\")) return false;\n        const full = join(scanDir, name);\n        try { return statSync(full).isDirectory(); } catch { return false; }\n      });\n\n      for (const child of children) {\n        const childPath = join(scanDir, child);\n        const childSlug = slugify(child);\n        const childEncoded = encodeDir(childPath);\n\n        const existing = db\n          .prepare(\"SELECT id FROM projects WHERE root_path = ?\")\n          .get(childPath) as { id: number } | undefined;\n\n        if (existing) {\n          result.projectsScanned++;\n          result.projectsUpdated++;\n\n          try { ensurePaiMarker(childPath, childSlug); } catch { /* non-fatal */ }\n\n          const notesDir = join(childPath, \"Notes\");\n          if (existsSync(notesDir)) {\n            const noteFiles = readdirSync(notesDir).filter((f) => f.endsWith(\".md\"));\n            for (const filename of noteFiles) {\n              const parsed = parseSessionFilename(filename);\n              if (!parsed) continue;\n              result.sessionsScanned++;\n              if (upsertSession(db, existing.id, parsed.number, parsed.date, parsed.slug, parsed.title, parsed.filename)) {\n                result.sessionsNew++;\n              }\n            }\n          }\n          continue;\n        }\n\n        const { id, isNew } = upsertProject(db, childSlug, childPath, childEncoded);\n        result.projectsScanned++;\n        if (isNew) result.projectsNew++;\n        else result.projectsUpdated++;\n\n        try { ensurePaiMarker(childPath, childSlug); } catch { /* non-fatal */ }\n\n        const notesDir = join(childPath, \"Notes\");\n        if (existsSync(notesDir)) {\n          const noteFiles = readdirSync(notesDir).filter((f) => f.endsWith(\".md\"));\n          for (const filename of noteFiles) {\n            const parsed = parseSessionFilename(filename);\n            if (!parsed) continue;\n            result.sessionsScanned++;\n            if (upsertSession(db, id, parsed.number, parsed.date, parsed.slug, parsed.title, parsed.filename)) {\n              result.sessionsNew++;\n            }\n          }\n        }\n      }\n    }\n  }\n\n  // Phase 4: Discover PAI.md markers in scan_dirs\n  if (config.scan_dirs.length) {\n    const resolvedScanDirs = config.scan_dirs.map(resolveHome).filter(existsSync);\n    const markers = discoverPaiMarkers(resolvedScanDirs);\n\n    for (const marker of markers) {\n      const registeredRow = db\n        .prepare(\"SELECT id, root_path, slug FROM projects WHERE slug = ?\")\n        .get(marker.slug) as { id: number; root_path: string; slug: string } | undefined;\n\n      if (!registeredRow) continue;\n\n      if (registeredRow.root_path !== marker.projectRoot) {\n        const newEncoded = encodeDir(marker.projectRoot);\n        const now4 = Date.now();\n\n        const encodedOwner = db\n          .prepare(\"SELECT id FROM projects WHERE encoded_dir = ?\")\n          .get(newEncoded) as { id: number } | undefined;\n        const pathOwner = db\n          .prepare(\"SELECT id FROM projects WHERE root_path = ?\")\n          .get(marker.projectRoot) as { id: number } | undefined;\n\n        const encodedSafe = !encodedOwner || encodedOwner.id === registeredRow.id;\n        const pathSafe = !pathOwner || pathOwner.id === registeredRow.id;\n\n        if (encodedSafe && pathSafe) {\n          db.prepare(\n            \"UPDATE projects SET root_path = ?, encoded_dir = ?, updated_at = ? WHERE id = ?\"\n          ).run(marker.projectRoot, newEncoded, now4, registeredRow.id);\n        } else if (pathSafe) {\n          db.prepare(\n            \"UPDATE projects SET root_path = ?, updated_at = ? WHERE id = ?\"\n          ).run(marker.projectRoot, now4, registeredRow.id);\n        }\n      }\n    }\n  }\n\n  return result;\n}\n\n// ---------------------------------------------------------------------------\n// cmdScan\n// ---------------------------------------------------------------------------\n\n/**\n * Run the registry scan CLI command.\n *\n * @param opts.quick  When true, skips verbose output (same scan, less noise).\n *                    The underlying scan is always incremental via upsert —\n *                    this flag exists for hook/daemon-triggered invocations that\n *                    want minimal log output.\n */\nexport function cmdScan(db: Database, opts: { quick?: boolean } = {}): void {\n  const config = loadScanConfig();\n  if (!opts.quick) {\n    console.log(dim(\"Scanning ~/.claude/projects/ ...\"));\n    if (config.scan_dirs.length) {\n      console.log(dim(`Scanning ${config.scan_dirs.length} extra dir(s): ${config.scan_dirs.join(\", \")}`));\n    }\n    console.log(dim(\"Scanning project-root Notes/ directories ...\"));\n  }\n\n  let result: ScanResult;\n  try {\n    result = performScan(db);\n  } catch (e) {\n    console.error(err(String(e)));\n    process.exit(1);\n  }\n\n  if (!opts.quick) {\n    console.log(\n      ok(`Scanned ${bold(String(result.projectsScanned))} projects, ${bold(String(result.sessionsScanned))} session notes.`)\n    );\n    console.log(dim(`  Projects: ${result.projectsNew} new, ${result.projectsUpdated} updated`));\n    console.log(dim(`  Sessions: ${result.sessionsNew} new`));\n\n    if (result.skipped.length) {\n      console.log();\n      console.log(warn(`  ${result.skipped.length} project(s) skipped (path not found on disk):`));\n      for (const s of result.skipped.slice(0, 10)) {\n        console.log(dim(`    ${s}`));\n      }\n      if (result.skipped.length > 10) {\n        console.log(dim(`    ... and ${result.skipped.length - 10} more`));\n      }\n    }\n  } else {\n    // Quick mode: single compact line (stderr so it doesn't pollute pipe output)\n    process.stderr.write(\n      `[registry-scan] ${result.projectsScanned} projects, ${result.sessionsScanned} sessions ` +\n      `(${result.projectsNew}+${result.sessionsNew} new).\\n`\n    );\n  }\n}\n","/**\n * triple-extraction-prompt.ts — Prompt template for KG triple extraction.\n *\n * Used by the session-summary-worker to extract structured facts from\n * a completed session summary and store them in the temporal knowledge graph.\n */\n\nexport function buildTripleExtractionPrompt(params: {\n  sessionContent: string;\n  projectSlug: string;\n  gitLog: string;\n}): string {\n  return `Extract structured entities and relations from this coding session.\n\nOutput a single JSON object with two arrays: \"entities\" and \"relations\".\n\nEntity types: project | person | concept | tool | file | version | decision | technology | organization\n\nRules:\n- Be SPECIFIC: entity names must be concrete (e.g., \"FSRS\", \"Glidr\", \"Matthias\")\n- Use snake_case relation verb phrases (e.g., \"uses_algorithm\", \"decided_to\", \"shipped_version\")\n- Skip opinions, speculation, and \"we should\" statements\n- Skip entities obvious from project metadata unless they have a meaningful relation\n- Maximum 15 relations per session — pick the most important\n- Each entity should have a brief description (1 sentence, what it is in this context)\n- Each relation must reference entity names that appear in the entities array\n\nExample output:\n{\n  \"entities\": [\n    {\"name\": \"Glidr\", \"type\": \"project\", \"description\": \"Flashcard app using FSRS spaced repetition\"},\n    {\"name\": \"FSRS\", \"type\": \"concept\", \"description\": \"Free Spaced Repetition Scheduler algorithm\"},\n    {\"name\": \"Matthias\", \"type\": \"person\", \"description\": \"Developer of Glidr and Quassl\"},\n    {\"name\": \"Quassl\", \"type\": \"project\", \"description\": \"iOS app being rewritten in Flutter\"},\n    {\"name\": \"Flutter\", \"type\": \"technology\", \"description\": \"Cross-platform mobile framework\"}\n  ],\n  \"relations\": [\n    {\"source\": \"Glidr\", \"relation\": \"uses_algorithm\", \"target\": \"FSRS\"},\n    {\"source\": \"Glidr\", \"relation\": \"shipped_version\", \"target\": \"1.0.5\"},\n    {\"source\": \"Matthias\", \"relation\": \"decided_to_rewrite\", \"target\": \"Quassl\"},\n    {\"source\": \"Quassl\", \"relation\": \"migrating_to\", \"target\": \"Flutter\"}\n  ]\n}\n\nPROJECT: ${params.projectSlug}\n\nSESSION CONTENT:\n${params.sessionContent}\n\nGIT COMMITS:\n${params.gitLog}\n\nJSON object (entities + relations):`;\n}\n","/**\n * kg-extraction.ts — Shared KG triple extraction logic.\n *\n * Extracted from session-summary-worker.ts so both the worker and the\n * CLI backfill (`pai kg backfill`) can use the same code path.\n *\n * Provides:\n *   - findClaudeBinary()       — locate the claude CLI\n *   - spawnClaude()            — generic prompt -> response runner (strips ANTHROPIC_API_KEY)\n *   - extractAndStoreTriples() — run the extractor prompt and persist triples to Postgres\n */\n\nimport { existsSync } from \"node:fs\";\nimport { join } from \"node:path\";\nimport { homedir } from \"node:os\";\nimport type { Pool } from \"pg\";\nimport type { Database } from \"better-sqlite3\";\n\nimport { buildTripleExtractionPrompt } from \"../daemon/templates/triple-extraction-prompt.js\";\nimport { kgAdd, kgQuery, kgInvalidate } from \"./kg.js\";\nimport { upsertKgEntity } from \"./kg-entity.js\";\n\n// ---------------------------------------------------------------------------\n// Claude CLI binary discovery\n// ---------------------------------------------------------------------------\n\n/**\n * Find the `claude` CLI binary. Checks common installation locations first\n * (launchd PATH is minimal so bare \"claude\" often won't resolve).\n */\nexport function findClaudeBinary(): string | null {\n  const candidates = [\n    join(homedir(), \".local\", \"bin\", \"claude\"),\n    join(homedir(), \".claude\", \"local\", \"claude\"),\n    \"/usr/local/bin/claude\",\n    \"/opt/homebrew/bin/claude\",\n  ];\n\n  for (const candidate of candidates) {\n    try {\n      if (existsSync(candidate)) return candidate;\n    } catch { /* skip */ }\n  }\n  return \"claude\";\n}\n\nconst CLAUDE_TIMEOUT_MS: Record<string, number> = {\n  haiku: 60_000,\n  sonnet: 120_000,\n  opus: 300_000,\n};\n\n/**\n * Spawn the claude CLI with a prompt on stdin and return stdout.\n *\n * IMPORTANT: ANTHROPIC_API_KEY is stripped from the spawned environment so\n * the CLI uses the user's Max plan (free) instead of billing the API key.\n */\nexport async function spawnClaude(\n  prompt: string,\n  model: \"haiku\" | \"sonnet\" | \"opus\" = \"sonnet\"\n): Promise<string | null> {\n  const claudeBin = findClaudeBinary();\n  if (!claudeBin) {\n    process.stderr.write(\"[kg-extraction] claude CLI not found.\\n\");\n    return null;\n  }\n\n  const { spawn } = await import(\"node:child_process\");\n\n  return new Promise((resolve) => {\n    let timer: ReturnType<typeof setTimeout> | null = null;\n\n    const { ANTHROPIC_API_KEY: _drop, ...envWithoutApiKey } = process.env;\n    const child = spawn(\n      claudeBin,\n      [\"--model\", model, \"-p\", \"--no-session-persistence\"],\n      { env: envWithoutApiKey, stdio: [\"pipe\", \"pipe\", \"pipe\"] }\n    );\n\n    let stdout = \"\";\n    let stderr = \"\";\n\n    child.stdout.on(\"data\", (chunk: Buffer) => { stdout += chunk.toString(); });\n    child.stderr.on(\"data\", (chunk: Buffer) => { stderr += chunk.toString(); });\n\n    child.on(\"error\", (err: Error) => {\n      if (timer) { clearTimeout(timer); timer = null; }\n      process.stderr.write(`[kg-extraction] ${model} spawn error: ${err.message}\\n`);\n      resolve(null);\n    });\n\n    child.on(\"close\", (code: number | null) => {\n      if (timer) { clearTimeout(timer); timer = null; }\n      if (code !== 0) {\n        process.stderr.write(\n          `[kg-extraction] ${model} exited ${code}: ${stderr.slice(0, 300)}\\n`\n        );\n        resolve(null);\n      } else {\n        resolve(stdout.trim() || null);\n      }\n    });\n\n    timer = setTimeout(() => {\n      process.stderr.write(`[kg-extraction] ${model} timed out — killing process.\\n`);\n      child.kill(\"SIGTERM\");\n      resolve(null);\n    }, CLAUDE_TIMEOUT_MS[model] ?? 120_000);\n\n    child.stdin.write(prompt);\n    child.stdin.end();\n  });\n}\n\n// ---------------------------------------------------------------------------\n// Triple extraction\n// ---------------------------------------------------------------------------\n\nexport interface ExtractTriplesParams {\n  summaryText: string;\n  projectSlug: string;\n  projectId: number | null;\n  sessionId: string;\n  gitLog?: string;\n  model?: \"haiku\" | \"sonnet\" | \"opus\";\n  /** Optional federation SQLite db — when provided, entities are upserted into kg_entities (QW1) */\n  federationDb?: Database;\n  /** Tenant ID for multi-tenant entity scoping (default: \"default\") */\n  tenantId?: string;\n}\n\nexport interface ExtractTriplesResult {\n  extracted: number;\n  added: number;\n  superseded: number;\n}\n\n/**\n * Extract structured KG triples from a session summary and store them in\n * Postgres. Idempotent: if a (subject, predicate) pair already has the same\n * object, no new row is added; if the object differs, the old triple is\n * invalidated (valid_to = NOW()) and a new one is inserted.\n *\n * Best-effort: per-triple errors are caught and logged but never thrown.\n * Returns a small stats object so callers can report progress.\n */\nexport async function extractAndStoreTriples(\n  pool: Pool,\n  params: ExtractTriplesParams\n): Promise<ExtractTriplesResult> {\n  const stats: ExtractTriplesResult = { extracted: 0, added: 0, superseded: 0 };\n\n  const prompt = buildTripleExtractionPrompt({\n    sessionContent: params.summaryText,\n    projectSlug: params.projectSlug,\n    gitLog: params.gitLog ?? \"\",\n  });\n\n  const jsonOutput = await spawnClaude(prompt, params.model ?? \"sonnet\");\n  if (!jsonOutput) return stats;\n\n  // Strip markdown code fences if Claude wrapped the JSON\n  const cleaned = jsonOutput\n    .replace(/^```json\\s*/m, \"\")\n    .replace(/^```\\s*/m, \"\")\n    .replace(/\\s*```$/m, \"\")\n    .trim();\n\n  // Support both legacy array format and new structured format\n  type LegacyTriple = { subject: string; predicate: string; object: string };\n  type NewRelation = { source: string; relation: string; target: string };\n  type NewEntity = { name: string; type: string; description: string };\n  type NewFormat = { entities: NewEntity[]; relations: NewRelation[] };\n\n  let triples: Array<LegacyTriple>;\n  try {\n    const parsed = JSON.parse(cleaned);\n\n    if (Array.isArray(parsed)) {\n      // Legacy format: [{subject, predicate, object}]\n      triples = parsed;\n    } else if (parsed && typeof parsed === \"object\" && Array.isArray(parsed.relations)) {\n      // New structured format: {entities: [...], relations: [...]}\n      const newFmt = parsed as NewFormat;\n\n      // QW1: Upsert entities into federation SQLite kg_entities table when db is available\n      if (params.federationDb && Array.isArray(newFmt.entities)) {\n        const tenantId = params.tenantId ?? \"default\";\n        for (const entity of newFmt.entities) {\n          if (!entity.name) continue;\n          try {\n            upsertKgEntity(params.federationDb, {\n              name: entity.name,\n              type: entity.type ?? \"unknown\",\n              description: entity.description,\n              tenantId,\n            });\n          } catch (entityErr) {\n            process.stderr.write(`[kg-extraction] entity upsert error (${entity.name}): ${entityErr}\\n`);\n          }\n        }\n      }\n\n      triples = newFmt.relations.map((r: NewRelation) => ({\n        subject: r.source,\n        predicate: r.relation,\n        object: r.target,\n      }));\n    } else {\n      process.stderr.write(`[kg-extraction] Unexpected JSON shape — neither array nor {entities,relations}\\n`);\n      return stats;\n    }\n  } catch (e) {\n    process.stderr.write(`[kg-extraction] JSON parse failed: ${e}\\n`);\n    return stats;\n  }\n\n  if (!Array.isArray(triples)) return stats;\n  stats.extracted = triples.length;\n\n  for (const t of triples) {\n    if (!t.subject || !t.predicate || !t.object) continue;\n\n    try {\n      const existing = await kgQuery(pool, {\n        subject: t.subject,\n        predicate: t.predicate,\n        project_id: params.projectId ?? undefined,\n      });\n\n      // If an identical (subject, predicate, object) is already valid, skip — idempotent\n      const alreadyValid = existing.find((e) => e.object === t.object && !e.valid_to);\n      if (alreadyValid) continue;\n\n      // Invalidate any superseded triple (same subject+predicate, different object)\n      const supersedes = existing.find((e) => e.object !== t.object && !e.valid_to);\n      if (supersedes) {\n        await kgInvalidate(pool, supersedes.id);\n        stats.superseded++;\n      }\n\n      await kgAdd(pool, {\n        subject: t.subject,\n        predicate: t.predicate,\n        object: t.object,\n        project_id: params.projectId ?? undefined,\n        source_session: params.sessionId,\n        confidence: \"EXTRACTED\",\n      });\n      stats.added++;\n    } catch (tripleErr) {\n      process.stderr.write(`[kg-extraction] store error (${t.subject}): ${tripleErr}\\n`);\n    }\n  }\n\n  return stats;\n}\n"],"mappings":";;;;;;;;;;;;;;;;;AAaA,SAAgB,cACd,IACA,MACA,UACA,YACgC;CAChC,MAAM,KAAK,KAAK;CAEhB,MAAM,SAAS,GACZ,QAAQ,8CAA8C,CACtD,IAAI,SAAS;AAEhB,KAAI,QAAQ;EACV,MAAM,eAAe,GAClB,QAAQ,gDAAgD,CACxD,IAAI,WAAW;AAElB,MAAI,CAAC,gBAAgB,aAAa,OAAO,OAAO,GAC9C,IAAG,QACD,mEACD,CAAC,IAAI,YAAY,IAAI,OAAO,GAAG;AAElC,SAAO;GAAE,IAAI,OAAO;GAAI,OAAO;GAAO;;CAGxC,MAAM,YAAY,GACf,QAAQ,gDAAgD,CACxD,IAAI,WAAW;AAElB,KAAI,WAAW;EACb,MAAM,YAAY,GACf,QAAQ,8CAA8C,CACtD,IAAI,SAAS;AAEhB,MAAI,CAAC,aAAa,UAAU,OAAO,UAAU,GAC3C,IAAG,QACD,iEACD,CAAC,IAAI,UAAU,IAAI,UAAU,GAAG;AAEnC,SAAO;GAAE,IAAI,UAAU;GAAI,OAAO;GAAO;;CAI3C,IAAI,YAAY;CAChB,IAAI,UAAU;AACd,QAAO,MAAM;AAIX,MAAI,CAHa,GACd,QAAQ,yCAAyC,CACjD,IAAI,UAAU,CACF;AACf;AACA,cAAY,GAAG,KAAK,GAAG;;CAGzB,MAAM,SAAS,GACZ,QACC;;qDAGD,CACA,IAAI,WAAW,WAAW,UAAU,YAAY,IAAI,GAAG;AAE1D,KAAI,OAAO,YAAY,GAAG;EACxB,MAAM,WACH,GAAG,QAAQ,gDAAgD,CAAC,IAAI,WAAW,IAC3E,GAAG,QAAQ,8CAA8C,CAAC,IAAI,SAAS;AAE1E,MAAI,SACF,QAAO;GAAE,IAAI,SAAS;GAAI,OAAO;GAAO;AAG1C,QAAM,IAAI,MACR,0FACiB,SAAS,eAAe,aAC1C;;AAGH,QAAO;EAAE,IAAI,OAAO;EAA2B,OAAO;EAAM;;;AAI9D,SAAgB,cACd,IACA,WACA,QACA,MACA,MACA,OACA,UACS;AAKT,KAJiB,GACd,QAAQ,8DAA8D,CACtE,IAAI,WAAW,OAAO,CAEX,QAAO;CAErB,MAAM,KAAK,KAAK;AAChB,IAAG,QACD;;gDAGD,CAAC,IAAI,WAAW,QAAQ,MAAM,MAAM,OAAO,UAAU,GAAG;AAEzD,QAAO;;;;;;;;;;;;;;;AC1FT,SAAS,iBAAsC;CAC7C,MAAM,sBAAM,IAAI,KAAqB;CACrC,MAAM,cAAc,KAAK,SAAS,EAAE,WAAW,eAAe;AAC9D,KAAI,CAAC,WAAW,YAAY,CAAE,QAAO;AAErC,KAAI;EACF,MAAM,MAAM,aAAa,aAAa,OAAO;EAC7C,MAAM,SAAS,KAAK,MAAM,IAAI;AAC9B,OAAK,MAAM,SAAS,OAAO,YAAY,EAAE,EAAE;GACzC,MAAM,MAAM,MAAM;AAClB,OAAI,CAAC,IAAK;AACV,OAAI;IACF,MAAM,OAAO,aAAa,IAAI;AAC9B,QAAI,CAAC,WAAW,KAAK,CAAE;IACvB,MAAM,UAAU,UAAU,KAAK;AAC/B,QAAI,IAAI,SAAS,KAAK;WAChB;;SAIJ;AAGR,QAAO;;AAOT,MAAM,sBAAsB,KAAK,SAAS,EAAE,WAAW,WAAW;AAClE,MAAM,iBAAiB,KAAK,SAAS,EAAE,OAAO;AAC9C,MAAM,kBAAkB,KAAK,gBAAgB,cAAc;AAM3D,SAAgB,iBAA4B;AAC1C,KAAI,CAAC,WAAW,gBAAgB,CAAE,QAAO,EAAE,WAAW,EAAE,EAAE;AAC1D,KAAI;AACF,SAAO,KAAK,MAAM,aAAa,iBAAiB,OAAO,CAAC;SAClD;AACN,SAAO,EAAE,WAAW,EAAE,EAAE;;;AAI5B,SAAgB,eAAe,QAAyB;AACtD,WAAU,gBAAgB,EAAE,WAAW,MAAM,CAAC;AAC9C,eAAc,iBAAiB,KAAK,UAAU,QAAQ,MAAM,EAAE,GAAG,MAAM,OAAO;;AAGhF,SAAgB,YAAY,GAAmB;AAC7C,KAAI,EAAE,WAAW,KAAK,CAAE,QAAO,KAAK,SAAS,EAAE,EAAE,MAAM,EAAE,CAAC;AAC1D,QAAO,QAAQ,EAAE;;;;;;AAWnB,SAAgB,cAAc,KAAuB;CACnD,MAAM,UAAoB,EAAE;AAC5B,KAAI,CAAC,WAAW,IAAI,CAAE,QAAO;AAE7B,MAAK,MAAM,SAAS,YAAY,KAAK,EAAE,eAAe,MAAM,CAAC,CAC3D,KAAI,MAAM,QAAQ,IAAI,MAAM,KAAK,SAAS,MAAM,CAC9C,SAAQ,KAAK,MAAM,KAAK;UACf,MAAM,aAAa,IAAI,UAAU,KAAK,MAAM,KAAK,EAAE;EAC5D,MAAM,UAAU,KAAK,KAAK,MAAM,KAAK;AACrC,OAAK,MAAM,cAAc,YAAY,SAAS,EAAE,eAAe,MAAM,CAAC,CACpE,KAAI,WAAW,aAAa,IAAI,UAAU,KAAK,WAAW,KAAK,EAAE;GAC/D,MAAM,WAAW,KAAK,SAAS,WAAW,KAAK;AAC/C,QAAK,MAAM,aAAa,YAAY,UAAU,EAAE,eAAe,MAAM,CAAC,CACpE,KAAI,UAAU,QAAQ,IAAI,UAAU,KAAK,SAAS,MAAM,CACtD,SAAQ,KAAK,UAAU,KAAK;;;AAOxC,QAAO;;AAoBT,SAAgB,YAAY,IAA0B;CACpD,MAAM,SAAqB;EACzB,iBAAiB;EACjB,aAAa;EACb,iBAAiB;EACjB,iBAAiB;EACjB,aAAa;EACb,SAAS,EAAE;EACZ;AAED,KAAI,CAAC,WAAW,oBAAoB,CAClC,OAAM,IAAI,MAAM,wCAAwC,sBAAsB;CAGhF,MAAM,UAAU,YAAY,oBAAoB,CAAC,QAAQ,SAAS;AAEhE,SAAO,SADM,KAAK,qBAAqB,KAAK,CACvB,CAAC,aAAa;GACnC;CAEF,MAAM,YAAY,oBAAoB;CACtC,MAAM,SAAS,gBAAgB;AAE/B,MAAK,MAAM,cAAc,SAAS;EAChC,IAAI,WAAW,iBAAiB,YAAY,UAAU;AAKtD,MAAI,CAAC,WAAW,SAAS,IAAI,OAAO,IAAI,WAAW,EAAE;GACnD,MAAM,UAAU,OAAO,IAAI,WAAW;AACtC,OAAI,WAAW,QAAQ,CACrB,YAAW;;AAIf,MAAI,CAAC,WAAW,SAAS,EAAE;AACzB,UAAO,QAAQ,KAAK,GAAG,WAAW,aAAa,SAAS,4BAA4B;AACpF,UAAO;AACP;;EAGF,MAAM,OAAO,QAAQ,SAAS,SAAS,IAAI,WAAW;EACtD,MAAM,EAAE,IAAI,UAAU,cAAc,IAAI,MAAM,UAAU,WAAW;AAEnE,SAAO;AACP,MAAI,MAAO,QAAO;MACb,QAAO;AAEZ,MAAI;AACF,mBAAgB,UAAU,KAAK;UACzB;EAIR,MAAM,iBAAiB,KAAK,qBAAqB,YAAY,QAAQ;AAErE,MAAI,WAAW,eAAe,EAE5B;OAAI,mBADiB,KAAK,UAAU,QAAQ,CAE1C,IAAG,QACD,wEACD,CAAC,IAAI,gBAAgB,KAAK,KAAK,EAAE,GAAG;;AAIzC,MAAI,CAAC,WAAW,eAAe,CAAE;EAEjC,MAAM,YAAY,cAAc,eAAe;AAE/C,OAAK,MAAM,YAAY,WAAW;GAChC,MAAM,SAAS,qBAAqB,SAAS;AAC7C,OAAI,CAAC,OAAQ;AAEb,UAAO;AAEP,OADqB,cAAc,IAAI,IAAI,OAAO,QAAQ,OAAO,MAAM,OAAO,MAAM,OAAO,OAAO,OAAO,SAAS,CAChG,QAAO;;;CAK7B;EACE,MAAM,iBAAiB,GACpB,QAAQ,mEAAmE,CAC3E,KAAK;AAER,OAAK,MAAM,WAAW,gBAAgB;GACpC,MAAM,WAAW,KAAK,QAAQ,WAAW,QAAQ;AACjD,OAAI,CAAC,WAAW,SAAS,CAAE;GAE3B,IAAI;AACJ,OAAI;AACF,YAAQ,cAAc,SAAS;WACzB;AACN;;AAGF,QAAK,MAAM,YAAY,OAAO;IAC5B,MAAM,SAAS,qBAAqB,SAAS;AAC7C,QAAI,CAAC,OAAQ;AAEb,WAAO;AAEP,QADqB,cAAc,IAAI,QAAQ,IAAI,OAAO,QAAQ,OAAO,MAAM,OAAO,MAAM,OAAO,OAAO,OAAO,SAAS,CACxG,QAAO;;;;CAM/B,MAAM,SAAS,gBAAgB;AAC/B,KAAI,OAAO,UAAU,OACnB,MAAK,MAAM,UAAU,OAAO,WAAW;EACrC,MAAM,UAAU,YAAY,OAAO;AACnC,MAAI,CAAC,WAAW,QAAQ,EAAE;AACxB,UAAO,QAAQ,KAAK,GAAG,OAAO,kCAAkC;AAChE;;EAGF,MAAM,WAAW,YAAY,QAAQ,CAAC,QAAQ,SAAS;AACrD,OAAI,KAAK,WAAW,IAAI,CAAE,QAAO;GACjC,MAAM,OAAO,KAAK,SAAS,KAAK;AAChC,OAAI;AAAE,WAAO,SAAS,KAAK,CAAC,aAAa;WAAU;AAAE,WAAO;;IAC5D;AAEF,OAAK,MAAM,SAAS,UAAU;GAC5B,MAAM,YAAY,KAAK,SAAS,MAAM;GACtC,MAAM,YAAY,QAAQ,MAAM;GAChC,MAAM,eAAe,UAAU,UAAU;GAEzC,MAAM,WAAW,GACd,QAAQ,8CAA8C,CACtD,IAAI,UAAU;AAEjB,OAAI,UAAU;AACZ,WAAO;AACP,WAAO;AAEP,QAAI;AAAE,qBAAgB,WAAW,UAAU;YAAU;IAErD,MAAM,WAAW,KAAK,WAAW,QAAQ;AACzC,QAAI,WAAW,SAAS,EAAE;KACxB,MAAM,YAAY,YAAY,SAAS,CAAC,QAAQ,MAAM,EAAE,SAAS,MAAM,CAAC;AACxE,UAAK,MAAM,YAAY,WAAW;MAChC,MAAM,SAAS,qBAAqB,SAAS;AAC7C,UAAI,CAAC,OAAQ;AACb,aAAO;AACP,UAAI,cAAc,IAAI,SAAS,IAAI,OAAO,QAAQ,OAAO,MAAM,OAAO,MAAM,OAAO,OAAO,OAAO,SAAS,CACxG,QAAO;;;AAIb;;GAGF,MAAM,EAAE,IAAI,UAAU,cAAc,IAAI,WAAW,WAAW,aAAa;AAC3E,UAAO;AACP,OAAI,MAAO,QAAO;OACb,QAAO;AAEZ,OAAI;AAAE,oBAAgB,WAAW,UAAU;WAAU;GAErD,MAAM,WAAW,KAAK,WAAW,QAAQ;AACzC,OAAI,WAAW,SAAS,EAAE;IACxB,MAAM,YAAY,YAAY,SAAS,CAAC,QAAQ,MAAM,EAAE,SAAS,MAAM,CAAC;AACxE,SAAK,MAAM,YAAY,WAAW;KAChC,MAAM,SAAS,qBAAqB,SAAS;AAC7C,SAAI,CAAC,OAAQ;AACb,YAAO;AACP,SAAI,cAAc,IAAI,IAAI,OAAO,QAAQ,OAAO,MAAM,OAAO,MAAM,OAAO,OAAO,OAAO,SAAS,CAC/F,QAAO;;;;;AASnB,KAAI,OAAO,UAAU,QAAQ;EAE3B,MAAM,UAAU,mBADS,OAAO,UAAU,IAAI,YAAY,CAAC,OAAO,WAAW,CACzB;AAEpD,OAAK,MAAM,UAAU,SAAS;GAC5B,MAAM,gBAAgB,GACnB,QAAQ,0DAA0D,CAClE,IAAI,OAAO,KAAK;AAEnB,OAAI,CAAC,cAAe;AAEpB,OAAI,cAAc,cAAc,OAAO,aAAa;IAClD,MAAM,aAAa,UAAU,OAAO,YAAY;IAChD,MAAM,OAAO,KAAK,KAAK;IAEvB,MAAM,eAAe,GAClB,QAAQ,gDAAgD,CACxD,IAAI,WAAW;IAClB,MAAM,YAAY,GACf,QAAQ,8CAA8C,CACtD,IAAI,OAAO,YAAY;IAE1B,MAAM,cAAc,CAAC,gBAAgB,aAAa,OAAO,cAAc;IACvE,MAAM,WAAW,CAAC,aAAa,UAAU,OAAO,cAAc;AAE9D,QAAI,eAAe,SACjB,IAAG,QACD,kFACD,CAAC,IAAI,OAAO,aAAa,YAAY,MAAM,cAAc,GAAG;aACpD,SACT,IAAG,QACD,iEACD,CAAC,IAAI,OAAO,aAAa,MAAM,cAAc,GAAG;;;;AAMzD,QAAO;;;;;;;;;;AAeT,SAAgB,QAAQ,IAAc,OAA4B,EAAE,EAAQ;CAC1E,MAAM,SAAS,gBAAgB;AAC/B,KAAI,CAAC,KAAK,OAAO;AACf,UAAQ,IAAI,IAAI,mCAAmC,CAAC;AACpD,MAAI,OAAO,UAAU,OACnB,SAAQ,IAAI,IAAI,YAAY,OAAO,UAAU,OAAO,iBAAiB,OAAO,UAAU,KAAK,KAAK,GAAG,CAAC;AAEtG,UAAQ,IAAI,IAAI,+CAA+C,CAAC;;CAGlE,IAAI;AACJ,KAAI;AACF,WAAS,YAAY,GAAG;UACjB,GAAG;AACV,UAAQ,MAAM,IAAI,OAAO,EAAE,CAAC,CAAC;AAC7B,UAAQ,KAAK,EAAE;;AAGjB,KAAI,CAAC,KAAK,OAAO;AACf,UAAQ,IACN,GAAG,WAAW,KAAK,OAAO,OAAO,gBAAgB,CAAC,CAAC,aAAa,KAAK,OAAO,OAAO,gBAAgB,CAAC,CAAC,iBAAiB,CACvH;AACD,UAAQ,IAAI,IAAI,eAAe,OAAO,YAAY,QAAQ,OAAO,gBAAgB,UAAU,CAAC;AAC5F,UAAQ,IAAI,IAAI,eAAe,OAAO,YAAY,MAAM,CAAC;AAEzD,MAAI,OAAO,QAAQ,QAAQ;AACzB,WAAQ,KAAK;AACb,WAAQ,IAAI,KAAK,KAAK,OAAO,QAAQ,OAAO,+CAA+C,CAAC;AAC5F,QAAK,MAAM,KAAK,OAAO,QAAQ,MAAM,GAAG,GAAG,CACzC,SAAQ,IAAI,IAAI,OAAO,IAAI,CAAC;AAE9B,OAAI,OAAO,QAAQ,SAAS,GAC1B,SAAQ,IAAI,IAAI,eAAe,OAAO,QAAQ,SAAS,GAAG,OAAO,CAAC;;OAKtE,SAAQ,OAAO,MACb,mBAAmB,OAAO,gBAAgB,aAAa,OAAO,gBAAgB,aAC1E,OAAO,YAAY,GAAG,OAAO,YAAY,UAC9C;;;;;;;;;;;AC3YL,SAAgB,4BAA4B,QAIjC;AACT,QAAO;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;WAgCE,OAAO,YAAY;;;EAG5B,OAAO,eAAe;;;EAGtB,OAAO,OAAO;;;;;;;;;;;;;;;;;;;;;;ACpBhB,SAAgB,mBAAkC;CAChD,MAAM,aAAa;EACjB,KAAK,SAAS,EAAE,UAAU,OAAO,SAAS;EAC1C,KAAK,SAAS,EAAE,WAAW,SAAS,SAAS;EAC7C;EACA;EACD;AAED,MAAK,MAAM,aAAa,WACtB,KAAI;AACF,MAAI,WAAW,UAAU,CAAE,QAAO;SAC5B;AAEV,QAAO;;AAGT,MAAM,oBAA4C;CAChD,OAAO;CACP,QAAQ;CACR,MAAM;CACP;;;;;;;AAQD,eAAsB,YACpB,QACA,QAAqC,UACb;CACxB,MAAM,YAAY,kBAAkB;AACpC,KAAI,CAAC,WAAW;AACd,UAAQ,OAAO,MAAM,0CAA0C;AAC/D,SAAO;;CAGT,MAAM,EAAE,UAAU,MAAM,OAAO;AAE/B,QAAO,IAAI,SAAS,YAAY;EAC9B,IAAI,QAA8C;EAElD,MAAM,EAAE,mBAAmB,OAAO,GAAG,qBAAqB,QAAQ;EAClE,MAAM,QAAQ,MACZ,WACA;GAAC;GAAW;GAAO;GAAM;GAA2B,EACpD;GAAE,KAAK;GAAkB,OAAO;IAAC;IAAQ;IAAQ;IAAO;GAAE,CAC3D;EAED,IAAI,SAAS;EACb,IAAI,SAAS;AAEb,QAAM,OAAO,GAAG,SAAS,UAAkB;AAAE,aAAU,MAAM,UAAU;IAAI;AAC3E,QAAM,OAAO,GAAG,SAAS,UAAkB;AAAE,aAAU,MAAM,UAAU;IAAI;AAE3E,QAAM,GAAG,UAAU,QAAe;AAChC,OAAI,OAAO;AAAE,iBAAa,MAAM;AAAE,YAAQ;;AAC1C,WAAQ,OAAO,MAAM,mBAAmB,MAAM,gBAAgB,IAAI,QAAQ,IAAI;AAC9E,WAAQ,KAAK;IACb;AAEF,QAAM,GAAG,UAAU,SAAwB;AACzC,OAAI,OAAO;AAAE,iBAAa,MAAM;AAAE,YAAQ;;AAC1C,OAAI,SAAS,GAAG;AACd,YAAQ,OAAO,MACb,mBAAmB,MAAM,UAAU,KAAK,IAAI,OAAO,MAAM,GAAG,IAAI,CAAC,IAClE;AACD,YAAQ,KAAK;SAEb,SAAQ,OAAO,MAAM,IAAI,KAAK;IAEhC;AAEF,UAAQ,iBAAiB;AACvB,WAAQ,OAAO,MAAM,mBAAmB,MAAM,iCAAiC;AAC/E,SAAM,KAAK,UAAU;AACrB,WAAQ,KAAK;KACZ,kBAAkB,UAAU,KAAQ;AAEvC,QAAM,MAAM,MAAM,OAAO;AACzB,QAAM,MAAM,KAAK;GACjB;;;;;;;;;;;AAmCJ,eAAsB,uBACpB,MACA,QAC+B;CAC/B,MAAM,QAA8B;EAAE,WAAW;EAAG,OAAO;EAAG,YAAY;EAAG;CAQ7E,MAAM,aAAa,MAAM,YANV,4BAA4B;EACzC,gBAAgB,OAAO;EACvB,aAAa,OAAO;EACpB,QAAQ,OAAO,UAAU;EAC1B,CAAC,EAE2C,OAAO,SAAS,SAAS;AACtE,KAAI,CAAC,WAAY,QAAO;CAGxB,MAAM,UAAU,WACb,QAAQ,gBAAgB,GAAG,CAC3B,QAAQ,YAAY,GAAG,CACvB,QAAQ,YAAY,GAAG,CACvB,MAAM;CAQT,IAAI;AACJ,KAAI;EACF,MAAM,SAAS,KAAK,MAAM,QAAQ;AAElC,MAAI,MAAM,QAAQ,OAAO,CAEvB,WAAU;WACD,UAAU,OAAO,WAAW,YAAY,MAAM,QAAQ,OAAO,UAAU,EAAE;GAElF,MAAM,SAAS;AAGf,OAAI,OAAO,gBAAgB,MAAM,QAAQ,OAAO,SAAS,EAAE;IACzD,MAAM,WAAW,OAAO,YAAY;AACpC,SAAK,MAAM,UAAU,OAAO,UAAU;AACpC,SAAI,CAAC,OAAO,KAAM;AAClB,SAAI;AACF,qBAAe,OAAO,cAAc;OAClC,MAAM,OAAO;OACb,MAAM,OAAO,QAAQ;OACrB,aAAa,OAAO;OACpB;OACD,CAAC;cACK,WAAW;AAClB,cAAQ,OAAO,MAAM,wCAAwC,OAAO,KAAK,KAAK,UAAU,IAAI;;;;AAKlG,aAAU,OAAO,UAAU,KAAK,OAAoB;IAClD,SAAS,EAAE;IACX,WAAW,EAAE;IACb,QAAQ,EAAE;IACX,EAAE;SACE;AACL,WAAQ,OAAO,MAAM,mFAAmF;AACxG,UAAO;;UAEF,GAAG;AACV,UAAQ,OAAO,MAAM,sCAAsC,EAAE,IAAI;AACjE,SAAO;;AAGT,KAAI,CAAC,MAAM,QAAQ,QAAQ,CAAE,QAAO;AACpC,OAAM,YAAY,QAAQ;AAE1B,MAAK,MAAM,KAAK,SAAS;AACvB,MAAI,CAAC,EAAE,WAAW,CAAC,EAAE,aAAa,CAAC,EAAE,OAAQ;AAE7C,MAAI;GACF,MAAM,WAAW,MAAM,QAAQ,MAAM;IACnC,SAAS,EAAE;IACX,WAAW,EAAE;IACb,YAAY,OAAO,aAAa;IACjC,CAAC;AAIF,OADqB,SAAS,MAAM,MAAM,EAAE,WAAW,EAAE,UAAU,CAAC,EAAE,SAAS,CAC7D;GAGlB,MAAM,aAAa,SAAS,MAAM,MAAM,EAAE,WAAW,EAAE,UAAU,CAAC,EAAE,SAAS;AAC7E,OAAI,YAAY;AACd,UAAM,aAAa,MAAM,WAAW,GAAG;AACvC,UAAM;;AAGR,SAAM,MAAM,MAAM;IAChB,SAAS,EAAE;IACX,WAAW,EAAE;IACb,QAAQ,EAAE;IACV,YAAY,OAAO,aAAa;IAChC,gBAAgB,OAAO;IACvB,YAAY;IACb,CAAC;AACF,SAAM;WACC,WAAW;AAClB,WAAQ,OAAO,MAAM,gCAAgC,EAAE,QAAQ,KAAK,UAAU,IAAI;;;AAItF,QAAO"}

package/dist/{sync-CdHSL9Kc.mjs → sync-CmBKOL3K.mjs}

@@ -1,4 +1,4 @@
-import { a as parseSessionTitleChunk, c as yieldToEventLoop, f as sha256File, i as isPathTooBroadForContentScan, l as chunkMarkdown, n as chunkId, o as walkContentFiles, r as detectTier, s as walkMdFiles, t as INDEX_YIELD_EVERY } from "./helpers-OCVFgprQ.mjs";
+import { a as parseSessionTitleChunk, c as yieldToEventLoop, f as sha256File, i as isPathTooBroadForContentScan, l as chunkMarkdown, n as chunkId, o as walkContentFiles, r as detectTier, s as walkMdFiles, t as INDEX_YIELD_EVERY } from "./helpers-crDEr6S2.mjs";
 import { existsSync, readFileSync, statSync } from "node:fs";
 import { basename, join, relative } from "node:path";
 
@@ -307,4 +307,4 @@ async function embedChunks(db, projectId, batchSize = 50, onProgress) {
 
 //#endregion
 export { indexProject as i, indexAll as n, indexFile as r, embedChunks as t };
-//# sourceMappingURL=sync-CdHSL9Kc.mjs.map
+//# sourceMappingURL=sync-CmBKOL3K.mjs.map

package/dist/{sync-CdHSL9Kc.mjs.map → sync-CmBKOL3K.mjs.map}

@@ -1 +1 @@
-{"version":3,"file":"sync-CdHSL9Kc.mjs","names":[],"sources":["../src/memory/indexer/sync.ts"],"sourcesContent":["/**\n * Synchronous (SQLite) indexer for the PAI federation memory engine.\n *\n * Scans project memory/ and Notes/ directories, chunks markdown files, and\n * inserts the resulting chunks into federation.db for BM25 search.\n *\n * Change detection: files whose SHA-256 hash has not changed since the last\n * index run are skipped, keeping incremental re-indexing fast.\n *\n * Uses raw better-sqlite3 Database directly for maximum SQLite performance\n * (synchronous transactions, no serialisation overhead).\n */\n\nimport { readFileSync, statSync, existsSync } from \"node:fs\";\nimport { join, relative, basename } from \"node:path\";\nimport type { Database } from \"better-sqlite3\";\nimport { chunkMarkdown } from \"../chunker.js\";\nimport {\n  sha256File,\n  chunkId,\n  detectTier,\n  walkMdFiles,\n  walkContentFiles,\n  isPathTooBroadForContentScan,\n  parseSessionTitleChunk,\n  yieldToEventLoop,\n  INDEX_YIELD_EVERY,\n} from \"./helpers.js\";\nimport type { IndexResult, EmbedResult } from \"./types.js\";\n\nexport type { IndexResult, EmbedResult };\n\n// Re-export detectTier for backward-compatibility (consumers import it from indexer.js)\nexport { detectTier };\n\n// ---------------------------------------------------------------------------\n// Single-file indexing\n// ---------------------------------------------------------------------------\n\n/**\n * Index a single file into the federation database.\n *\n * @returns true if the file was re-indexed (changed or new), false if skipped.\n */\nexport function indexFile(\n  db: Database,\n  projectId: number,\n  rootPath: string,\n  relativePath: string,\n  source: string,\n  tier: string,\n): boolean {\n  const absPath = join(rootPath, relativePath);\n\n  // Read file content\n  let content: string;\n  let stat: ReturnType<typeof statSync>;\n  try {\n    content = readFileSync(absPath, \"utf8\");\n    stat = statSync(absPath);\n  } catch {\n    // File unreadable or missing — skip silently\n    return false;\n  }\n\n  const hash = sha256File(content);\n  const mtime = Math.floor(stat.mtimeMs);\n  const size = stat.size;\n\n  // Check if the file has changed since last index\n  const existing = db\n    .prepare(\n      \"SELECT hash FROM memory_files WHERE project_id = ? AND path = ?\",\n    )\n    .get(projectId, relativePath) as { hash: string } | undefined;\n\n  if (existing?.hash === hash) {\n    // Unchanged — skip\n    return false;\n  }\n\n  // Delete old chunks for this file from both tables\n  const oldChunkIds = db\n    .prepare(\n      \"SELECT id FROM memory_chunks WHERE project_id = ? AND path = ?\",\n    )\n    .all(projectId, relativePath) as Array<{ id: string }>;\n\n  const deleteFts = db.prepare(\"DELETE FROM memory_fts WHERE id = ?\");\n  const deleteChunk = db.prepare(\n    \"DELETE FROM memory_chunks WHERE project_id = ? AND path = ?\",\n  );\n\n  db.transaction(() => {\n    for (const row of oldChunkIds) {\n      deleteFts.run(row.id);\n    }\n    deleteChunk.run(projectId, relativePath);\n  })();\n\n  // Chunk the new content\n  const chunks = chunkMarkdown(content);\n\n  // Insert new chunks into memory_chunks and memory_fts\n  const insertChunk = db.prepare(`\n    INSERT INTO memory_chunks (id, project_id, source, tier, path, start_line, end_line, hash, text, updated_at)\n    VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)\n  `);\n\n  const insertFts = db.prepare(`\n    INSERT INTO memory_fts (text, id, project_id, path, source, tier, start_line, end_line)\n    VALUES (?, ?, ?, ?, ?, ?, ?, ?)\n  `);\n\n  const upsertFile = db.prepare(`\n    INSERT INTO memory_files (project_id, path, source, tier, hash, mtime, size)\n    VALUES (?, ?, ?, ?, ?, ?, ?)\n    ON CONFLICT(project_id, path) DO UPDATE SET\n      source = excluded.source,\n      tier   = excluded.tier,\n      hash   = excluded.hash,\n      mtime  = excluded.mtime,\n      size   = excluded.size\n  `);\n\n  const updatedAt = Date.now();\n\n  db.transaction(() => {\n    for (let i = 0; i < chunks.length; i++) {\n      const chunk = chunks[i]!;\n      const id = chunkId(projectId, relativePath, i, chunk.startLine, chunk.endLine);\n      insertChunk.run(\n        id,\n        projectId,\n        source,\n        tier,\n        relativePath,\n        chunk.startLine,\n        chunk.endLine,\n        chunk.hash,\n        chunk.text,\n        updatedAt,\n      );\n      insertFts.run(\n        chunk.text,\n        id,\n        projectId,\n        relativePath,\n        source,\n        tier,\n        chunk.startLine,\n        chunk.endLine,\n      );\n    }\n    upsertFile.run(projectId, relativePath, source, tier, hash, mtime, size);\n  })();\n\n  return true;\n}\n\n// ---------------------------------------------------------------------------\n// Project-level indexing\n// ---------------------------------------------------------------------------\n\n/**\n * Index all memory, Notes, and content files for a single registered project.\n *\n * Scans:\n *  - {rootPath}/MEMORY.md    → source='memory', tier='evergreen'\n *  - {rootPath}/memory/      → source='memory', tier from detectTier()\n *  - {rootPath}/Notes/       → source='notes',  tier='session'\n *  - {rootPath}/**\\/*.md    → source='content', tier='topic'  (all other .md files, recursive)\n *  - {claudeNotesDir}/       → source='notes',  tier='session'  (if set and different)\n */\nexport async function indexProject(\n  db: Database,\n  projectId: number,\n  rootPath: string,\n  claudeNotesDir?: string | null,\n): Promise<IndexResult> {\n  const result: IndexResult = {\n    filesProcessed: 0,\n    chunksCreated: 0,\n    filesSkipped: 0,\n  };\n\n  const filesToIndex: Array<{ absPath: string; rootBase: string; source: string; tier: string }> = [];\n\n  // Root-level MEMORY.md\n  const rootMemoryMd = join(rootPath, \"MEMORY.md\");\n  if (existsSync(rootMemoryMd)) {\n    filesToIndex.push({ absPath: rootMemoryMd, rootBase: rootPath, source: \"memory\", tier: \"evergreen\" });\n  }\n\n  // memory/ directory\n  const memoryDir = join(rootPath, \"memory\");\n  for (const absPath of walkMdFiles(memoryDir)) {\n    const relPath = relative(rootPath, absPath);\n    const tier = detectTier(relPath);\n    filesToIndex.push({ absPath, rootBase: rootPath, source: \"memory\", tier });\n  }\n\n  // {rootPath}/Notes/ directory\n  const notesDir = join(rootPath, \"Notes\");\n  for (const absPath of walkMdFiles(notesDir)) {\n    filesToIndex.push({ absPath, rootBase: rootPath, source: \"notes\", tier: \"session\" });\n  }\n\n  // Synthetic session-title chunks for Notes files with the standard filename format:\n  // \"NNNN - YYYY-MM-DD - Descriptive Title.md\"\n  {\n    const titleInsertChunk = db.prepare(`\n      INSERT OR IGNORE INTO memory_chunks (id, project_id, source, tier, path, start_line, end_line, hash, text, updated_at)\n      VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)\n    `);\n    const titleInsertFts = db.prepare(`\n      INSERT OR IGNORE INTO memory_fts (text, id, project_id, path, source, tier, start_line, end_line)\n      VALUES (?, ?, ?, ?, ?, ?, ?, ?)\n    `);\n    const updatedAt = Date.now();\n    for (const absPath of walkMdFiles(notesDir)) {\n      const fileName = basename(absPath);\n      const text = parseSessionTitleChunk(fileName);\n      if (!text) continue;\n      const relPath = relative(rootPath, absPath);\n      const syntheticPath = `${relPath}::title`;\n      const id = chunkId(projectId, syntheticPath, 0, 0, 0);\n      const hash = sha256File(text);\n      db.transaction(() => {\n        titleInsertChunk.run(id, projectId, \"notes\", \"session\", syntheticPath, 0, 0, hash, text, updatedAt);\n        titleInsertFts.run(text, id, projectId, syntheticPath, \"notes\", \"session\", 0, 0);\n      })();\n    }\n  }\n\n  // {rootPath}/**/*.md — all other markdown content\n  if (!isPathTooBroadForContentScan(rootPath)) {\n    for (const absPath of walkContentFiles(rootPath)) {\n      filesToIndex.push({ absPath, rootBase: rootPath, source: \"content\", tier: \"topic\" });\n    }\n  }\n\n  // Claude Code session notes directory (~/.claude/projects/{encoded}/Notes/)\n  if (claudeNotesDir && claudeNotesDir !== notesDir) {\n    for (const absPath of walkMdFiles(claudeNotesDir)) {\n      filesToIndex.push({ absPath, rootBase: claudeNotesDir, source: \"notes\", tier: \"session\" });\n    }\n\n    // Synthetic title chunks for claude notes dir\n    {\n      const updatedAt = Date.now();\n      const titleInsertChunk2 = db.prepare(`\n        INSERT OR IGNORE INTO memory_chunks (id, project_id, source, tier, path, start_line, end_line, hash, text, updated_at)\n        VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)\n      `);\n      const titleInsertFts2 = db.prepare(`\n        INSERT OR IGNORE INTO memory_fts (text, id, project_id, path, source, tier, start_line, end_line)\n        VALUES (?, ?, ?, ?, ?, ?, ?, ?)\n      `);\n      for (const absPath of walkMdFiles(claudeNotesDir)) {\n        const fileName = basename(absPath);\n        const text = parseSessionTitleChunk(fileName);\n        if (!text) continue;\n        const relPath = relative(claudeNotesDir, absPath);\n        const syntheticPath = `${relPath}::title`;\n        const id = chunkId(projectId, syntheticPath, 0, 0, 0);\n        const hash = sha256File(text);\n        db.transaction(() => {\n          titleInsertChunk2.run(id, projectId, \"notes\", \"session\", syntheticPath, 0, 0, hash, text, updatedAt);\n          titleInsertFts2.run(text, id, projectId, syntheticPath, \"notes\", \"session\", 0, 0);\n        })();\n      }\n    }\n\n    // Derive the sibling memory/ directory: .../Notes/ → .../memory/\n    if (claudeNotesDir.endsWith(\"/Notes\")) {\n      const claudeProjectDir = claudeNotesDir.slice(0, -\"/Notes\".length);\n      const claudeMemoryDir = join(claudeProjectDir, \"memory\");\n\n      const claudeMemoryMd = join(claudeProjectDir, \"MEMORY.md\");\n      if (existsSync(claudeMemoryMd)) {\n        filesToIndex.push({\n          absPath: claudeMemoryMd,\n          rootBase: claudeProjectDir,\n          source: \"memory\",\n          tier: \"evergreen\",\n        });\n      }\n\n      for (const absPath of walkMdFiles(claudeMemoryDir)) {\n        const relPath = relative(claudeProjectDir, absPath);\n        const tier = detectTier(relPath);\n        filesToIndex.push({ absPath, rootBase: claudeProjectDir, source: \"memory\", tier });\n      }\n    }\n  }\n\n  // Yield after collection phase before processing\n  await yieldToEventLoop();\n\n  let filesSinceYield = 0;\n\n  for (const { absPath, rootBase, source, tier } of filesToIndex) {\n    if (filesSinceYield >= INDEX_YIELD_EVERY) {\n      await yieldToEventLoop();\n      filesSinceYield = 0;\n    }\n    filesSinceYield++;\n\n    const relPath = relative(rootBase, absPath);\n    const changed = indexFile(db, projectId, rootBase, relPath, source, tier);\n\n    if (changed) {\n      const count = db\n        .prepare(\n          \"SELECT COUNT(*) as n FROM memory_chunks WHERE project_id = ? AND path = ?\",\n        )\n        .get(projectId, relPath) as { n: number };\n\n      result.filesProcessed++;\n      result.chunksCreated += count.n;\n    } else {\n      result.filesSkipped++;\n    }\n  }\n\n  // Prune stale paths: remove DB entries for files that no longer exist on disk.\n  const livePaths = new Set<string>();\n  for (const { absPath, rootBase } of filesToIndex) {\n    livePaths.add(relative(rootBase, absPath));\n  }\n\n  const dbChunkPaths = db\n    .prepare(\"SELECT DISTINCT path FROM memory_chunks WHERE project_id = ?\")\n    .all(projectId) as Array<{ path: string }>;\n\n  const stalePaths: string[] = [];\n  for (const row of dbChunkPaths) {\n    const basePath = row.path.endsWith(\"::title\")\n      ? row.path.slice(0, -\"::title\".length)\n      : row.path;\n    if (!livePaths.has(basePath)) {\n      stalePaths.push(row.path);\n    }\n  }\n\n  if (stalePaths.length > 0) {\n    const deleteChunksFts = db.prepare(\"DELETE FROM memory_fts WHERE id = ?\");\n    const deleteChunks = db.prepare(\n      \"DELETE FROM memory_chunks WHERE project_id = ? AND path = ?\",\n    );\n    const deleteFile = db.prepare(\n      \"DELETE FROM memory_files WHERE project_id = ? AND path = ?\",\n    );\n\n    db.transaction(() => {\n      for (const stalePath of stalePaths) {\n        const chunkIds = db\n          .prepare(\"SELECT id FROM memory_chunks WHERE project_id = ? AND path = ?\")\n          .all(projectId, stalePath) as Array<{ id: string }>;\n        for (const { id } of chunkIds) {\n          deleteChunksFts.run(id);\n        }\n        deleteChunks.run(projectId, stalePath);\n        deleteFile.run(projectId, stalePath);\n      }\n    })();\n  }\n\n  return result;\n}\n\n// ---------------------------------------------------------------------------\n// Global indexing (all registered projects)\n// ---------------------------------------------------------------------------\n\n/**\n * Index all active projects registered in the registry DB.\n *\n * Async: yields to the event loop between each project so that the daemon's\n * Unix socket server can process IPC requests (e.g. status) while indexing.\n */\nexport async function indexAll(\n  db: Database,\n  registryDb: Database,\n): Promise<{ projects: number; result: IndexResult }> {\n  const projects = registryDb\n    .prepare(\"SELECT id, root_path, claude_notes_dir FROM projects WHERE status = 'active'\")\n    .all() as Array<{ id: number; root_path: string; claude_notes_dir: string | null }>;\n\n  const totals: IndexResult = {\n    filesProcessed: 0,\n    chunksCreated: 0,\n    filesSkipped: 0,\n  };\n\n  for (const project of projects) {\n    await yieldToEventLoop();\n    const r = await indexProject(db, project.id, project.root_path, project.claude_notes_dir);\n    totals.filesProcessed += r.filesProcessed;\n    totals.chunksCreated += r.chunksCreated;\n    totals.filesSkipped += r.filesSkipped;\n  }\n\n  return { projects: projects.length, result: totals };\n}\n\n// ---------------------------------------------------------------------------\n// Embedding generation\n// ---------------------------------------------------------------------------\n\n/**\n * Generate and store embeddings for chunks that do not yet have one.\n *\n * Because better-sqlite3 is synchronous but the embedding pipeline is async,\n * we fetch all unembedded chunk texts first, generate embeddings in batches,\n * and then write them back in a transaction.\n *\n * @param db         Open federation database.\n * @param projectId  Optional — restrict to a specific project.\n * @param batchSize  Number of chunks to embed per round. Default 50.\n * @param onProgress Optional callback called after each batch with running totals.\n */\nexport async function embedChunks(\n  db: Database,\n  projectId?: number,\n  batchSize = 50,\n  onProgress?: (embedded: number, total: number) => void,\n): Promise<EmbedResult> {\n  // Dynamic import — keeps the heavy ML runtime out of the module load path\n  const { generateEmbedding, serializeEmbedding } = await import(\"../embeddings.js\");\n\n  const conditions = [\"embedding IS NULL\"];\n  const params: (string | number)[] = [];\n\n  if (projectId !== undefined) {\n    conditions.push(\"project_id = ?\");\n    params.push(projectId);\n  }\n\n  const where = \"WHERE \" + conditions.join(\" AND \");\n\n  const rows = db\n    .prepare(`SELECT id, text FROM memory_chunks ${where} ORDER BY id`)\n    .all(...params) as Array<{ id: string; text: string }>;\n\n  if (rows.length === 0) {\n    return { chunksEmbedded: 0, chunksSkipped: 0 };\n  }\n\n  const updateStmt = db.prepare(\n    \"UPDATE memory_chunks SET embedding = ? WHERE id = ?\",\n  );\n\n  let embedded = 0;\n  const total = rows.length;\n\n  for (let i = 0; i < rows.length; i += batchSize) {\n    const batch = rows.slice(i, i + batchSize);\n\n    // Generate embeddings for the batch (async — must happen OUTSIDE transaction)\n    const embeddings: Array<{ id: string; blob: Buffer }> = [];\n    for (const row of batch) {\n      const vec = await generateEmbedding(row.text);\n      const blob = serializeEmbedding(vec);\n      embeddings.push({ id: row.id, blob });\n    }\n\n    // Write the batch in a single transaction\n    db.transaction(() => {\n      for (const { id, blob } of embeddings) {\n        updateStmt.run(blob, id);\n      }\n    })();\n\n    embedded += embeddings.length;\n    onProgress?.(embedded, total);\n  }\n\n  return { chunksEmbedded: embedded, chunksSkipped: 0 };\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;AA4CA,SAAgB,UACd,IACA,WACA,UACA,cACA,QACA,MACS;CACT,MAAM,UAAU,KAAK,UAAU,aAAa;CAG5C,IAAI;CACJ,IAAI;AACJ,KAAI;AACF,YAAU,aAAa,SAAS,OAAO;AACvC,SAAO,SAAS,QAAQ;SAClB;AAEN,SAAO;;CAGT,MAAM,OAAO,WAAW,QAAQ;CAChC,MAAM,QAAQ,KAAK,MAAM,KAAK,QAAQ;CACtC,MAAM,OAAO,KAAK;AASlB,KANiB,GACd,QACC,kEACD,CACA,IAAI,WAAW,aAAa,EAEjB,SAAS,KAErB,QAAO;CAIT,MAAM,cAAc,GACjB,QACC,iEACD,CACA,IAAI,WAAW,aAAa;CAE/B,MAAM,YAAY,GAAG,QAAQ,sCAAsC;CACnE,MAAM,cAAc,GAAG,QACrB,8DACD;AAED,IAAG,kBAAkB;AACnB,OAAK,MAAM,OAAO,YAChB,WAAU,IAAI,IAAI,GAAG;AAEvB,cAAY,IAAI,WAAW,aAAa;GACxC,EAAE;CAGJ,MAAM,SAAS,cAAc,QAAQ;CAGrC,MAAM,cAAc,GAAG,QAAQ;;;IAG7B;CAEF,MAAM,YAAY,GAAG,QAAQ;;;IAG3B;CAEF,MAAM,aAAa,GAAG,QAAQ;;;;;;;;;IAS5B;CAEF,MAAM,YAAY,KAAK,KAAK;AAE5B,IAAG,kBAAkB;AACnB,OAAK,IAAI,IAAI,GAAG,IAAI,OAAO,QAAQ,KAAK;GACtC,MAAM,QAAQ,OAAO;GACrB,MAAM,KAAK,QAAQ,WAAW,cAAc,GAAG,MAAM,WAAW,MAAM,QAAQ;AAC9E,eAAY,IACV,IACA,WACA,QACA,MACA,cACA,MAAM,WACN,MAAM,SACN,MAAM,MACN,MAAM,MACN,UACD;AACD,aAAU,IACR,MAAM,MACN,IACA,WACA,cACA,QACA,MACA,MAAM,WACN,MAAM,QACP;;AAEH,aAAW,IAAI,WAAW,cAAc,QAAQ,MAAM,MAAM,OAAO,KAAK;GACxE,EAAE;AAEJ,QAAO;;;;;;;;;;;;AAiBT,eAAsB,aACpB,IACA,WACA,UACA,gBACsB;CACtB,MAAM,SAAsB;EAC1B,gBAAgB;EAChB,eAAe;EACf,cAAc;EACf;CAED,MAAM,eAA2F,EAAE;CAGnG,MAAM,eAAe,KAAK,UAAU,YAAY;AAChD,KAAI,WAAW,aAAa,CAC1B,cAAa,KAAK;EAAE,SAAS;EAAc,UAAU;EAAU,QAAQ;EAAU,MAAM;EAAa,CAAC;CAIvG,MAAM,YAAY,KAAK,UAAU,SAAS;AAC1C,MAAK,MAAM,WAAW,YAAY,UAAU,EAAE;EAE5C,MAAM,OAAO,WADG,SAAS,UAAU,QAAQ,CACX;AAChC,eAAa,KAAK;GAAE;GAAS,UAAU;GAAU,QAAQ;GAAU;GAAM,CAAC;;CAI5E,MAAM,WAAW,KAAK,UAAU,QAAQ;AACxC,MAAK,MAAM,WAAW,YAAY,SAAS,CACzC,cAAa,KAAK;EAAE;EAAS,UAAU;EAAU,QAAQ;EAAS,MAAM;EAAW,CAAC;CAKtF;EACE,MAAM,mBAAmB,GAAG,QAAQ;;;MAGlC;EACF,MAAM,iBAAiB,GAAG,QAAQ;;;MAGhC;EACF,MAAM,YAAY,KAAK,KAAK;AAC5B,OAAK,MAAM,WAAW,YAAY,SAAS,EAAE;GAE3C,MAAM,OAAO,uBADI,SAAS,QAAQ,CACW;AAC7C,OAAI,CAAC,KAAM;GAEX,MAAM,gBAAgB,GADN,SAAS,UAAU,QAAQ,CACV;GACjC,MAAM,KAAK,QAAQ,WAAW,eAAe,GAAG,GAAG,EAAE;GACrD,MAAM,OAAO,WAAW,KAAK;AAC7B,MAAG,kBAAkB;AACnB,qBAAiB,IAAI,IAAI,WAAW,SAAS,WAAW,eAAe,GAAG,GAAG,MAAM,MAAM,UAAU;AACnG,mBAAe,IAAI,MAAM,IAAI,WAAW,eAAe,SAAS,WAAW,GAAG,EAAE;KAChF,EAAE;;;AAKR,KAAI,CAAC,6BAA6B,SAAS,CACzC,MAAK,MAAM,WAAW,iBAAiB,SAAS,CAC9C,cAAa,KAAK;EAAE;EAAS,UAAU;EAAU,QAAQ;EAAW,MAAM;EAAS,CAAC;AAKxF,KAAI,kBAAkB,mBAAmB,UAAU;AACjD,OAAK,MAAM,WAAW,YAAY,eAAe,CAC/C,cAAa,KAAK;GAAE;GAAS,UAAU;GAAgB,QAAQ;GAAS,MAAM;GAAW,CAAC;EAI5F;GACE,MAAM,YAAY,KAAK,KAAK;GAC5B,MAAM,oBAAoB,GAAG,QAAQ;;;QAGnC;GACF,MAAM,kBAAkB,GAAG,QAAQ;;;QAGjC;AACF,QAAK,MAAM,WAAW,YAAY,eAAe,EAAE;IAEjD,MAAM,OAAO,uBADI,SAAS,QAAQ,CACW;AAC7C,QAAI,CAAC,KAAM;IAEX,MAAM,gBAAgB,GADN,SAAS,gBAAgB,QAAQ,CAChB;IACjC,MAAM,KAAK,QAAQ,WAAW,eAAe,GAAG,GAAG,EAAE;IACrD,MAAM,OAAO,WAAW,KAAK;AAC7B,OAAG,kBAAkB;AACnB,uBAAkB,IAAI,IAAI,WAAW,SAAS,WAAW,eAAe,GAAG,GAAG,MAAM,MAAM,UAAU;AACpG,qBAAgB,IAAI,MAAM,IAAI,WAAW,eAAe,SAAS,WAAW,GAAG,EAAE;MACjF,EAAE;;;AAKR,MAAI,eAAe,SAAS,SAAS,EAAE;GACrC,MAAM,mBAAmB,eAAe,MAAM,GAAG,GAAiB;GAClE,MAAM,kBAAkB,KAAK,kBAAkB,SAAS;GAExD,MAAM,iBAAiB,KAAK,kBAAkB,YAAY;AAC1D,OAAI,WAAW,eAAe,CAC5B,cAAa,KAAK;IAChB,SAAS;IACT,UAAU;IACV,QAAQ;IACR,MAAM;IACP,CAAC;AAGJ,QAAK,MAAM,WAAW,YAAY,gBAAgB,EAAE;IAElD,MAAM,OAAO,WADG,SAAS,kBAAkB,QAAQ,CACnB;AAChC,iBAAa,KAAK;KAAE;KAAS,UAAU;KAAkB,QAAQ;KAAU;KAAM,CAAC;;;;AAMxF,OAAM,kBAAkB;CAExB,IAAI,kBAAkB;AAEtB,MAAK,MAAM,EAAE,SAAS,UAAU,QAAQ,UAAU,cAAc;AAC9D,MAAI,mBAAmB,mBAAmB;AACxC,SAAM,kBAAkB;AACxB,qBAAkB;;AAEpB;EAEA,MAAM,UAAU,SAAS,UAAU,QAAQ;AAG3C,MAFgB,UAAU,IAAI,WAAW,UAAU,SAAS,QAAQ,KAAK,EAE5D;GACX,MAAM,QAAQ,GACX,QACC,4EACD,CACA,IAAI,WAAW,QAAQ;AAE1B,UAAO;AACP,UAAO,iBAAiB,MAAM;QAE9B,QAAO;;CAKX,MAAM,4BAAY,IAAI,KAAa;AACnC,MAAK,MAAM,EAAE,SAAS,cAAc,aAClC,WAAU,IAAI,SAAS,UAAU,QAAQ,CAAC;CAG5C,MAAM,eAAe,GAClB,QAAQ,+DAA+D,CACvE,IAAI,UAAU;CAEjB,MAAM,aAAuB,EAAE;AAC/B,MAAK,MAAM,OAAO,cAAc;EAC9B,MAAM,WAAW,IAAI,KAAK,SAAS,UAAU,GACzC,IAAI,KAAK,MAAM,GAAG,GAAkB,GACpC,IAAI;AACR,MAAI,CAAC,UAAU,IAAI,SAAS,CAC1B,YAAW,KAAK,IAAI,KAAK;;AAI7B,KAAI,WAAW,SAAS,GAAG;EACzB,MAAM,kBAAkB,GAAG,QAAQ,sCAAsC;EACzE,MAAM,eAAe,GAAG,QACtB,8DACD;EACD,MAAM,aAAa,GAAG,QACpB,6DACD;AAED,KAAG,kBAAkB;AACnB,QAAK,MAAM,aAAa,YAAY;IAClC,MAAM,WAAW,GACd,QAAQ,iEAAiE,CACzE,IAAI,WAAW,UAAU;AAC5B,SAAK,MAAM,EAAE,QAAQ,SACnB,iBAAgB,IAAI,GAAG;AAEzB,iBAAa,IAAI,WAAW,UAAU;AACtC,eAAW,IAAI,WAAW,UAAU;;IAEtC,EAAE;;AAGN,QAAO;;;;;;;;AAaT,eAAsB,SACpB,IACA,YACoD;CACpD,MAAM,WAAW,WACd,QAAQ,+EAA+E,CACvF,KAAK;CAER,MAAM,SAAsB;EAC1B,gBAAgB;EAChB,eAAe;EACf,cAAc;EACf;AAED,MAAK,MAAM,WAAW,UAAU;AAC9B,QAAM,kBAAkB;EACxB,MAAM,IAAI,MAAM,aAAa,IAAI,QAAQ,IAAI,QAAQ,WAAW,QAAQ,iBAAiB;AACzF,SAAO,kBAAkB,EAAE;AAC3B,SAAO,iBAAiB,EAAE;AAC1B,SAAO,gBAAgB,EAAE;;AAG3B,QAAO;EAAE,UAAU,SAAS;EAAQ,QAAQ;EAAQ;;;;;;;;;;;;;;AAmBtD,eAAsB,YACpB,IACA,WACA,YAAY,IACZ,YACsB;CAEtB,MAAM,EAAE,mBAAmB,uBAAuB,MAAM,OAAO;CAE/D,MAAM,aAAa,CAAC,oBAAoB;CACxC,MAAM,SAA8B,EAAE;AAEtC,KAAI,cAAc,QAAW;AAC3B,aAAW,KAAK,iBAAiB;AACjC,SAAO,KAAK,UAAU;;CAGxB,MAAM,QAAQ,WAAW,WAAW,KAAK,QAAQ;CAEjD,MAAM,OAAO,GACV,QAAQ,sCAAsC,MAAM,cAAc,CAClE,IAAI,GAAG,OAAO;AAEjB,KAAI,KAAK,WAAW,EAClB,QAAO;EAAE,gBAAgB;EAAG,eAAe;EAAG;CAGhD,MAAM,aAAa,GAAG,QACpB,sDACD;CAED,IAAI,WAAW;CACf,MAAM,QAAQ,KAAK;AAEnB,MAAK,IAAI,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK,WAAW;EAC/C,MAAM,QAAQ,KAAK,MAAM,GAAG,IAAI,UAAU;EAG1C,MAAM,aAAkD,EAAE;AAC1D,OAAK,MAAM,OAAO,OAAO;GAEvB,MAAM,OAAO,mBADD,MAAM,kBAAkB,IAAI,KAAK,CACT;AACpC,cAAW,KAAK;IAAE,IAAI,IAAI;IAAI;IAAM,CAAC;;AAIvC,KAAG,kBAAkB;AACnB,QAAK,MAAM,EAAE,IAAI,UAAU,WACzB,YAAW,IAAI,MAAM,GAAG;IAE1B,EAAE;AAEJ,cAAY,WAAW;AACvB,eAAa,UAAU,MAAM;;AAG/B,QAAO;EAAE,gBAAgB;EAAU,eAAe;EAAG"}
+{"version":3,"file":"sync-CmBKOL3K.mjs","names":[],"sources":["../src/memory/indexer/sync.ts"],"sourcesContent":["/**\n * Synchronous (SQLite) indexer for the PAI federation memory engine.\n *\n * Scans project memory/ and Notes/ directories, chunks markdown files, and\n * inserts the resulting chunks into federation.db for BM25 search.\n *\n * Change detection: files whose SHA-256 hash has not changed since the last\n * index run are skipped, keeping incremental re-indexing fast.\n *\n * Uses raw better-sqlite3 Database directly for maximum SQLite performance\n * (synchronous transactions, no serialisation overhead).\n */\n\nimport { readFileSync, statSync, existsSync } from \"node:fs\";\nimport { join, relative, basename } from \"node:path\";\nimport type { Database } from \"better-sqlite3\";\nimport { chunkMarkdown } from \"../chunker.js\";\nimport {\n  sha256File,\n  chunkId,\n  detectTier,\n  walkMdFiles,\n  walkContentFiles,\n  isPathTooBroadForContentScan,\n  parseSessionTitleChunk,\n  yieldToEventLoop,\n  INDEX_YIELD_EVERY,\n} from \"./helpers.js\";\nimport type { IndexResult, EmbedResult } from \"./types.js\";\n\nexport type { IndexResult, EmbedResult };\n\n// Re-export detectTier for backward-compatibility (consumers import it from indexer.js)\nexport { detectTier };\n\n// ---------------------------------------------------------------------------\n// Single-file indexing\n// ---------------------------------------------------------------------------\n\n/**\n * Index a single file into the federation database.\n *\n * @returns true if the file was re-indexed (changed or new), false if skipped.\n */\nexport function indexFile(\n  db: Database,\n  projectId: number,\n  rootPath: string,\n  relativePath: string,\n  source: string,\n  tier: string,\n): boolean {\n  const absPath = join(rootPath, relativePath);\n\n  // Read file content\n  let content: string;\n  let stat: ReturnType<typeof statSync>;\n  try {\n    content = readFileSync(absPath, \"utf8\");\n    stat = statSync(absPath);\n  } catch {\n    // File unreadable or missing — skip silently\n    return false;\n  }\n\n  const hash = sha256File(content);\n  const mtime = Math.floor(stat.mtimeMs);\n  const size = stat.size;\n\n  // Check if the file has changed since last index\n  const existing = db\n    .prepare(\n      \"SELECT hash FROM memory_files WHERE project_id = ? AND path = ?\",\n    )\n    .get(projectId, relativePath) as { hash: string } | undefined;\n\n  if (existing?.hash === hash) {\n    // Unchanged — skip\n    return false;\n  }\n\n  // Delete old chunks for this file from both tables\n  const oldChunkIds = db\n    .prepare(\n      \"SELECT id FROM memory_chunks WHERE project_id = ? AND path = ?\",\n    )\n    .all(projectId, relativePath) as Array<{ id: string }>;\n\n  const deleteFts = db.prepare(\"DELETE FROM memory_fts WHERE id = ?\");\n  const deleteChunk = db.prepare(\n    \"DELETE FROM memory_chunks WHERE project_id = ? AND path = ?\",\n  );\n\n  db.transaction(() => {\n    for (const row of oldChunkIds) {\n      deleteFts.run(row.id);\n    }\n    deleteChunk.run(projectId, relativePath);\n  })();\n\n  // Chunk the new content\n  const chunks = chunkMarkdown(content);\n\n  // Insert new chunks into memory_chunks and memory_fts\n  const insertChunk = db.prepare(`\n    INSERT INTO memory_chunks (id, project_id, source, tier, path, start_line, end_line, hash, text, updated_at)\n    VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)\n  `);\n\n  const insertFts = db.prepare(`\n    INSERT INTO memory_fts (text, id, project_id, path, source, tier, start_line, end_line)\n    VALUES (?, ?, ?, ?, ?, ?, ?, ?)\n  `);\n\n  const upsertFile = db.prepare(`\n    INSERT INTO memory_files (project_id, path, source, tier, hash, mtime, size)\n    VALUES (?, ?, ?, ?, ?, ?, ?)\n    ON CONFLICT(project_id, path) DO UPDATE SET\n      source = excluded.source,\n      tier   = excluded.tier,\n      hash   = excluded.hash,\n      mtime  = excluded.mtime,\n      size   = excluded.size\n  `);\n\n  const updatedAt = Date.now();\n\n  db.transaction(() => {\n    for (let i = 0; i < chunks.length; i++) {\n      const chunk = chunks[i]!;\n      const id = chunkId(projectId, relativePath, i, chunk.startLine, chunk.endLine);\n      insertChunk.run(\n        id,\n        projectId,\n        source,\n        tier,\n        relativePath,\n        chunk.startLine,\n        chunk.endLine,\n        chunk.hash,\n        chunk.text,\n        updatedAt,\n      );\n      insertFts.run(\n        chunk.text,\n        id,\n        projectId,\n        relativePath,\n        source,\n        tier,\n        chunk.startLine,\n        chunk.endLine,\n      );\n    }\n    upsertFile.run(projectId, relativePath, source, tier, hash, mtime, size);\n  })();\n\n  return true;\n}\n\n// ---------------------------------------------------------------------------\n// Project-level indexing\n// ---------------------------------------------------------------------------\n\n/**\n * Index all memory, Notes, and content files for a single registered project.\n *\n * Scans:\n *  - {rootPath}/MEMORY.md    → source='memory', tier='evergreen'\n *  - {rootPath}/memory/      → source='memory', tier from detectTier()\n *  - {rootPath}/Notes/       → source='notes',  tier='session'\n *  - {rootPath}/**\\/*.md    → source='content', tier='topic'  (all other .md files, recursive)\n *  - {claudeNotesDir}/       → source='notes',  tier='session'  (if set and different)\n */\nexport async function indexProject(\n  db: Database,\n  projectId: number,\n  rootPath: string,\n  claudeNotesDir?: string | null,\n): Promise<IndexResult> {\n  const result: IndexResult = {\n    filesProcessed: 0,\n    chunksCreated: 0,\n    filesSkipped: 0,\n  };\n\n  const filesToIndex: Array<{ absPath: string; rootBase: string; source: string; tier: string }> = [];\n\n  // Root-level MEMORY.md\n  const rootMemoryMd = join(rootPath, \"MEMORY.md\");\n  if (existsSync(rootMemoryMd)) {\n    filesToIndex.push({ absPath: rootMemoryMd, rootBase: rootPath, source: \"memory\", tier: \"evergreen\" });\n  }\n\n  // memory/ directory\n  const memoryDir = join(rootPath, \"memory\");\n  for (const absPath of walkMdFiles(memoryDir)) {\n    const relPath = relative(rootPath, absPath);\n    const tier = detectTier(relPath);\n    filesToIndex.push({ absPath, rootBase: rootPath, source: \"memory\", tier });\n  }\n\n  // {rootPath}/Notes/ directory\n  const notesDir = join(rootPath, \"Notes\");\n  for (const absPath of walkMdFiles(notesDir)) {\n    filesToIndex.push({ absPath, rootBase: rootPath, source: \"notes\", tier: \"session\" });\n  }\n\n  // Synthetic session-title chunks for Notes files with the standard filename format:\n  // \"NNNN - YYYY-MM-DD - Descriptive Title.md\"\n  {\n    const titleInsertChunk = db.prepare(`\n      INSERT OR IGNORE INTO memory_chunks (id, project_id, source, tier, path, start_line, end_line, hash, text, updated_at)\n      VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)\n    `);\n    const titleInsertFts = db.prepare(`\n      INSERT OR IGNORE INTO memory_fts (text, id, project_id, path, source, tier, start_line, end_line)\n      VALUES (?, ?, ?, ?, ?, ?, ?, ?)\n    `);\n    const updatedAt = Date.now();\n    for (const absPath of walkMdFiles(notesDir)) {\n      const fileName = basename(absPath);\n      const text = parseSessionTitleChunk(fileName);\n      if (!text) continue;\n      const relPath = relative(rootPath, absPath);\n      const syntheticPath = `${relPath}::title`;\n      const id = chunkId(projectId, syntheticPath, 0, 0, 0);\n      const hash = sha256File(text);\n      db.transaction(() => {\n        titleInsertChunk.run(id, projectId, \"notes\", \"session\", syntheticPath, 0, 0, hash, text, updatedAt);\n        titleInsertFts.run(text, id, projectId, syntheticPath, \"notes\", \"session\", 0, 0);\n      })();\n    }\n  }\n\n  // {rootPath}/**/*.md — all other markdown content\n  if (!isPathTooBroadForContentScan(rootPath)) {\n    for (const absPath of walkContentFiles(rootPath)) {\n      filesToIndex.push({ absPath, rootBase: rootPath, source: \"content\", tier: \"topic\" });\n    }\n  }\n\n  // Claude Code session notes directory (~/.claude/projects/{encoded}/Notes/)\n  if (claudeNotesDir && claudeNotesDir !== notesDir) {\n    for (const absPath of walkMdFiles(claudeNotesDir)) {\n      filesToIndex.push({ absPath, rootBase: claudeNotesDir, source: \"notes\", tier: \"session\" });\n    }\n\n    // Synthetic title chunks for claude notes dir\n    {\n      const updatedAt = Date.now();\n      const titleInsertChunk2 = db.prepare(`\n        INSERT OR IGNORE INTO memory_chunks (id, project_id, source, tier, path, start_line, end_line, hash, text, updated_at)\n        VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)\n      `);\n      const titleInsertFts2 = db.prepare(`\n        INSERT OR IGNORE INTO memory_fts (text, id, project_id, path, source, tier, start_line, end_line)\n        VALUES (?, ?, ?, ?, ?, ?, ?, ?)\n      `);\n      for (const absPath of walkMdFiles(claudeNotesDir)) {\n        const fileName = basename(absPath);\n        const text = parseSessionTitleChunk(fileName);\n        if (!text) continue;\n        const relPath = relative(claudeNotesDir, absPath);\n        const syntheticPath = `${relPath}::title`;\n        const id = chunkId(projectId, syntheticPath, 0, 0, 0);\n        const hash = sha256File(text);\n        db.transaction(() => {\n          titleInsertChunk2.run(id, projectId, \"notes\", \"session\", syntheticPath, 0, 0, hash, text, updatedAt);\n          titleInsertFts2.run(text, id, projectId, syntheticPath, \"notes\", \"session\", 0, 0);\n        })();\n      }\n    }\n\n    // Derive the sibling memory/ directory: .../Notes/ → .../memory/\n    if (claudeNotesDir.endsWith(\"/Notes\")) {\n      const claudeProjectDir = claudeNotesDir.slice(0, -\"/Notes\".length);\n      const claudeMemoryDir = join(claudeProjectDir, \"memory\");\n\n      const claudeMemoryMd = join(claudeProjectDir, \"MEMORY.md\");\n      if (existsSync(claudeMemoryMd)) {\n        filesToIndex.push({\n          absPath: claudeMemoryMd,\n          rootBase: claudeProjectDir,\n          source: \"memory\",\n          tier: \"evergreen\",\n        });\n      }\n\n      for (const absPath of walkMdFiles(claudeMemoryDir)) {\n        const relPath = relative(claudeProjectDir, absPath);\n        const tier = detectTier(relPath);\n        filesToIndex.push({ absPath, rootBase: claudeProjectDir, source: \"memory\", tier });\n      }\n    }\n  }\n\n  // Yield after collection phase before processing\n  await yieldToEventLoop();\n\n  let filesSinceYield = 0;\n\n  for (const { absPath, rootBase, source, tier } of filesToIndex) {\n    if (filesSinceYield >= INDEX_YIELD_EVERY) {\n      await yieldToEventLoop();\n      filesSinceYield = 0;\n    }\n    filesSinceYield++;\n\n    const relPath = relative(rootBase, absPath);\n    const changed = indexFile(db, projectId, rootBase, relPath, source, tier);\n\n    if (changed) {\n      const count = db\n        .prepare(\n          \"SELECT COUNT(*) as n FROM memory_chunks WHERE project_id = ? AND path = ?\",\n        )\n        .get(projectId, relPath) as { n: number };\n\n      result.filesProcessed++;\n      result.chunksCreated += count.n;\n    } else {\n      result.filesSkipped++;\n    }\n  }\n\n  // Prune stale paths: remove DB entries for files that no longer exist on disk.\n  const livePaths = new Set<string>();\n  for (const { absPath, rootBase } of filesToIndex) {\n    livePaths.add(relative(rootBase, absPath));\n  }\n\n  const dbChunkPaths = db\n    .prepare(\"SELECT DISTINCT path FROM memory_chunks WHERE project_id = ?\")\n    .all(projectId) as Array<{ path: string }>;\n\n  const stalePaths: string[] = [];\n  for (const row of dbChunkPaths) {\n    const basePath = row.path.endsWith(\"::title\")\n      ? row.path.slice(0, -\"::title\".length)\n      : row.path;\n    if (!livePaths.has(basePath)) {\n      stalePaths.push(row.path);\n    }\n  }\n\n  if (stalePaths.length > 0) {\n    const deleteChunksFts = db.prepare(\"DELETE FROM memory_fts WHERE id = ?\");\n    const deleteChunks = db.prepare(\n      \"DELETE FROM memory_chunks WHERE project_id = ? AND path = ?\",\n    );\n    const deleteFile = db.prepare(\n      \"DELETE FROM memory_files WHERE project_id = ? AND path = ?\",\n    );\n\n    db.transaction(() => {\n      for (const stalePath of stalePaths) {\n        const chunkIds = db\n          .prepare(\"SELECT id FROM memory_chunks WHERE project_id = ? AND path = ?\")\n          .all(projectId, stalePath) as Array<{ id: string }>;\n        for (const { id } of chunkIds) {\n          deleteChunksFts.run(id);\n        }\n        deleteChunks.run(projectId, stalePath);\n        deleteFile.run(projectId, stalePath);\n      }\n    })();\n  }\n\n  return result;\n}\n\n// ---------------------------------------------------------------------------\n// Global indexing (all registered projects)\n// ---------------------------------------------------------------------------\n\n/**\n * Index all active projects registered in the registry DB.\n *\n * Async: yields to the event loop between each project so that the daemon's\n * Unix socket server can process IPC requests (e.g. status) while indexing.\n */\nexport async function indexAll(\n  db: Database,\n  registryDb: Database,\n): Promise<{ projects: number; result: IndexResult }> {\n  const projects = registryDb\n    .prepare(\"SELECT id, root_path, claude_notes_dir FROM projects WHERE status = 'active'\")\n    .all() as Array<{ id: number; root_path: string; claude_notes_dir: string | null }>;\n\n  const totals: IndexResult = {\n    filesProcessed: 0,\n    chunksCreated: 0,\n    filesSkipped: 0,\n  };\n\n  for (const project of projects) {\n    await yieldToEventLoop();\n    const r = await indexProject(db, project.id, project.root_path, project.claude_notes_dir);\n    totals.filesProcessed += r.filesProcessed;\n    totals.chunksCreated += r.chunksCreated;\n    totals.filesSkipped += r.filesSkipped;\n  }\n\n  return { projects: projects.length, result: totals };\n}\n\n// ---------------------------------------------------------------------------\n// Embedding generation\n// ---------------------------------------------------------------------------\n\n/**\n * Generate and store embeddings for chunks that do not yet have one.\n *\n * Because better-sqlite3 is synchronous but the embedding pipeline is async,\n * we fetch all unembedded chunk texts first, generate embeddings in batches,\n * and then write them back in a transaction.\n *\n * @param db         Open federation database.\n * @param projectId  Optional — restrict to a specific project.\n * @param batchSize  Number of chunks to embed per round. Default 50.\n * @param onProgress Optional callback called after each batch with running totals.\n */\nexport async function embedChunks(\n  db: Database,\n  projectId?: number,\n  batchSize = 50,\n  onProgress?: (embedded: number, total: number) => void,\n): Promise<EmbedResult> {\n  // Dynamic import — keeps the heavy ML runtime out of the module load path\n  const { generateEmbedding, serializeEmbedding } = await import(\"../embeddings.js\");\n\n  const conditions = [\"embedding IS NULL\"];\n  const params: (string | number)[] = [];\n\n  if (projectId !== undefined) {\n    conditions.push(\"project_id = ?\");\n    params.push(projectId);\n  }\n\n  const where = \"WHERE \" + conditions.join(\" AND \");\n\n  const rows = db\n    .prepare(`SELECT id, text FROM memory_chunks ${where} ORDER BY id`)\n    .all(...params) as Array<{ id: string; text: string }>;\n\n  if (rows.length === 0) {\n    return { chunksEmbedded: 0, chunksSkipped: 0 };\n  }\n\n  const updateStmt = db.prepare(\n    \"UPDATE memory_chunks SET embedding = ? WHERE id = ?\",\n  );\n\n  let embedded = 0;\n  const total = rows.length;\n\n  for (let i = 0; i < rows.length; i += batchSize) {\n    const batch = rows.slice(i, i + batchSize);\n\n    // Generate embeddings for the batch (async — must happen OUTSIDE transaction)\n    const embeddings: Array<{ id: string; blob: Buffer }> = [];\n    for (const row of batch) {\n      const vec = await generateEmbedding(row.text);\n      const blob = serializeEmbedding(vec);\n      embeddings.push({ id: row.id, blob });\n    }\n\n    // Write the batch in a single transaction\n    db.transaction(() => {\n      for (const { id, blob } of embeddings) {\n        updateStmt.run(blob, id);\n      }\n    })();\n\n    embedded += embeddings.length;\n    onProgress?.(embedded, total);\n  }\n\n  return { chunksEmbedded: embedded, chunksSkipped: 0 };\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;AA4CA,SAAgB,UACd,IACA,WACA,UACA,cACA,QACA,MACS;CACT,MAAM,UAAU,KAAK,UAAU,aAAa;CAG5C,IAAI;CACJ,IAAI;AACJ,KAAI;AACF,YAAU,aAAa,SAAS,OAAO;AACvC,SAAO,SAAS,QAAQ;SAClB;AAEN,SAAO;;CAGT,MAAM,OAAO,WAAW,QAAQ;CAChC,MAAM,QAAQ,KAAK,MAAM,KAAK,QAAQ;CACtC,MAAM,OAAO,KAAK;AASlB,KANiB,GACd,QACC,kEACD,CACA,IAAI,WAAW,aAAa,EAEjB,SAAS,KAErB,QAAO;CAIT,MAAM,cAAc,GACjB,QACC,iEACD,CACA,IAAI,WAAW,aAAa;CAE/B,MAAM,YAAY,GAAG,QAAQ,sCAAsC;CACnE,MAAM,cAAc,GAAG,QACrB,8DACD;AAED,IAAG,kBAAkB;AACnB,OAAK,MAAM,OAAO,YAChB,WAAU,IAAI,IAAI,GAAG;AAEvB,cAAY,IAAI,WAAW,aAAa;GACxC,EAAE;CAGJ,MAAM,SAAS,cAAc,QAAQ;CAGrC,MAAM,cAAc,GAAG,QAAQ;;;IAG7B;CAEF,MAAM,YAAY,GAAG,QAAQ;;;IAG3B;CAEF,MAAM,aAAa,GAAG,QAAQ;;;;;;;;;IAS5B;CAEF,MAAM,YAAY,KAAK,KAAK;AAE5B,IAAG,kBAAkB;AACnB,OAAK,IAAI,IAAI,GAAG,IAAI,OAAO,QAAQ,KAAK;GACtC,MAAM,QAAQ,OAAO;GACrB,MAAM,KAAK,QAAQ,WAAW,cAAc,GAAG,MAAM,WAAW,MAAM,QAAQ;AAC9E,eAAY,IACV,IACA,WACA,QACA,MACA,cACA,MAAM,WACN,MAAM,SACN,MAAM,MACN,MAAM,MACN,UACD;AACD,aAAU,IACR,MAAM,MACN,IACA,WACA,cACA,QACA,MACA,MAAM,WACN,MAAM,QACP;;AAEH,aAAW,IAAI,WAAW,cAAc,QAAQ,MAAM,MAAM,OAAO,KAAK;GACxE,EAAE;AAEJ,QAAO;;;;;;;;;;;;AAiBT,eAAsB,aACpB,IACA,WACA,UACA,gBACsB;CACtB,MAAM,SAAsB;EAC1B,gBAAgB;EAChB,eAAe;EACf,cAAc;EACf;CAED,MAAM,eAA2F,EAAE;CAGnG,MAAM,eAAe,KAAK,UAAU,YAAY;AAChD,KAAI,WAAW,aAAa,CAC1B,cAAa,KAAK;EAAE,SAAS;EAAc,UAAU;EAAU,QAAQ;EAAU,MAAM;EAAa,CAAC;CAIvG,MAAM,YAAY,KAAK,UAAU,SAAS;AAC1C,MAAK,MAAM,WAAW,YAAY,UAAU,EAAE;EAE5C,MAAM,OAAO,WADG,SAAS,UAAU,QAAQ,CACX;AAChC,eAAa,KAAK;GAAE;GAAS,UAAU;GAAU,QAAQ;GAAU;GAAM,CAAC;;CAI5E,MAAM,WAAW,KAAK,UAAU,QAAQ;AACxC,MAAK,MAAM,WAAW,YAAY,SAAS,CACzC,cAAa,KAAK;EAAE;EAAS,UAAU;EAAU,QAAQ;EAAS,MAAM;EAAW,CAAC;CAKtF;EACE,MAAM,mBAAmB,GAAG,QAAQ;;;MAGlC;EACF,MAAM,iBAAiB,GAAG,QAAQ;;;MAGhC;EACF,MAAM,YAAY,KAAK,KAAK;AAC5B,OAAK,MAAM,WAAW,YAAY,SAAS,EAAE;GAE3C,MAAM,OAAO,uBADI,SAAS,QAAQ,CACW;AAC7C,OAAI,CAAC,KAAM;GAEX,MAAM,gBAAgB,GADN,SAAS,UAAU,QAAQ,CACV;GACjC,MAAM,KAAK,QAAQ,WAAW,eAAe,GAAG,GAAG,EAAE;GACrD,MAAM,OAAO,WAAW,KAAK;AAC7B,MAAG,kBAAkB;AACnB,qBAAiB,IAAI,IAAI,WAAW,SAAS,WAAW,eAAe,GAAG,GAAG,MAAM,MAAM,UAAU;AACnG,mBAAe,IAAI,MAAM,IAAI,WAAW,eAAe,SAAS,WAAW,GAAG,EAAE;KAChF,EAAE;;;AAKR,KAAI,CAAC,6BAA6B,SAAS,CACzC,MAAK,MAAM,WAAW,iBAAiB,SAAS,CAC9C,cAAa,KAAK;EAAE;EAAS,UAAU;EAAU,QAAQ;EAAW,MAAM;EAAS,CAAC;AAKxF,KAAI,kBAAkB,mBAAmB,UAAU;AACjD,OAAK,MAAM,WAAW,YAAY,eAAe,CAC/C,cAAa,KAAK;GAAE;GAAS,UAAU;GAAgB,QAAQ;GAAS,MAAM;GAAW,CAAC;EAI5F;GACE,MAAM,YAAY,KAAK,KAAK;GAC5B,MAAM,oBAAoB,GAAG,QAAQ;;;QAGnC;GACF,MAAM,kBAAkB,GAAG,QAAQ;;;QAGjC;AACF,QAAK,MAAM,WAAW,YAAY,eAAe,EAAE;IAEjD,MAAM,OAAO,uBADI,SAAS,QAAQ,CACW;AAC7C,QAAI,CAAC,KAAM;IAEX,MAAM,gBAAgB,GADN,SAAS,gBAAgB,QAAQ,CAChB;IACjC,MAAM,KAAK,QAAQ,WAAW,eAAe,GAAG,GAAG,EAAE;IACrD,MAAM,OAAO,WAAW,KAAK;AAC7B,OAAG,kBAAkB;AACnB,uBAAkB,IAAI,IAAI,WAAW,SAAS,WAAW,eAAe,GAAG,GAAG,MAAM,MAAM,UAAU;AACpG,qBAAgB,IAAI,MAAM,IAAI,WAAW,eAAe,SAAS,WAAW,GAAG,EAAE;MACjF,EAAE;;;AAKR,MAAI,eAAe,SAAS,SAAS,EAAE;GACrC,MAAM,mBAAmB,eAAe,MAAM,GAAG,GAAiB;GAClE,MAAM,kBAAkB,KAAK,kBAAkB,SAAS;GAExD,MAAM,iBAAiB,KAAK,kBAAkB,YAAY;AAC1D,OAAI,WAAW,eAAe,CAC5B,cAAa,KAAK;IAChB,SAAS;IACT,UAAU;IACV,QAAQ;IACR,MAAM;IACP,CAAC;AAGJ,QAAK,MAAM,WAAW,YAAY,gBAAgB,EAAE;IAElD,MAAM,OAAO,WADG,SAAS,kBAAkB,QAAQ,CACnB;AAChC,iBAAa,KAAK;KAAE;KAAS,UAAU;KAAkB,QAAQ;KAAU;KAAM,CAAC;;;;AAMxF,OAAM,kBAAkB;CAExB,IAAI,kBAAkB;AAEtB,MAAK,MAAM,EAAE,SAAS,UAAU,QAAQ,UAAU,cAAc;AAC9D,MAAI,mBAAmB,mBAAmB;AACxC,SAAM,kBAAkB;AACxB,qBAAkB;;AAEpB;EAEA,MAAM,UAAU,SAAS,UAAU,QAAQ;AAG3C,MAFgB,UAAU,IAAI,WAAW,UAAU,SAAS,QAAQ,KAAK,EAE5D;GACX,MAAM,QAAQ,GACX,QACC,4EACD,CACA,IAAI,WAAW,QAAQ;AAE1B,UAAO;AACP,UAAO,iBAAiB,MAAM;QAE9B,QAAO;;CAKX,MAAM,4BAAY,IAAI,KAAa;AACnC,MAAK,MAAM,EAAE,SAAS,cAAc,aAClC,WAAU,IAAI,SAAS,UAAU,QAAQ,CAAC;CAG5C,MAAM,eAAe,GAClB,QAAQ,+DAA+D,CACvE,IAAI,UAAU;CAEjB,MAAM,aAAuB,EAAE;AAC/B,MAAK,MAAM,OAAO,cAAc;EAC9B,MAAM,WAAW,IAAI,KAAK,SAAS,UAAU,GACzC,IAAI,KAAK,MAAM,GAAG,GAAkB,GACpC,IAAI;AACR,MAAI,CAAC,UAAU,IAAI,SAAS,CAC1B,YAAW,KAAK,IAAI,KAAK;;AAI7B,KAAI,WAAW,SAAS,GAAG;EACzB,MAAM,kBAAkB,GAAG,QAAQ,sCAAsC;EACzE,MAAM,eAAe,GAAG,QACtB,8DACD;EACD,MAAM,aAAa,GAAG,QACpB,6DACD;AAED,KAAG,kBAAkB;AACnB,QAAK,MAAM,aAAa,YAAY;IAClC,MAAM,WAAW,GACd,QAAQ,iEAAiE,CACzE,IAAI,WAAW,UAAU;AAC5B,SAAK,MAAM,EAAE,QAAQ,SACnB,iBAAgB,IAAI,GAAG;AAEzB,iBAAa,IAAI,WAAW,UAAU;AACtC,eAAW,IAAI,WAAW,UAAU;;IAEtC,EAAE;;AAGN,QAAO;;;;;;;;AAaT,eAAsB,SACpB,IACA,YACoD;CACpD,MAAM,WAAW,WACd,QAAQ,+EAA+E,CACvF,KAAK;CAER,MAAM,SAAsB;EAC1B,gBAAgB;EAChB,eAAe;EACf,cAAc;EACf;AAED,MAAK,MAAM,WAAW,UAAU;AAC9B,QAAM,kBAAkB;EACxB,MAAM,IAAI,MAAM,aAAa,IAAI,QAAQ,IAAI,QAAQ,WAAW,QAAQ,iBAAiB;AACzF,SAAO,kBAAkB,EAAE;AAC3B,SAAO,iBAAiB,EAAE;AAC1B,SAAO,gBAAgB,EAAE;;AAG3B,QAAO;EAAE,UAAU,SAAS;EAAQ,QAAQ;EAAQ;;;;;;;;;;;;;;AAmBtD,eAAsB,YACpB,IACA,WACA,YAAY,IACZ,YACsB;CAEtB,MAAM,EAAE,mBAAmB,uBAAuB,MAAM,OAAO;CAE/D,MAAM,aAAa,CAAC,oBAAoB;CACxC,MAAM,SAA8B,EAAE;AAEtC,KAAI,cAAc,QAAW;AAC3B,aAAW,KAAK,iBAAiB;AACjC,SAAO,KAAK,UAAU;;CAGxB,MAAM,QAAQ,WAAW,WAAW,KAAK,QAAQ;CAEjD,MAAM,OAAO,GACV,QAAQ,sCAAsC,MAAM,cAAc,CAClE,IAAI,GAAG,OAAO;AAEjB,KAAI,KAAK,WAAW,EAClB,QAAO;EAAE,gBAAgB;EAAG,eAAe;EAAG;CAGhD,MAAM,aAAa,GAAG,QACpB,sDACD;CAED,IAAI,WAAW;CACf,MAAM,QAAQ,KAAK;AAEnB,MAAK,IAAI,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK,WAAW;EAC/C,MAAM,QAAQ,KAAK,MAAM,GAAG,IAAI,UAAU;EAG1C,MAAM,aAAkD,EAAE;AAC1D,OAAK,MAAM,OAAO,OAAO;GAEvB,MAAM,OAAO,mBADD,MAAM,kBAAkB,IAAI,KAAK,CACT;AACpC,cAAW,KAAK;IAAE,IAAI,IAAI;IAAI;IAAM,CAAC;;AAIvC,KAAG,kBAAkB;AACnB,QAAK,MAAM,EAAE,IAAI,UAAU,WACzB,YAAW,IAAI,MAAM,GAAG;IAE1B,EAAE;AAEJ,cAAY,WAAW;AACvB,eAAa,UAAU,MAAM;;AAG/B,QAAO;EAAE,gBAAgB;EAAU,eAAe;EAAG"}

package/dist/{vault-indexer-DsPC6wQf.mjs → vault-indexer-vNnL3Azb.mjs}

@@ -1,4 +1,4 @@
-import { c as yieldToEventLoop, f as sha256File, l as chunkMarkdown, n as chunkId } from "./helpers-OCVFgprQ.mjs";
+import { c as yieldToEventLoop, f as sha256File, l as chunkMarkdown, n as chunkId } from "./helpers-crDEr6S2.mjs";
 import { existsSync, readFileSync, readdirSync, statSync } from "node:fs";
 import { basename, dirname, join, normalize, relative } from "node:path";
 
@@ -533,4 +533,4 @@ async function indexVault(backend, vaultProjectId, vaultRoot) {
 
 //#endregion
 export { indexVault };
-//# sourceMappingURL=vault-indexer-DsPC6wQf.mjs.map
+//# sourceMappingURL=vault-indexer-vNnL3Azb.mjs.map

package/dist/{vault-indexer-DsPC6wQf.mjs.map → vault-indexer-vNnL3Azb.mjs.map}

@@ -1 +1 @@
-{"version":3,"file":"vault-indexer-DsPC6wQf.mjs","names":[],"sources":["../src/memory/vault/walk.ts","../src/memory/vault/deduplicate.ts","../src/memory/vault/parse-links.ts","../src/memory/vault/name-index.ts","../src/memory/vault/resolve.ts","../src/memory/vault/indexer.ts"],"sourcesContent":["/**\n * Vault directory walker — follows symlinks with cycle detection.\n */\n\nimport { statSync, readdirSync, existsSync } from \"node:fs\";\nimport { join, relative } from \"node:path\";\nimport type { VaultFile } from \"./types.js\";\n\n/** Maximum number of .md files to collect from a vault. */\nconst VAULT_MAX_FILES = 10_000;\n\n/** Maximum recursion depth for vault directory walks. */\nconst VAULT_MAX_DEPTH = 10;\n\n/**\n * Directories to always skip, at any depth, during vault walks.\n * Includes standard build/VCS noise plus Obsidian-specific directories.\n */\nconst VAULT_SKIP_DIRS = new Set([\n  // Version control\n  \".git\",\n  // Dependency directories (any language)\n  \"node_modules\",\n  \"vendor\",\n  \"Pods\",\n  // Build / compile output\n  \"dist\",\n  \"build\",\n  \"out\",\n  \"DerivedData\",\n  \".next\",\n  // Python virtual environments and caches\n  \".venv\",\n  \"venv\",\n  \"__pycache__\",\n  // General caches\n  \".cache\",\n  \".bun\",\n  // Obsidian internals\n  \".obsidian\",\n  \".trash\",\n]);\n\n/**\n * Recursively collect all .md files under a vault root, following symlinks.\n *\n * Symlink-following behaviour:\n *  - Symbolic links to files: followed if the target is a .md file\n *  - Symbolic links to directories: followed with cycle detection via inode\n *\n * Cycle detection is based on the real inode of each visited directory.\n * Using the real stat (not lstat) ensures that symlinked dirs resolve to\n * their actual inode, preventing infinite loops.\n *\n * @param vaultRoot  Absolute path to the vault root directory.\n * @param opts       Optional overrides for maxFiles and maxDepth.\n */\nexport function walkVaultMdFiles(\n  vaultRoot: string,\n  opts?: { maxFiles?: number; maxDepth?: number },\n): VaultFile[] {\n  const maxFiles = opts?.maxFiles ?? VAULT_MAX_FILES;\n  const maxDepth = opts?.maxDepth ?? VAULT_MAX_DEPTH;\n\n  const results: VaultFile[] = [];\n  const visitedDirs = new Set<string>();\n\n  function walk(dir: string, depth: number): void {\n    if (results.length >= maxFiles) return;\n    if (depth > maxDepth) return;\n\n    // Get the real inode of this directory (follows symlinks on the dir itself)\n    let dirStat: ReturnType<typeof statSync>;\n    try {\n      dirStat = statSync(dir);\n    } catch {\n      return; // Unreadable or broken symlink — skip\n    }\n\n    const dirKey = `${dirStat.dev}:${dirStat.ino}`;\n    if (visitedDirs.has(dirKey)) return; // Cycle detected\n    visitedDirs.add(dirKey);\n\n    let entries: import(\"node:fs\").Dirent<string>[];\n    try {\n      entries = readdirSync(dir, { withFileTypes: true, encoding: \"utf8\" });\n    } catch {\n      return; // Unreadable directory — skip\n    }\n\n    for (const entry of entries) {\n      if (results.length >= maxFiles) break;\n      if (VAULT_SKIP_DIRS.has(entry.name)) continue;\n\n      const full = join(dir, entry.name);\n\n      if (entry.isSymbolicLink()) {\n        // Follow the symlink — resolve to real target\n        let targetStat: ReturnType<typeof statSync>;\n        try {\n          targetStat = statSync(full); // statSync follows symlinks\n        } catch {\n          continue; // Broken symlink — skip\n        }\n\n        if (targetStat.isDirectory()) {\n          if (!VAULT_SKIP_DIRS.has(entry.name)) {\n            walk(full, depth + 1);\n          }\n        } else if (targetStat.isFile() && entry.name.endsWith(\".md\")) {\n          results.push({\n            absPath: full,\n            vaultRelPath: relative(vaultRoot, full),\n            inode: targetStat.ino,\n            device: targetStat.dev,\n          });\n        }\n      } else if (entry.isDirectory()) {\n        walk(full, depth + 1);\n      } else if (entry.isFile() && entry.name.endsWith(\".md\")) {\n        let fileStat: ReturnType<typeof statSync>;\n        try {\n          fileStat = statSync(full);\n        } catch {\n          continue;\n        }\n        results.push({\n          absPath: full,\n          vaultRelPath: relative(vaultRoot, full),\n          inode: fileStat.ino,\n          device: fileStat.dev,\n        });\n      }\n    }\n  }\n\n  if (existsSync(vaultRoot)) {\n    walk(vaultRoot, 0);\n  }\n\n  return results;\n}\n","/**\n * Inode-based deduplication for vault files.\n */\n\nimport type { VaultFile, InodeGroup } from \"./types.js\";\n\n/**\n * Group vault files by inode identity (device + inode).\n *\n * Within each group, the canonical file is chosen as the one with the\n * fewest path separators (shallowest), breaking ties by shortest string.\n * All other group members become aliases.\n */\nexport function deduplicateByInode(files: VaultFile[]): InodeGroup[] {\n  const groups = new Map<string, VaultFile[]>();\n\n  for (const file of files) {\n    const key = `${file.device}:${file.inode}`;\n    const existing = groups.get(key);\n    if (existing) {\n      existing.push(file);\n    } else {\n      groups.set(key, [file]);\n    }\n  }\n\n  const result: InodeGroup[] = [];\n\n  for (const group of groups.values()) {\n    if (group.length === 0) continue;\n\n    // Sort: fewest path separators first, then shortest string\n    const sorted = [...group].sort((a, b) => {\n      const aDepth = (a.vaultRelPath.match(/\\//g) ?? []).length;\n      const bDepth = (b.vaultRelPath.match(/\\//g) ?? []).length;\n      if (aDepth !== bDepth) return aDepth - bDepth;\n      return a.vaultRelPath.length - b.vaultRelPath.length;\n    });\n\n    const [canonical, ...aliases] = sorted as [VaultFile, ...VaultFile[]];\n    result.push({ canonical, aliases });\n  }\n\n  return result;\n}\n","/**\n * Markdown link parser for vault files.\n *\n * Handles wikilinks ([[target]]), markdown links ([text](path)),\n * embeds (![[target]]), and frontmatter wikilinks.\n */\n\nimport type { ParsedLink } from \"./types.js\";\n\n/**\n * Parse all links from markdown content.\n *\n * Handles:\n *  - Standard wikilinks: [[Target Note]]\n *  - Aliased wikilinks: [[Target Note|Display Text]]\n *  - Heading anchors: [[Target Note#Heading]] (stripped for resolution)\n *  - Embeds: ![[Target Note]]\n *  - Frontmatter wikilinks (YAML between --- delimiters)\n *  - Markdown links: [text](path/to/note.md)\n *  - Markdown embeds: ![alt](image.png)\n *\n * External URLs (http://, https://, mailto:, etc.) are excluded — only\n * relative paths are treated as vault links.\n *\n * @param content  Raw markdown file content.\n * @returns        Array of parsed links in document order.\n */\nexport function parseLinks(content: string): ParsedLink[] {\n  const links: ParsedLink[] = [];\n  const lines = content.split(\"\\n\");\n\n  // Determine frontmatter range (YAML between opening and closing ---)\n  let frontmatterEnd = 0;\n  if (content.startsWith(\"---\")) {\n    const closingIdx = content.indexOf(\"\\n---\", 3);\n    if (closingIdx !== -1) {\n      frontmatterEnd = content.slice(0, closingIdx + 4).split(\"\\n\").length - 1;\n    }\n  }\n\n  // Regex for [[wikilinks]] and ![[embeds]]\n  const wikilinkRe = /(!?)\\[\\[([^\\]]+?)\\]\\]/g;\n\n  // Regex for markdown links [text](target) and embeds ![alt](target)\n  const mdLinkRe = /(!)?\\[([^\\]]*)\\]\\(([^)]+)\\)/g;\n\n  for (let lineIdx = 0; lineIdx < lines.length; lineIdx++) {\n    const line = lines[lineIdx]!;\n    const lineNumber = lineIdx + 1; // 1-indexed\n    const isFrontmatter = lineIdx < frontmatterEnd;\n\n    // --- Wikilinks ---\n    wikilinkRe.lastIndex = 0;\n    let match: RegExpExecArray | null;\n    while ((match = wikilinkRe.exec(line)) !== null) {\n      const isEmbed = match[1] === \"!\";\n      const inner = match[2]!;\n\n      // Split on first | for alias\n      const pipeIdx = inner.indexOf(\"|\");\n      const beforePipe = pipeIdx === -1 ? inner : inner.slice(0, pipeIdx);\n      const alias = pipeIdx === -1 ? null : inner.slice(pipeIdx + 1);\n\n      // Strip heading anchor (everything after #)\n      const hashIdx = beforePipe.indexOf(\"#\");\n      const raw = hashIdx === -1 ? beforePipe.trim() : beforePipe.slice(0, hashIdx).trim();\n\n      if (!raw) continue; // Skip links with empty targets (e.g. [[#Heading]])\n\n      links.push({\n        raw,\n        alias: alias?.trim() ?? null,\n        lineNumber,\n        isEmbed: isEmbed && !isFrontmatter,\n        isMdLink: false,\n      });\n    }\n\n    // --- Markdown links --- (skip inside frontmatter)\n    if (!isFrontmatter) {\n      mdLinkRe.lastIndex = 0;\n      while ((match = mdLinkRe.exec(line)) !== null) {\n        const isEmbed = match[1] === \"!\";\n        const displayText = match[2]!;\n        let target = match[3]!.trim();\n\n        // Skip external URLs\n        if (/^[a-z][a-z0-9+.-]*:/i.test(target)) continue;\n\n        // Skip pure anchor links (#heading)\n        if (target.startsWith(\"#\")) continue;\n\n        // Strip heading anchor from target\n        const hashIdx = target.indexOf(\"#\");\n        if (hashIdx !== -1) target = target.slice(0, hashIdx);\n\n        // URL-decode (Obsidian encodes spaces as %20 in md links)\n        try {\n          target = decodeURIComponent(target);\n        } catch {\n          // Malformed encoding — use as-is\n        }\n\n        // Strip .md extension for resolution (resolveWikilink adds it back)\n        const raw = target.replace(/\\.md$/i, \"\").trim();\n        if (!raw) continue;\n\n        links.push({\n          raw,\n          alias: displayText || null,\n          lineNumber,\n          isEmbed,\n          isMdLink: true,\n        });\n      }\n    }\n  }\n\n  return links;\n}\n\n/** @deprecated Use {@link parseLinks} instead. */\nexport const parseWikilinks = parseLinks;\n","/**\n * Name index builder for Obsidian wikilink resolution.\n */\n\nimport { basename } from \"node:path\";\nimport type { VaultFile } from \"./types.js\";\n\n/**\n * Build a name lookup index for Obsidian wikilink resolution.\n *\n * Maps lowercase filename (without .md extension) to all vault-relative paths\n * that share that name. Includes both canonical paths and alias paths so that\n * wikilinks resolve regardless of which path the file is accessed through.\n */\nexport function buildNameIndex(files: VaultFile[]): Map<string, string[]> {\n  const index = new Map<string, string[]>();\n\n  for (const file of files) {\n    const name = basename(file.vaultRelPath, \".md\").toLowerCase();\n    const existing = index.get(name);\n    if (existing) {\n      existing.push(file.vaultRelPath);\n    } else {\n      index.set(name, [file.vaultRelPath]);\n    }\n  }\n\n  return index;\n}\n","/**\n * Wikilink resolver using Obsidian's shortest-match algorithm.\n */\n\nimport { normalize, basename, dirname } from \"node:path\";\n\n/**\n * Resolve a wikilink target to a vault-relative path using Obsidian's rules.\n *\n * Resolution algorithm:\n *  1. If raw contains \"/\", attempt exact path match (with and without .md).\n *  2. Normalize: lowercase the raw target, strip .md extension.\n *  3. Look up in the name index (all files with that basename).\n *  4. If exactly one match, return it.\n *  5. If multiple matches, pick the one closest to the source file\n *     (longest common directory prefix, then shortest overall path).\n *  6. If no matches, return null (dead link).\n *\n * @param raw         The raw link target (heading-stripped, pipe-stripped).\n * @param nameIndex   Map from lowercase basename-without-ext to vault paths.\n * @param sourcePath  Vault-relative path of the file containing the link.\n * @returns           Vault-relative path of the resolved target, or null.\n */\nexport function resolveWikilink(\n  raw: string,\n  nameIndex: Map<string, string[]>,\n  sourcePath: string,\n): string | null {\n  if (!raw) return null;\n\n  // Case 1: path contains \"/\" — try exact match with and without .md\n  if (raw.includes(\"/\")) {\n    const normalized = normalize(raw);\n    const normalizedMd = normalized.endsWith(\".md\") ? normalized : normalized + \".md\";\n\n    // Check if any indexed path matches (case-insensitive for macOS compatibility)\n    for (const [, paths] of nameIndex) {\n      for (const p of paths) {\n        if (p === normalizedMd || p === normalized) return p;\n        if (p.toLowerCase() === normalizedMd.toLowerCase()) return p;\n      }\n    }\n    // Fall through to name lookup in case the path prefix was wrong\n  }\n\n  // Normalize the raw target for name lookup.\n  // Use the basename only — Obsidian resolves by filename, not full path.\n  const rawBase = basename(raw)\n    .replace(/\\.md$/i, \"\")\n    .toLowerCase()\n    .trim();\n\n  if (!rawBase) return null;\n\n  const candidates = nameIndex.get(rawBase);\n\n  if (!candidates || candidates.length === 0) {\n    return null; // Dead link\n  }\n\n  if (candidates.length === 1) {\n    return candidates[0]!;\n  }\n\n  // Multiple matches — pick the one closest to the source file\n  const sourceDir = dirname(sourcePath);\n\n  let bestPath: string | null = null;\n  let bestPrefixLen = -1;\n  let bestPathLen = Infinity;\n\n  for (const candidate of candidates) {\n    const candidateDir = dirname(candidate);\n    const prefixLen = commonPrefixLength(sourceDir, candidateDir);\n    const pathLen = candidate.length;\n\n    if (\n      prefixLen > bestPrefixLen ||\n      (prefixLen === bestPrefixLen && pathLen < bestPathLen)\n    ) {\n      bestPrefixLen = prefixLen;\n      bestPathLen = pathLen;\n      bestPath = candidate;\n    }\n  }\n\n  return bestPath;\n}\n\n/**\n * Compute the length of the common prefix between two directory paths,\n * measured in path segments (not raw characters).\n *\n * Example: \"a/b/c\" and \"a/b/d\" → 2 (common: \"a\", \"b\")\n */\nfunction commonPrefixLength(a: string, b: string): number {\n  if (a === \".\" && b === \".\") return 0;\n  const aParts = a === \".\" ? [] : a.split(\"/\");\n  const bParts = b === \".\" ? [] : b.split(\"/\");\n  let count = 0;\n  const len = Math.min(aParts.length, bParts.length);\n  for (let i = 0; i < len; i++) {\n    if (aParts[i] === bParts[i]) {\n      count++;\n    } else {\n      break;\n    }\n  }\n  return count;\n}\n","/**\n * Main vault indexing orchestrator.\n *\n * Indexes an entire Obsidian vault (or any markdown knowledge base), following\n * symlinks, deduplicating files by inode, parsing wikilinks, and computing\n * per-file health metrics (orphan detection, dead links).\n *\n * Key differences from the project indexer (indexer/sync.ts):\n *  - Follows symbolic links (project indexer skips them)\n *  - Deduplicates files with the same inode (same content reachable via multiple paths)\n *  - Parses [[wikilinks]] and builds a directed link graph\n *  - Resolves wikilinks using Obsidian's shortest-match algorithm\n *  - Computes health metrics per file: inbound/outbound link counts, dead links, orphans\n */\n\nimport { readFileSync } from \"node:fs\";\nimport { basename } from \"node:path\";\nimport type {\n  StorageBackend,\n  VaultFileRow,\n  VaultAliasRow,\n  VaultLinkRow,\n  VaultHealthRow,\n  VaultNameEntry,\n  ChunkRow,\n} from \"../../storage/interface.js\";\nimport { chunkMarkdown } from \"../chunker.js\";\nimport { sha256File, chunkId, yieldToEventLoop } from \"../indexer/helpers.js\";\nimport { walkVaultMdFiles } from \"./walk.js\";\nimport { deduplicateByInode } from \"./deduplicate.js\";\nimport { parseLinks } from \"./parse-links.js\";\nimport { buildNameIndex } from \"./name-index.js\";\nimport { resolveWikilink } from \"./resolve.js\";\nimport type { VaultIndexResult } from \"./types.js\";\n\n/** Number of files to process before yielding to the event loop. */\nconst VAULT_YIELD_EVERY = 1;\n\n/**\n * Index an entire Obsidian vault (or markdown knowledge base) using the\n * async StorageBackend interface.\n *\n * Steps:\n *  1. Walk vault root, following symlinks.\n *  2. Deduplicate by inode — each unique file is indexed once.\n *  3. Build a name index for wikilink resolution.\n *  4. For each canonical file:\n *     a. SHA-256 hash for change detection — skip unchanged files.\n *     b. Read content, chunk with chunkMarkdown().\n *     c. Insert chunks into backend (memory_chunks and memory_fts).\n *     d. Upsert vault_files row.\n *  5. Record aliases in vault_aliases.\n *  6. Rebuild vault_name_index table.\n *  7. Rebuild vault_links:\n *     a. Parse [[wikilinks]] from each canonical file.\n *     b. Resolve each link with resolveWikilink().\n *     c. Insert into vault_links.\n *  8. Compute and upsert health metrics (vault_health).\n *  9. Return statistics.\n *\n * @param backend         StorageBackend to write to.\n * @param vaultProjectId  Registry project ID for the vault \"project\".\n * @param vaultRoot       Absolute path to the vault root directory.\n */\nexport async function indexVault(\n  backend: StorageBackend,\n  vaultProjectId: number,\n  vaultRoot: string,\n): Promise<VaultIndexResult> {\n  const startTime = Date.now();\n\n  const result: VaultIndexResult = {\n    filesIndexed: 0,\n    chunksCreated: 0,\n    filesSkipped: 0,\n    aliasesRecorded: 0,\n    linksExtracted: 0,\n    deadLinksFound: 0,\n    orphansFound: 0,\n    elapsed: 0,\n  };\n\n  // Step 1: Walk vault, collecting all .md files (follows symlinks)\n  const allFiles = walkVaultMdFiles(vaultRoot);\n\n  // Step 2: Deduplicate by inode\n  const inodeGroups = deduplicateByInode(allFiles);\n\n  // Step 3: Build name index (from all files including aliases, for resolution)\n  const nameIndex = buildNameIndex(allFiles);\n\n  // Step 4: Index each canonical file\n  await yieldToEventLoop();\n  let filesSinceYield = 0;\n\n  for (const group of inodeGroups) {\n    if (filesSinceYield >= VAULT_YIELD_EVERY) {\n      await yieldToEventLoop();\n      filesSinceYield = 0;\n    }\n    filesSinceYield++;\n\n    const { canonical } = group;\n\n    let content: string;\n    try {\n      content = readFileSync(canonical.absPath, \"utf8\");\n    } catch {\n      result.filesSkipped++;\n      continue;\n    }\n\n    const hash = sha256File(content);\n\n    // Change detection: skip if hash is unchanged\n    const existing = await backend.getVaultFile(canonical.vaultRelPath);\n    if (existing?.hash === hash) {\n      result.filesSkipped++;\n      continue;\n    }\n\n    // Delete old chunks for this vault path\n    await backend.deleteChunksForFile(vaultProjectId, canonical.vaultRelPath);\n\n    // Chunk the content\n    const chunks = chunkMarkdown(content);\n    const updatedAt = Date.now();\n\n    // Extract title from first H1 heading or filename\n    const titleMatch = /^#\\s+(.+)$/m.exec(content);\n    const title = titleMatch\n      ? titleMatch[1]!.trim()\n      : basename(canonical.vaultRelPath, \".md\");\n\n    // Build chunk rows\n    const chunkRows: ChunkRow[] = [];\n    for (let i = 0; i < chunks.length; i++) {\n      const chunk = chunks[i]!;\n      const id = chunkId(\n        vaultProjectId,\n        canonical.vaultRelPath,\n        i,\n        chunk.startLine,\n        chunk.endLine,\n      );\n      chunkRows.push({\n        id,\n        projectId: vaultProjectId,\n        source: \"vault\",\n        tier: \"topic\",\n        path: canonical.vaultRelPath,\n        startLine: chunk.startLine,\n        endLine: chunk.endLine,\n        hash: chunk.hash,\n        text: chunk.text,\n        updatedAt,\n      });\n    }\n\n    if (chunkRows.length > 0) {\n      await backend.insertChunks(chunkRows);\n    }\n\n    // Upsert vault file record\n    const vaultFileRow: VaultFileRow = {\n      vaultPath: canonical.vaultRelPath,\n      inode: canonical.inode,\n      device: canonical.device,\n      hash,\n      title,\n      indexedAt: updatedAt,\n    };\n    await backend.upsertVaultFile(vaultFileRow);\n\n    result.filesIndexed++;\n    result.chunksCreated += chunks.length;\n  }\n\n  // Step 5: Record aliases in vault_aliases\n  await yieldToEventLoop();\n\n  const allAliases: VaultAliasRow[] = [];\n  for (const group of inodeGroups) {\n    for (const alias of group.aliases) {\n      allAliases.push({\n        vaultPath: alias.vaultRelPath,\n        canonicalPath: group.canonical.vaultRelPath,\n        inode: alias.inode,\n        device: alias.device,\n      });\n      result.aliasesRecorded++;\n    }\n  }\n\n  const canonicalPaths = new Set(inodeGroups.map((g) => g.canonical.vaultRelPath));\n  for (const canonPath of canonicalPaths) {\n    await backend.deleteVaultAliases(canonPath);\n  }\n  if (allAliases.length > 0) {\n    await backend.upsertVaultAliases(allAliases);\n  }\n\n  // Step 6: Rebuild vault_name_index\n  await yieldToEventLoop();\n\n  const nameEntries: VaultNameEntry[] = [];\n  for (const [name, paths] of nameIndex) {\n    for (const path of paths) {\n      nameEntries.push({ name, vaultPath: path });\n    }\n  }\n  await backend.replaceNameIndex(nameEntries);\n\n  // Step 7: Rebuild vault_links\n  await yieldToEventLoop();\n\n  const linkRows: VaultLinkRow[] = [];\n  const allSourcePaths: string[] = [];\n\n  let linkParseYield = 0;\n  for (const group of inodeGroups) {\n    if (linkParseYield++ % VAULT_YIELD_EVERY === 0) {\n      await yieldToEventLoop();\n    }\n\n    const { canonical } = group;\n    allSourcePaths.push(canonical.vaultRelPath);\n\n    let content: string;\n    try {\n      content = readFileSync(canonical.absPath, \"utf8\");\n    } catch {\n      continue;\n    }\n\n    const parsedLinks = parseLinks(content);\n    for (const link of parsedLinks) {\n      const target = resolveWikilink(link.raw, nameIndex, canonical.vaultRelPath);\n      let linkType: string;\n      if (link.isMdLink) {\n        linkType = link.isEmbed ? \"md-embed\" : \"md-link\";\n      } else {\n        linkType = link.isEmbed ? \"embed\" : \"wikilink\";\n      }\n      linkRows.push({\n        sourcePath: canonical.vaultRelPath,\n        targetRaw: link.raw,\n        targetPath: target,\n        linkType,\n        lineNumber: link.lineNumber,\n        confidence: \"EXTRACTED\",\n      });\n    }\n  }\n\n  // Replace all links for all sources in batches of 500\n  const LINK_BATCH_SIZE = 500;\n  for (let i = 0; i < allSourcePaths.length; i += LINK_BATCH_SIZE) {\n    const batchSources = allSourcePaths.slice(i, i + LINK_BATCH_SIZE);\n    const batchLinks = linkRows.filter((r) => batchSources.includes(r.sourcePath));\n    await backend.replaceLinksForSources(batchSources, batchLinks);\n    await yieldToEventLoop();\n  }\n\n  result.linksExtracted = linkRows.length;\n  result.deadLinksFound = linkRows.filter((r) => r.targetPath === null).length;\n\n  // Step 8: Compute and upsert vault_health metrics\n  await yieldToEventLoop();\n\n  const outboundMap = new Map<string, number>();\n  const deadMap = new Map<string, number>();\n  const inboundMap = new Map<string, number>();\n\n  for (const row of linkRows) {\n    outboundMap.set(row.sourcePath, (outboundMap.get(row.sourcePath) ?? 0) + 1);\n    if (row.targetPath === null) {\n      deadMap.set(row.sourcePath, (deadMap.get(row.sourcePath) ?? 0) + 1);\n    } else {\n      inboundMap.set(row.targetPath, (inboundMap.get(row.targetPath) ?? 0) + 1);\n    }\n  }\n\n  const computedAt = Date.now();\n  let orphanCount = 0;\n\n  const HEALTH_BATCH_SIZE = 500;\n  for (let i = 0; i < inodeGroups.length; i += HEALTH_BATCH_SIZE) {\n    const batch = inodeGroups.slice(i, i + HEALTH_BATCH_SIZE);\n    const healthRows: VaultHealthRow[] = batch.map((group) => {\n      const path = group.canonical.vaultRelPath;\n      const inbound = inboundMap.get(path) ?? 0;\n      const outbound = outboundMap.get(path) ?? 0;\n      const dead = deadMap.get(path) ?? 0;\n      const isOrphan = inbound === 0;\n      if (isOrphan) orphanCount++;\n      return {\n        vaultPath: path,\n        inboundCount: inbound,\n        outboundCount: outbound,\n        deadLinkCount: dead,\n        isOrphan,\n        computedAt,\n      };\n    });\n    await backend.upsertVaultHealth(healthRows);\n    await yieldToEventLoop();\n  }\n\n  result.orphansFound = orphanCount;\n  result.elapsed = Date.now() - startTime;\n\n  return result;\n}\n"],"mappings":";;;;;;;;;AASA,MAAM,kBAAkB;;AAGxB,MAAM,kBAAkB;;;;;AAMxB,MAAM,kBAAkB,IAAI,IAAI;CAE9B;CAEA;CACA;CACA;CAEA;CACA;CACA;CACA;CACA;CAEA;CACA;CACA;CAEA;CACA;CAEA;CACA;CACD,CAAC;;;;;;;;;;;;;;;AAgBF,SAAgB,iBACd,WACA,MACa;CACb,MAAM,WAAW,MAAM,YAAY;CACnC,MAAM,WAAW,MAAM,YAAY;CAEnC,MAAM,UAAuB,EAAE;CAC/B,MAAM,8BAAc,IAAI,KAAa;CAErC,SAAS,KAAK,KAAa,OAAqB;AAC9C,MAAI,QAAQ,UAAU,SAAU;AAChC,MAAI,QAAQ,SAAU;EAGtB,IAAI;AACJ,MAAI;AACF,aAAU,SAAS,IAAI;UACjB;AACN;;EAGF,MAAM,SAAS,GAAG,QAAQ,IAAI,GAAG,QAAQ;AACzC,MAAI,YAAY,IAAI,OAAO,CAAE;AAC7B,cAAY,IAAI,OAAO;EAEvB,IAAI;AACJ,MAAI;AACF,aAAU,YAAY,KAAK;IAAE,eAAe;IAAM,UAAU;IAAQ,CAAC;UAC/D;AACN;;AAGF,OAAK,MAAM,SAAS,SAAS;AAC3B,OAAI,QAAQ,UAAU,SAAU;AAChC,OAAI,gBAAgB,IAAI,MAAM,KAAK,CAAE;GAErC,MAAM,OAAO,KAAK,KAAK,MAAM,KAAK;AAElC,OAAI,MAAM,gBAAgB,EAAE;IAE1B,IAAI;AACJ,QAAI;AACF,kBAAa,SAAS,KAAK;YACrB;AACN;;AAGF,QAAI,WAAW,aAAa,EAC1B;SAAI,CAAC,gBAAgB,IAAI,MAAM,KAAK,CAClC,MAAK,MAAM,QAAQ,EAAE;eAEd,WAAW,QAAQ,IAAI,MAAM,KAAK,SAAS,MAAM,CAC1D,SAAQ,KAAK;KACX,SAAS;KACT,cAAc,SAAS,WAAW,KAAK;KACvC,OAAO,WAAW;KAClB,QAAQ,WAAW;KACpB,CAAC;cAEK,MAAM,aAAa,CAC5B,MAAK,MAAM,QAAQ,EAAE;YACZ,MAAM,QAAQ,IAAI,MAAM,KAAK,SAAS,MAAM,EAAE;IACvD,IAAI;AACJ,QAAI;AACF,gBAAW,SAAS,KAAK;YACnB;AACN;;AAEF,YAAQ,KAAK;KACX,SAAS;KACT,cAAc,SAAS,WAAW,KAAK;KACvC,OAAO,SAAS;KAChB,QAAQ,SAAS;KAClB,CAAC;;;;AAKR,KAAI,WAAW,UAAU,CACvB,MAAK,WAAW,EAAE;AAGpB,QAAO;;;;;;;;;;;;AC/HT,SAAgB,mBAAmB,OAAkC;CACnE,MAAM,yBAAS,IAAI,KAA0B;AAE7C,MAAK,MAAM,QAAQ,OAAO;EACxB,MAAM,MAAM,GAAG,KAAK,OAAO,GAAG,KAAK;EACnC,MAAM,WAAW,OAAO,IAAI,IAAI;AAChC,MAAI,SACF,UAAS,KAAK,KAAK;MAEnB,QAAO,IAAI,KAAK,CAAC,KAAK,CAAC;;CAI3B,MAAM,SAAuB,EAAE;AAE/B,MAAK,MAAM,SAAS,OAAO,QAAQ,EAAE;AACnC,MAAI,MAAM,WAAW,EAAG;EAUxB,MAAM,CAAC,WAAW,GAAG,WAPN,CAAC,GAAG,MAAM,CAAC,MAAM,GAAG,MAAM;GACvC,MAAM,UAAU,EAAE,aAAa,MAAM,MAAM,IAAI,EAAE,EAAE;GACnD,MAAM,UAAU,EAAE,aAAa,MAAM,MAAM,IAAI,EAAE,EAAE;AACnD,OAAI,WAAW,OAAQ,QAAO,SAAS;AACvC,UAAO,EAAE,aAAa,SAAS,EAAE,aAAa;IAC9C;AAGF,SAAO,KAAK;GAAE;GAAW;GAAS,CAAC;;AAGrC,QAAO;;;;;;;;;;;;;;;;;;;;;;;AChBT,SAAgB,WAAW,SAA+B;CACxD,MAAM,QAAsB,EAAE;CAC9B,MAAM,QAAQ,QAAQ,MAAM,KAAK;CAGjC,IAAI,iBAAiB;AACrB,KAAI,QAAQ,WAAW,MAAM,EAAE;EAC7B,MAAM,aAAa,QAAQ,QAAQ,SAAS,EAAE;AAC9C,MAAI,eAAe,GACjB,kBAAiB,QAAQ,MAAM,GAAG,aAAa,EAAE,CAAC,MAAM,KAAK,CAAC,SAAS;;CAK3E,MAAM,aAAa;CAGnB,MAAM,WAAW;AAEjB,MAAK,IAAI,UAAU,GAAG,UAAU,MAAM,QAAQ,WAAW;EACvD,MAAM,OAAO,MAAM;EACnB,MAAM,aAAa,UAAU;EAC7B,MAAM,gBAAgB,UAAU;AAGhC,aAAW,YAAY;EACvB,IAAI;AACJ,UAAQ,QAAQ,WAAW,KAAK,KAAK,MAAM,MAAM;GAC/C,MAAM,UAAU,MAAM,OAAO;GAC7B,MAAM,QAAQ,MAAM;GAGpB,MAAM,UAAU,MAAM,QAAQ,IAAI;GAClC,MAAM,aAAa,YAAY,KAAK,QAAQ,MAAM,MAAM,GAAG,QAAQ;GACnE,MAAM,QAAQ,YAAY,KAAK,OAAO,MAAM,MAAM,UAAU,EAAE;GAG9D,MAAM,UAAU,WAAW,QAAQ,IAAI;GACvC,MAAM,MAAM,YAAY,KAAK,WAAW,MAAM,GAAG,WAAW,MAAM,GAAG,QAAQ,CAAC,MAAM;AAEpF,OAAI,CAAC,IAAK;AAEV,SAAM,KAAK;IACT;IACA,OAAO,OAAO,MAAM,IAAI;IACxB;IACA,SAAS,WAAW,CAAC;IACrB,UAAU;IACX,CAAC;;AAIJ,MAAI,CAAC,eAAe;AAClB,YAAS,YAAY;AACrB,WAAQ,QAAQ,SAAS,KAAK,KAAK,MAAM,MAAM;IAC7C,MAAM,UAAU,MAAM,OAAO;IAC7B,MAAM,cAAc,MAAM;IAC1B,IAAI,SAAS,MAAM,GAAI,MAAM;AAG7B,QAAI,uBAAuB,KAAK,OAAO,CAAE;AAGzC,QAAI,OAAO,WAAW,IAAI,CAAE;IAG5B,MAAM,UAAU,OAAO,QAAQ,IAAI;AACnC,QAAI,YAAY,GAAI,UAAS,OAAO,MAAM,GAAG,QAAQ;AAGrD,QAAI;AACF,cAAS,mBAAmB,OAAO;YAC7B;IAKR,MAAM,MAAM,OAAO,QAAQ,UAAU,GAAG,CAAC,MAAM;AAC/C,QAAI,CAAC,IAAK;AAEV,UAAM,KAAK;KACT;KACA,OAAO,eAAe;KACtB;KACA;KACA,UAAU;KACX,CAAC;;;;AAKR,QAAO;;;;;;;;;;;;;;;ACxGT,SAAgB,eAAe,OAA2C;CACxE,MAAM,wBAAQ,IAAI,KAAuB;AAEzC,MAAK,MAAM,QAAQ,OAAO;EACxB,MAAM,OAAO,SAAS,KAAK,cAAc,MAAM,CAAC,aAAa;EAC7D,MAAM,WAAW,MAAM,IAAI,KAAK;AAChC,MAAI,SACF,UAAS,KAAK,KAAK,aAAa;MAEhC,OAAM,IAAI,MAAM,CAAC,KAAK,aAAa,CAAC;;AAIxC,QAAO;;;;;;;;;;;;;;;;;;;;;;;;;ACJT,SAAgB,gBACd,KACA,WACA,YACe;AACf,KAAI,CAAC,IAAK,QAAO;AAGjB,KAAI,IAAI,SAAS,IAAI,EAAE;EACrB,MAAM,aAAa,UAAU,IAAI;EACjC,MAAM,eAAe,WAAW,SAAS,MAAM,GAAG,aAAa,aAAa;AAG5E,OAAK,MAAM,GAAG,UAAU,UACtB,MAAK,MAAM,KAAK,OAAO;AACrB,OAAI,MAAM,gBAAgB,MAAM,WAAY,QAAO;AACnD,OAAI,EAAE,aAAa,KAAK,aAAa,aAAa,CAAE,QAAO;;;CAQjE,MAAM,UAAU,SAAS,IAAI,CAC1B,QAAQ,UAAU,GAAG,CACrB,aAAa,CACb,MAAM;AAET,KAAI,CAAC,QAAS,QAAO;CAErB,MAAM,aAAa,UAAU,IAAI,QAAQ;AAEzC,KAAI,CAAC,cAAc,WAAW,WAAW,EACvC,QAAO;AAGT,KAAI,WAAW,WAAW,EACxB,QAAO,WAAW;CAIpB,MAAM,YAAY,QAAQ,WAAW;CAErC,IAAI,WAA0B;CAC9B,IAAI,gBAAgB;CACpB,IAAI,cAAc;AAElB,MAAK,MAAM,aAAa,YAAY;EAElC,MAAM,YAAY,mBAAmB,WADhB,QAAQ,UAAU,CACsB;EAC7D,MAAM,UAAU,UAAU;AAE1B,MACE,YAAY,iBACX,cAAc,iBAAiB,UAAU,aAC1C;AACA,mBAAgB;AAChB,iBAAc;AACd,cAAW;;;AAIf,QAAO;;;;;;;;AAST,SAAS,mBAAmB,GAAW,GAAmB;AACxD,KAAI,MAAM,OAAO,MAAM,IAAK,QAAO;CACnC,MAAM,SAAS,MAAM,MAAM,EAAE,GAAG,EAAE,MAAM,IAAI;CAC5C,MAAM,SAAS,MAAM,MAAM,EAAE,GAAG,EAAE,MAAM,IAAI;CAC5C,IAAI,QAAQ;CACZ,MAAM,MAAM,KAAK,IAAI,OAAO,QAAQ,OAAO,OAAO;AAClD,MAAK,IAAI,IAAI,GAAG,IAAI,KAAK,IACvB,KAAI,OAAO,OAAO,OAAO,GACvB;KAEA;AAGJ,QAAO;;;;;;;;;;;;;;;;;;;;ACxET,MAAM,oBAAoB;;;;;;;;;;;;;;;;;;;;;;;;;;;AA4B1B,eAAsB,WACpB,SACA,gBACA,WAC2B;CAC3B,MAAM,YAAY,KAAK,KAAK;CAE5B,MAAM,SAA2B;EAC/B,cAAc;EACd,eAAe;EACf,cAAc;EACd,iBAAiB;EACjB,gBAAgB;EAChB,gBAAgB;EAChB,cAAc;EACd,SAAS;EACV;CAGD,MAAM,WAAW,iBAAiB,UAAU;CAG5C,MAAM,cAAc,mBAAmB,SAAS;CAGhD,MAAM,YAAY,eAAe,SAAS;AAG1C,OAAM,kBAAkB;CACxB,IAAI,kBAAkB;AAEtB,MAAK,MAAM,SAAS,aAAa;AAC/B,MAAI,mBAAmB,mBAAmB;AACxC,SAAM,kBAAkB;AACxB,qBAAkB;;AAEpB;EAEA,MAAM,EAAE,cAAc;EAEtB,IAAI;AACJ,MAAI;AACF,aAAU,aAAa,UAAU,SAAS,OAAO;UAC3C;AACN,UAAO;AACP;;EAGF,MAAM,OAAO,WAAW,QAAQ;AAIhC,OADiB,MAAM,QAAQ,aAAa,UAAU,aAAa,GACrD,SAAS,MAAM;AAC3B,UAAO;AACP;;AAIF,QAAM,QAAQ,oBAAoB,gBAAgB,UAAU,aAAa;EAGzE,MAAM,SAAS,cAAc,QAAQ;EACrC,MAAM,YAAY,KAAK,KAAK;EAG5B,MAAM,aAAa,cAAc,KAAK,QAAQ;EAC9C,MAAM,QAAQ,aACV,WAAW,GAAI,MAAM,GACrB,SAAS,UAAU,cAAc,MAAM;EAG3C,MAAM,YAAwB,EAAE;AAChC,OAAK,IAAI,IAAI,GAAG,IAAI,OAAO,QAAQ,KAAK;GACtC,MAAM,QAAQ,OAAO;GACrB,MAAM,KAAK,QACT,gBACA,UAAU,cACV,GACA,MAAM,WACN,MAAM,QACP;AACD,aAAU,KAAK;IACb;IACA,WAAW;IACX,QAAQ;IACR,MAAM;IACN,MAAM,UAAU;IAChB,WAAW,MAAM;IACjB,SAAS,MAAM;IACf,MAAM,MAAM;IACZ,MAAM,MAAM;IACZ;IACD,CAAC;;AAGJ,MAAI,UAAU,SAAS,EACrB,OAAM,QAAQ,aAAa,UAAU;EAIvC,MAAM,eAA6B;GACjC,WAAW,UAAU;GACrB,OAAO,UAAU;GACjB,QAAQ,UAAU;GAClB;GACA;GACA,WAAW;GACZ;AACD,QAAM,QAAQ,gBAAgB,aAAa;AAE3C,SAAO;AACP,SAAO,iBAAiB,OAAO;;AAIjC,OAAM,kBAAkB;CAExB,MAAM,aAA8B,EAAE;AACtC,MAAK,MAAM,SAAS,YAClB,MAAK,MAAM,SAAS,MAAM,SAAS;AACjC,aAAW,KAAK;GACd,WAAW,MAAM;GACjB,eAAe,MAAM,UAAU;GAC/B,OAAO,MAAM;GACb,QAAQ,MAAM;GACf,CAAC;AACF,SAAO;;CAIX,MAAM,iBAAiB,IAAI,IAAI,YAAY,KAAK,MAAM,EAAE,UAAU,aAAa,CAAC;AAChF,MAAK,MAAM,aAAa,eACtB,OAAM,QAAQ,mBAAmB,UAAU;AAE7C,KAAI,WAAW,SAAS,EACtB,OAAM,QAAQ,mBAAmB,WAAW;AAI9C,OAAM,kBAAkB;CAExB,MAAM,cAAgC,EAAE;AACxC,MAAK,MAAM,CAAC,MAAM,UAAU,UAC1B,MAAK,MAAM,QAAQ,MACjB,aAAY,KAAK;EAAE;EAAM,WAAW;EAAM,CAAC;AAG/C,OAAM,QAAQ,iBAAiB,YAAY;AAG3C,OAAM,kBAAkB;CAExB,MAAM,WAA2B,EAAE;CACnC,MAAM,iBAA2B,EAAE;CAEnC,IAAI,iBAAiB;AACrB,MAAK,MAAM,SAAS,aAAa;AAC/B,MAAI,mBAAmB,sBAAsB,EAC3C,OAAM,kBAAkB;EAG1B,MAAM,EAAE,cAAc;AACtB,iBAAe,KAAK,UAAU,aAAa;EAE3C,IAAI;AACJ,MAAI;AACF,aAAU,aAAa,UAAU,SAAS,OAAO;UAC3C;AACN;;EAGF,MAAM,cAAc,WAAW,QAAQ;AACvC,OAAK,MAAM,QAAQ,aAAa;GAC9B,MAAM,SAAS,gBAAgB,KAAK,KAAK,WAAW,UAAU,aAAa;GAC3E,IAAI;AACJ,OAAI,KAAK,SACP,YAAW,KAAK,UAAU,aAAa;OAEvC,YAAW,KAAK,UAAU,UAAU;AAEtC,YAAS,KAAK;IACZ,YAAY,UAAU;IACtB,WAAW,KAAK;IAChB,YAAY;IACZ;IACA,YAAY,KAAK;IACjB,YAAY;IACb,CAAC;;;CAKN,MAAM,kBAAkB;AACxB,MAAK,IAAI,IAAI,GAAG,IAAI,eAAe,QAAQ,KAAK,iBAAiB;EAC/D,MAAM,eAAe,eAAe,MAAM,GAAG,IAAI,gBAAgB;EACjE,MAAM,aAAa,SAAS,QAAQ,MAAM,aAAa,SAAS,EAAE,WAAW,CAAC;AAC9E,QAAM,QAAQ,uBAAuB,cAAc,WAAW;AAC9D,QAAM,kBAAkB;;AAG1B,QAAO,iBAAiB,SAAS;AACjC,QAAO,iBAAiB,SAAS,QAAQ,MAAM,EAAE,eAAe,KAAK,CAAC;AAGtE,OAAM,kBAAkB;CAExB,MAAM,8BAAc,IAAI,KAAqB;CAC7C,MAAM,0BAAU,IAAI,KAAqB;CACzC,MAAM,6BAAa,IAAI,KAAqB;AAE5C,MAAK,MAAM,OAAO,UAAU;AAC1B,cAAY,IAAI,IAAI,aAAa,YAAY,IAAI,IAAI,WAAW,IAAI,KAAK,EAAE;AAC3E,MAAI,IAAI,eAAe,KACrB,SAAQ,IAAI,IAAI,aAAa,QAAQ,IAAI,IAAI,WAAW,IAAI,KAAK,EAAE;MAEnE,YAAW,IAAI,IAAI,aAAa,WAAW,IAAI,IAAI,WAAW,IAAI,KAAK,EAAE;;CAI7E,MAAM,aAAa,KAAK,KAAK;CAC7B,IAAI,cAAc;CAElB,MAAM,oBAAoB;AAC1B,MAAK,IAAI,IAAI,GAAG,IAAI,YAAY,QAAQ,KAAK,mBAAmB;EAE9D,MAAM,aADQ,YAAY,MAAM,GAAG,IAAI,kBAAkB,CACd,KAAK,UAAU;GACxD,MAAM,OAAO,MAAM,UAAU;GAC7B,MAAM,UAAU,WAAW,IAAI,KAAK,IAAI;GACxC,MAAM,WAAW,YAAY,IAAI,KAAK,IAAI;GAC1C,MAAM,OAAO,QAAQ,IAAI,KAAK,IAAI;GAClC,MAAM,WAAW,YAAY;AAC7B,OAAI,SAAU;AACd,UAAO;IACL,WAAW;IACX,cAAc;IACd,eAAe;IACf,eAAe;IACf;IACA;IACD;IACD;AACF,QAAM,QAAQ,kBAAkB,WAAW;AAC3C,QAAM,kBAAkB;;AAG1B,QAAO,eAAe;AACtB,QAAO,UAAU,KAAK,KAAK,GAAG;AAE9B,QAAO"}
+{"version":3,"file":"vault-indexer-vNnL3Azb.mjs","names":[],"sources":["../src/memory/vault/walk.ts","../src/memory/vault/deduplicate.ts","../src/memory/vault/parse-links.ts","../src/memory/vault/name-index.ts","../src/memory/vault/resolve.ts","../src/memory/vault/indexer.ts"],"sourcesContent":["/**\n * Vault directory walker — follows symlinks with cycle detection.\n */\n\nimport { statSync, readdirSync, existsSync } from \"node:fs\";\nimport { join, relative } from \"node:path\";\nimport type { VaultFile } from \"./types.js\";\n\n/** Maximum number of .md files to collect from a vault. */\nconst VAULT_MAX_FILES = 10_000;\n\n/** Maximum recursion depth for vault directory walks. */\nconst VAULT_MAX_DEPTH = 10;\n\n/**\n * Directories to always skip, at any depth, during vault walks.\n * Includes standard build/VCS noise plus Obsidian-specific directories.\n */\nconst VAULT_SKIP_DIRS = new Set([\n  // Version control\n  \".git\",\n  // Dependency directories (any language)\n  \"node_modules\",\n  \"vendor\",\n  \"Pods\",\n  // Build / compile output\n  \"dist\",\n  \"build\",\n  \"out\",\n  \"DerivedData\",\n  \".next\",\n  // Python virtual environments and caches\n  \".venv\",\n  \"venv\",\n  \"__pycache__\",\n  // General caches\n  \".cache\",\n  \".bun\",\n  // Obsidian internals\n  \".obsidian\",\n  \".trash\",\n]);\n\n/**\n * Recursively collect all .md files under a vault root, following symlinks.\n *\n * Symlink-following behaviour:\n *  - Symbolic links to files: followed if the target is a .md file\n *  - Symbolic links to directories: followed with cycle detection via inode\n *\n * Cycle detection is based on the real inode of each visited directory.\n * Using the real stat (not lstat) ensures that symlinked dirs resolve to\n * their actual inode, preventing infinite loops.\n *\n * @param vaultRoot  Absolute path to the vault root directory.\n * @param opts       Optional overrides for maxFiles and maxDepth.\n */\nexport function walkVaultMdFiles(\n  vaultRoot: string,\n  opts?: { maxFiles?: number; maxDepth?: number },\n): VaultFile[] {\n  const maxFiles = opts?.maxFiles ?? VAULT_MAX_FILES;\n  const maxDepth = opts?.maxDepth ?? VAULT_MAX_DEPTH;\n\n  const results: VaultFile[] = [];\n  const visitedDirs = new Set<string>();\n\n  function walk(dir: string, depth: number): void {\n    if (results.length >= maxFiles) return;\n    if (depth > maxDepth) return;\n\n    // Get the real inode of this directory (follows symlinks on the dir itself)\n    let dirStat: ReturnType<typeof statSync>;\n    try {\n      dirStat = statSync(dir);\n    } catch {\n      return; // Unreadable or broken symlink — skip\n    }\n\n    const dirKey = `${dirStat.dev}:${dirStat.ino}`;\n    if (visitedDirs.has(dirKey)) return; // Cycle detected\n    visitedDirs.add(dirKey);\n\n    let entries: import(\"node:fs\").Dirent<string>[];\n    try {\n      entries = readdirSync(dir, { withFileTypes: true, encoding: \"utf8\" });\n    } catch {\n      return; // Unreadable directory — skip\n    }\n\n    for (const entry of entries) {\n      if (results.length >= maxFiles) break;\n      if (VAULT_SKIP_DIRS.has(entry.name)) continue;\n\n      const full = join(dir, entry.name);\n\n      if (entry.isSymbolicLink()) {\n        // Follow the symlink — resolve to real target\n        let targetStat: ReturnType<typeof statSync>;\n        try {\n          targetStat = statSync(full); // statSync follows symlinks\n        } catch {\n          continue; // Broken symlink — skip\n        }\n\n        if (targetStat.isDirectory()) {\n          if (!VAULT_SKIP_DIRS.has(entry.name)) {\n            walk(full, depth + 1);\n          }\n        } else if (targetStat.isFile() && entry.name.endsWith(\".md\")) {\n          results.push({\n            absPath: full,\n            vaultRelPath: relative(vaultRoot, full),\n            inode: targetStat.ino,\n            device: targetStat.dev,\n          });\n        }\n      } else if (entry.isDirectory()) {\n        walk(full, depth + 1);\n      } else if (entry.isFile() && entry.name.endsWith(\".md\")) {\n        let fileStat: ReturnType<typeof statSync>;\n        try {\n          fileStat = statSync(full);\n        } catch {\n          continue;\n        }\n        results.push({\n          absPath: full,\n          vaultRelPath: relative(vaultRoot, full),\n          inode: fileStat.ino,\n          device: fileStat.dev,\n        });\n      }\n    }\n  }\n\n  if (existsSync(vaultRoot)) {\n    walk(vaultRoot, 0);\n  }\n\n  return results;\n}\n","/**\n * Inode-based deduplication for vault files.\n */\n\nimport type { VaultFile, InodeGroup } from \"./types.js\";\n\n/**\n * Group vault files by inode identity (device + inode).\n *\n * Within each group, the canonical file is chosen as the one with the\n * fewest path separators (shallowest), breaking ties by shortest string.\n * All other group members become aliases.\n */\nexport function deduplicateByInode(files: VaultFile[]): InodeGroup[] {\n  const groups = new Map<string, VaultFile[]>();\n\n  for (const file of files) {\n    const key = `${file.device}:${file.inode}`;\n    const existing = groups.get(key);\n    if (existing) {\n      existing.push(file);\n    } else {\n      groups.set(key, [file]);\n    }\n  }\n\n  const result: InodeGroup[] = [];\n\n  for (const group of groups.values()) {\n    if (group.length === 0) continue;\n\n    // Sort: fewest path separators first, then shortest string\n    const sorted = [...group].sort((a, b) => {\n      const aDepth = (a.vaultRelPath.match(/\\//g) ?? []).length;\n      const bDepth = (b.vaultRelPath.match(/\\//g) ?? []).length;\n      if (aDepth !== bDepth) return aDepth - bDepth;\n      return a.vaultRelPath.length - b.vaultRelPath.length;\n    });\n\n    const [canonical, ...aliases] = sorted as [VaultFile, ...VaultFile[]];\n    result.push({ canonical, aliases });\n  }\n\n  return result;\n}\n","/**\n * Markdown link parser for vault files.\n *\n * Handles wikilinks ([[target]]), markdown links ([text](path)),\n * embeds (![[target]]), and frontmatter wikilinks.\n */\n\nimport type { ParsedLink } from \"./types.js\";\n\n/**\n * Parse all links from markdown content.\n *\n * Handles:\n *  - Standard wikilinks: [[Target Note]]\n *  - Aliased wikilinks: [[Target Note|Display Text]]\n *  - Heading anchors: [[Target Note#Heading]] (stripped for resolution)\n *  - Embeds: ![[Target Note]]\n *  - Frontmatter wikilinks (YAML between --- delimiters)\n *  - Markdown links: [text](path/to/note.md)\n *  - Markdown embeds: ![alt](image.png)\n *\n * External URLs (http://, https://, mailto:, etc.) are excluded — only\n * relative paths are treated as vault links.\n *\n * @param content  Raw markdown file content.\n * @returns        Array of parsed links in document order.\n */\nexport function parseLinks(content: string): ParsedLink[] {\n  const links: ParsedLink[] = [];\n  const lines = content.split(\"\\n\");\n\n  // Determine frontmatter range (YAML between opening and closing ---)\n  let frontmatterEnd = 0;\n  if (content.startsWith(\"---\")) {\n    const closingIdx = content.indexOf(\"\\n---\", 3);\n    if (closingIdx !== -1) {\n      frontmatterEnd = content.slice(0, closingIdx + 4).split(\"\\n\").length - 1;\n    }\n  }\n\n  // Regex for [[wikilinks]] and ![[embeds]]\n  const wikilinkRe = /(!?)\\[\\[([^\\]]+?)\\]\\]/g;\n\n  // Regex for markdown links [text](target) and embeds ![alt](target)\n  const mdLinkRe = /(!)?\\[([^\\]]*)\\]\\(([^)]+)\\)/g;\n\n  for (let lineIdx = 0; lineIdx < lines.length; lineIdx++) {\n    const line = lines[lineIdx]!;\n    const lineNumber = lineIdx + 1; // 1-indexed\n    const isFrontmatter = lineIdx < frontmatterEnd;\n\n    // --- Wikilinks ---\n    wikilinkRe.lastIndex = 0;\n    let match: RegExpExecArray | null;\n    while ((match = wikilinkRe.exec(line)) !== null) {\n      const isEmbed = match[1] === \"!\";\n      const inner = match[2]!;\n\n      // Split on first | for alias\n      const pipeIdx = inner.indexOf(\"|\");\n      const beforePipe = pipeIdx === -1 ? inner : inner.slice(0, pipeIdx);\n      const alias = pipeIdx === -1 ? null : inner.slice(pipeIdx + 1);\n\n      // Strip heading anchor (everything after #)\n      const hashIdx = beforePipe.indexOf(\"#\");\n      const raw = hashIdx === -1 ? beforePipe.trim() : beforePipe.slice(0, hashIdx).trim();\n\n      if (!raw) continue; // Skip links with empty targets (e.g. [[#Heading]])\n\n      links.push({\n        raw,\n        alias: alias?.trim() ?? null,\n        lineNumber,\n        isEmbed: isEmbed && !isFrontmatter,\n        isMdLink: false,\n      });\n    }\n\n    // --- Markdown links --- (skip inside frontmatter)\n    if (!isFrontmatter) {\n      mdLinkRe.lastIndex = 0;\n      while ((match = mdLinkRe.exec(line)) !== null) {\n        const isEmbed = match[1] === \"!\";\n        const displayText = match[2]!;\n        let target = match[3]!.trim();\n\n        // Skip external URLs\n        if (/^[a-z][a-z0-9+.-]*:/i.test(target)) continue;\n\n        // Skip pure anchor links (#heading)\n        if (target.startsWith(\"#\")) continue;\n\n        // Strip heading anchor from target\n        const hashIdx = target.indexOf(\"#\");\n        if (hashIdx !== -1) target = target.slice(0, hashIdx);\n\n        // URL-decode (Obsidian encodes spaces as %20 in md links)\n        try {\n          target = decodeURIComponent(target);\n        } catch {\n          // Malformed encoding — use as-is\n        }\n\n        // Strip .md extension for resolution (resolveWikilink adds it back)\n        const raw = target.replace(/\\.md$/i, \"\").trim();\n        if (!raw) continue;\n\n        links.push({\n          raw,\n          alias: displayText || null,\n          lineNumber,\n          isEmbed,\n          isMdLink: true,\n        });\n      }\n    }\n  }\n\n  return links;\n}\n\n/** @deprecated Use {@link parseLinks} instead. */\nexport const parseWikilinks = parseLinks;\n","/**\n * Name index builder for Obsidian wikilink resolution.\n */\n\nimport { basename } from \"node:path\";\nimport type { VaultFile } from \"./types.js\";\n\n/**\n * Build a name lookup index for Obsidian wikilink resolution.\n *\n * Maps lowercase filename (without .md extension) to all vault-relative paths\n * that share that name. Includes both canonical paths and alias paths so that\n * wikilinks resolve regardless of which path the file is accessed through.\n */\nexport function buildNameIndex(files: VaultFile[]): Map<string, string[]> {\n  const index = new Map<string, string[]>();\n\n  for (const file of files) {\n    const name = basename(file.vaultRelPath, \".md\").toLowerCase();\n    const existing = index.get(name);\n    if (existing) {\n      existing.push(file.vaultRelPath);\n    } else {\n      index.set(name, [file.vaultRelPath]);\n    }\n  }\n\n  return index;\n}\n","/**\n * Wikilink resolver using Obsidian's shortest-match algorithm.\n */\n\nimport { normalize, basename, dirname } from \"node:path\";\n\n/**\n * Resolve a wikilink target to a vault-relative path using Obsidian's rules.\n *\n * Resolution algorithm:\n *  1. If raw contains \"/\", attempt exact path match (with and without .md).\n *  2. Normalize: lowercase the raw target, strip .md extension.\n *  3. Look up in the name index (all files with that basename).\n *  4. If exactly one match, return it.\n *  5. If multiple matches, pick the one closest to the source file\n *     (longest common directory prefix, then shortest overall path).\n *  6. If no matches, return null (dead link).\n *\n * @param raw         The raw link target (heading-stripped, pipe-stripped).\n * @param nameIndex   Map from lowercase basename-without-ext to vault paths.\n * @param sourcePath  Vault-relative path of the file containing the link.\n * @returns           Vault-relative path of the resolved target, or null.\n */\nexport function resolveWikilink(\n  raw: string,\n  nameIndex: Map<string, string[]>,\n  sourcePath: string,\n): string | null {\n  if (!raw) return null;\n\n  // Case 1: path contains \"/\" — try exact match with and without .md\n  if (raw.includes(\"/\")) {\n    const normalized = normalize(raw);\n    const normalizedMd = normalized.endsWith(\".md\") ? normalized : normalized + \".md\";\n\n    // Check if any indexed path matches (case-insensitive for macOS compatibility)\n    for (const [, paths] of nameIndex) {\n      for (const p of paths) {\n        if (p === normalizedMd || p === normalized) return p;\n        if (p.toLowerCase() === normalizedMd.toLowerCase()) return p;\n      }\n    }\n    // Fall through to name lookup in case the path prefix was wrong\n  }\n\n  // Normalize the raw target for name lookup.\n  // Use the basename only — Obsidian resolves by filename, not full path.\n  const rawBase = basename(raw)\n    .replace(/\\.md$/i, \"\")\n    .toLowerCase()\n    .trim();\n\n  if (!rawBase) return null;\n\n  const candidates = nameIndex.get(rawBase);\n\n  if (!candidates || candidates.length === 0) {\n    return null; // Dead link\n  }\n\n  if (candidates.length === 1) {\n    return candidates[0]!;\n  }\n\n  // Multiple matches — pick the one closest to the source file\n  const sourceDir = dirname(sourcePath);\n\n  let bestPath: string | null = null;\n  let bestPrefixLen = -1;\n  let bestPathLen = Infinity;\n\n  for (const candidate of candidates) {\n    const candidateDir = dirname(candidate);\n    const prefixLen = commonPrefixLength(sourceDir, candidateDir);\n    const pathLen = candidate.length;\n\n    if (\n      prefixLen > bestPrefixLen ||\n      (prefixLen === bestPrefixLen && pathLen < bestPathLen)\n    ) {\n      bestPrefixLen = prefixLen;\n      bestPathLen = pathLen;\n      bestPath = candidate;\n    }\n  }\n\n  return bestPath;\n}\n\n/**\n * Compute the length of the common prefix between two directory paths,\n * measured in path segments (not raw characters).\n *\n * Example: \"a/b/c\" and \"a/b/d\" → 2 (common: \"a\", \"b\")\n */\nfunction commonPrefixLength(a: string, b: string): number {\n  if (a === \".\" && b === \".\") return 0;\n  const aParts = a === \".\" ? [] : a.split(\"/\");\n  const bParts = b === \".\" ? [] : b.split(\"/\");\n  let count = 0;\n  const len = Math.min(aParts.length, bParts.length);\n  for (let i = 0; i < len; i++) {\n    if (aParts[i] === bParts[i]) {\n      count++;\n    } else {\n      break;\n    }\n  }\n  return count;\n}\n","/**\n * Main vault indexing orchestrator.\n *\n * Indexes an entire Obsidian vault (or any markdown knowledge base), following\n * symlinks, deduplicating files by inode, parsing wikilinks, and computing\n * per-file health metrics (orphan detection, dead links).\n *\n * Key differences from the project indexer (indexer/sync.ts):\n *  - Follows symbolic links (project indexer skips them)\n *  - Deduplicates files with the same inode (same content reachable via multiple paths)\n *  - Parses [[wikilinks]] and builds a directed link graph\n *  - Resolves wikilinks using Obsidian's shortest-match algorithm\n *  - Computes health metrics per file: inbound/outbound link counts, dead links, orphans\n */\n\nimport { readFileSync } from \"node:fs\";\nimport { basename } from \"node:path\";\nimport type {\n  StorageBackend,\n  VaultFileRow,\n  VaultAliasRow,\n  VaultLinkRow,\n  VaultHealthRow,\n  VaultNameEntry,\n  ChunkRow,\n} from \"../../storage/interface.js\";\nimport { chunkMarkdown } from \"../chunker.js\";\nimport { sha256File, chunkId, yieldToEventLoop } from \"../indexer/helpers.js\";\nimport { walkVaultMdFiles } from \"./walk.js\";\nimport { deduplicateByInode } from \"./deduplicate.js\";\nimport { parseLinks } from \"./parse-links.js\";\nimport { buildNameIndex } from \"./name-index.js\";\nimport { resolveWikilink } from \"./resolve.js\";\nimport type { VaultIndexResult } from \"./types.js\";\n\n/** Number of files to process before yielding to the event loop. */\nconst VAULT_YIELD_EVERY = 1;\n\n/**\n * Index an entire Obsidian vault (or markdown knowledge base) using the\n * async StorageBackend interface.\n *\n * Steps:\n *  1. Walk vault root, following symlinks.\n *  2. Deduplicate by inode — each unique file is indexed once.\n *  3. Build a name index for wikilink resolution.\n *  4. For each canonical file:\n *     a. SHA-256 hash for change detection — skip unchanged files.\n *     b. Read content, chunk with chunkMarkdown().\n *     c. Insert chunks into backend (memory_chunks and memory_fts).\n *     d. Upsert vault_files row.\n *  5. Record aliases in vault_aliases.\n *  6. Rebuild vault_name_index table.\n *  7. Rebuild vault_links:\n *     a. Parse [[wikilinks]] from each canonical file.\n *     b. Resolve each link with resolveWikilink().\n *     c. Insert into vault_links.\n *  8. Compute and upsert health metrics (vault_health).\n *  9. Return statistics.\n *\n * @param backend         StorageBackend to write to.\n * @param vaultProjectId  Registry project ID for the vault \"project\".\n * @param vaultRoot       Absolute path to the vault root directory.\n */\nexport async function indexVault(\n  backend: StorageBackend,\n  vaultProjectId: number,\n  vaultRoot: string,\n): Promise<VaultIndexResult> {\n  const startTime = Date.now();\n\n  const result: VaultIndexResult = {\n    filesIndexed: 0,\n    chunksCreated: 0,\n    filesSkipped: 0,\n    aliasesRecorded: 0,\n    linksExtracted: 0,\n    deadLinksFound: 0,\n    orphansFound: 0,\n    elapsed: 0,\n  };\n\n  // Step 1: Walk vault, collecting all .md files (follows symlinks)\n  const allFiles = walkVaultMdFiles(vaultRoot);\n\n  // Step 2: Deduplicate by inode\n  const inodeGroups = deduplicateByInode(allFiles);\n\n  // Step 3: Build name index (from all files including aliases, for resolution)\n  const nameIndex = buildNameIndex(allFiles);\n\n  // Step 4: Index each canonical file\n  await yieldToEventLoop();\n  let filesSinceYield = 0;\n\n  for (const group of inodeGroups) {\n    if (filesSinceYield >= VAULT_YIELD_EVERY) {\n      await yieldToEventLoop();\n      filesSinceYield = 0;\n    }\n    filesSinceYield++;\n\n    const { canonical } = group;\n\n    let content: string;\n    try {\n      content = readFileSync(canonical.absPath, \"utf8\");\n    } catch {\n      result.filesSkipped++;\n      continue;\n    }\n\n    const hash = sha256File(content);\n\n    // Change detection: skip if hash is unchanged\n    const existing = await backend.getVaultFile(canonical.vaultRelPath);\n    if (existing?.hash === hash) {\n      result.filesSkipped++;\n      continue;\n    }\n\n    // Delete old chunks for this vault path\n    await backend.deleteChunksForFile(vaultProjectId, canonical.vaultRelPath);\n\n    // Chunk the content\n    const chunks = chunkMarkdown(content);\n    const updatedAt = Date.now();\n\n    // Extract title from first H1 heading or filename\n    const titleMatch = /^#\\s+(.+)$/m.exec(content);\n    const title = titleMatch\n      ? titleMatch[1]!.trim()\n      : basename(canonical.vaultRelPath, \".md\");\n\n    // Build chunk rows\n    const chunkRows: ChunkRow[] = [];\n    for (let i = 0; i < chunks.length; i++) {\n      const chunk = chunks[i]!;\n      const id = chunkId(\n        vaultProjectId,\n        canonical.vaultRelPath,\n        i,\n        chunk.startLine,\n        chunk.endLine,\n      );\n      chunkRows.push({\n        id,\n        projectId: vaultProjectId,\n        source: \"vault\",\n        tier: \"topic\",\n        path: canonical.vaultRelPath,\n        startLine: chunk.startLine,\n        endLine: chunk.endLine,\n        hash: chunk.hash,\n        text: chunk.text,\n        updatedAt,\n      });\n    }\n\n    if (chunkRows.length > 0) {\n      await backend.insertChunks(chunkRows);\n    }\n\n    // Upsert vault file record\n    const vaultFileRow: VaultFileRow = {\n      vaultPath: canonical.vaultRelPath,\n      inode: canonical.inode,\n      device: canonical.device,\n      hash,\n      title,\n      indexedAt: updatedAt,\n    };\n    await backend.upsertVaultFile(vaultFileRow);\n\n    result.filesIndexed++;\n    result.chunksCreated += chunks.length;\n  }\n\n  // Step 5: Record aliases in vault_aliases\n  await yieldToEventLoop();\n\n  const allAliases: VaultAliasRow[] = [];\n  for (const group of inodeGroups) {\n    for (const alias of group.aliases) {\n      allAliases.push({\n        vaultPath: alias.vaultRelPath,\n        canonicalPath: group.canonical.vaultRelPath,\n        inode: alias.inode,\n        device: alias.device,\n      });\n      result.aliasesRecorded++;\n    }\n  }\n\n  const canonicalPaths = new Set(inodeGroups.map((g) => g.canonical.vaultRelPath));\n  for (const canonPath of canonicalPaths) {\n    await backend.deleteVaultAliases(canonPath);\n  }\n  if (allAliases.length > 0) {\n    await backend.upsertVaultAliases(allAliases);\n  }\n\n  // Step 6: Rebuild vault_name_index\n  await yieldToEventLoop();\n\n  const nameEntries: VaultNameEntry[] = [];\n  for (const [name, paths] of nameIndex) {\n    for (const path of paths) {\n      nameEntries.push({ name, vaultPath: path });\n    }\n  }\n  await backend.replaceNameIndex(nameEntries);\n\n  // Step 7: Rebuild vault_links\n  await yieldToEventLoop();\n\n  const linkRows: VaultLinkRow[] = [];\n  const allSourcePaths: string[] = [];\n\n  let linkParseYield = 0;\n  for (const group of inodeGroups) {\n    if (linkParseYield++ % VAULT_YIELD_EVERY === 0) {\n      await yieldToEventLoop();\n    }\n\n    const { canonical } = group;\n    allSourcePaths.push(canonical.vaultRelPath);\n\n    let content: string;\n    try {\n      content = readFileSync(canonical.absPath, \"utf8\");\n    } catch {\n      continue;\n    }\n\n    const parsedLinks = parseLinks(content);\n    for (const link of parsedLinks) {\n      const target = resolveWikilink(link.raw, nameIndex, canonical.vaultRelPath);\n      let linkType: string;\n      if (link.isMdLink) {\n        linkType = link.isEmbed ? \"md-embed\" : \"md-link\";\n      } else {\n        linkType = link.isEmbed ? \"embed\" : \"wikilink\";\n      }\n      linkRows.push({\n        sourcePath: canonical.vaultRelPath,\n        targetRaw: link.raw,\n        targetPath: target,\n        linkType,\n        lineNumber: link.lineNumber,\n        confidence: \"EXTRACTED\",\n      });\n    }\n  }\n\n  // Replace all links for all sources in batches of 500\n  const LINK_BATCH_SIZE = 500;\n  for (let i = 0; i < allSourcePaths.length; i += LINK_BATCH_SIZE) {\n    const batchSources = allSourcePaths.slice(i, i + LINK_BATCH_SIZE);\n    const batchLinks = linkRows.filter((r) => batchSources.includes(r.sourcePath));\n    await backend.replaceLinksForSources(batchSources, batchLinks);\n    await yieldToEventLoop();\n  }\n\n  result.linksExtracted = linkRows.length;\n  result.deadLinksFound = linkRows.filter((r) => r.targetPath === null).length;\n\n  // Step 8: Compute and upsert vault_health metrics\n  await yieldToEventLoop();\n\n  const outboundMap = new Map<string, number>();\n  const deadMap = new Map<string, number>();\n  const inboundMap = new Map<string, number>();\n\n  for (const row of linkRows) {\n    outboundMap.set(row.sourcePath, (outboundMap.get(row.sourcePath) ?? 0) + 1);\n    if (row.targetPath === null) {\n      deadMap.set(row.sourcePath, (deadMap.get(row.sourcePath) ?? 0) + 1);\n    } else {\n      inboundMap.set(row.targetPath, (inboundMap.get(row.targetPath) ?? 0) + 1);\n    }\n  }\n\n  const computedAt = Date.now();\n  let orphanCount = 0;\n\n  const HEALTH_BATCH_SIZE = 500;\n  for (let i = 0; i < inodeGroups.length; i += HEALTH_BATCH_SIZE) {\n    const batch = inodeGroups.slice(i, i + HEALTH_BATCH_SIZE);\n    const healthRows: VaultHealthRow[] = batch.map((group) => {\n      const path = group.canonical.vaultRelPath;\n      const inbound = inboundMap.get(path) ?? 0;\n      const outbound = outboundMap.get(path) ?? 0;\n      const dead = deadMap.get(path) ?? 0;\n      const isOrphan = inbound === 0;\n      if (isOrphan) orphanCount++;\n      return {\n        vaultPath: path,\n        inboundCount: inbound,\n        outboundCount: outbound,\n        deadLinkCount: dead,\n        isOrphan,\n        computedAt,\n      };\n    });\n    await backend.upsertVaultHealth(healthRows);\n    await yieldToEventLoop();\n  }\n\n  result.orphansFound = orphanCount;\n  result.elapsed = Date.now() - startTime;\n\n  return result;\n}\n"],"mappings":";;;;;;;;;AASA,MAAM,kBAAkB;;AAGxB,MAAM,kBAAkB;;;;;AAMxB,MAAM,kBAAkB,IAAI,IAAI;CAE9B;CAEA;CACA;CACA;CAEA;CACA;CACA;CACA;CACA;CAEA;CACA;CACA;CAEA;CACA;CAEA;CACA;CACD,CAAC;;;;;;;;;;;;;;;AAgBF,SAAgB,iBACd,WACA,MACa;CACb,MAAM,WAAW,MAAM,YAAY;CACnC,MAAM,WAAW,MAAM,YAAY;CAEnC,MAAM,UAAuB,EAAE;CAC/B,MAAM,8BAAc,IAAI,KAAa;CAErC,SAAS,KAAK,KAAa,OAAqB;AAC9C,MAAI,QAAQ,UAAU,SAAU;AAChC,MAAI,QAAQ,SAAU;EAGtB,IAAI;AACJ,MAAI;AACF,aAAU,SAAS,IAAI;UACjB;AACN;;EAGF,MAAM,SAAS,GAAG,QAAQ,IAAI,GAAG,QAAQ;AACzC,MAAI,YAAY,IAAI,OAAO,CAAE;AAC7B,cAAY,IAAI,OAAO;EAEvB,IAAI;AACJ,MAAI;AACF,aAAU,YAAY,KAAK;IAAE,eAAe;IAAM,UAAU;IAAQ,CAAC;UAC/D;AACN;;AAGF,OAAK,MAAM,SAAS,SAAS;AAC3B,OAAI,QAAQ,UAAU,SAAU;AAChC,OAAI,gBAAgB,IAAI,MAAM,KAAK,CAAE;GAErC,MAAM,OAAO,KAAK,KAAK,MAAM,KAAK;AAElC,OAAI,MAAM,gBAAgB,EAAE;IAE1B,IAAI;AACJ,QAAI;AACF,kBAAa,SAAS,KAAK;YACrB;AACN;;AAGF,QAAI,WAAW,aAAa,EAC1B;SAAI,CAAC,gBAAgB,IAAI,MAAM,KAAK,CAClC,MAAK,MAAM,QAAQ,EAAE;eAEd,WAAW,QAAQ,IAAI,MAAM,KAAK,SAAS,MAAM,CAC1D,SAAQ,KAAK;KACX,SAAS;KACT,cAAc,SAAS,WAAW,KAAK;KACvC,OAAO,WAAW;KAClB,QAAQ,WAAW;KACpB,CAAC;cAEK,MAAM,aAAa,CAC5B,MAAK,MAAM,QAAQ,EAAE;YACZ,MAAM,QAAQ,IAAI,MAAM,KAAK,SAAS,MAAM,EAAE;IACvD,IAAI;AACJ,QAAI;AACF,gBAAW,SAAS,KAAK;YACnB;AACN;;AAEF,YAAQ,KAAK;KACX,SAAS;KACT,cAAc,SAAS,WAAW,KAAK;KACvC,OAAO,SAAS;KAChB,QAAQ,SAAS;KAClB,CAAC;;;;AAKR,KAAI,WAAW,UAAU,CACvB,MAAK,WAAW,EAAE;AAGpB,QAAO;;;;;;;;;;;;AC/HT,SAAgB,mBAAmB,OAAkC;CACnE,MAAM,yBAAS,IAAI,KAA0B;AAE7C,MAAK,MAAM,QAAQ,OAAO;EACxB,MAAM,MAAM,GAAG,KAAK,OAAO,GAAG,KAAK;EACnC,MAAM,WAAW,OAAO,IAAI,IAAI;AAChC,MAAI,SACF,UAAS,KAAK,KAAK;MAEnB,QAAO,IAAI,KAAK,CAAC,KAAK,CAAC;;CAI3B,MAAM,SAAuB,EAAE;AAE/B,MAAK,MAAM,SAAS,OAAO,QAAQ,EAAE;AACnC,MAAI,MAAM,WAAW,EAAG;EAUxB,MAAM,CAAC,WAAW,GAAG,WAPN,CAAC,GAAG,MAAM,CAAC,MAAM,GAAG,MAAM;GACvC,MAAM,UAAU,EAAE,aAAa,MAAM,MAAM,IAAI,EAAE,EAAE;GACnD,MAAM,UAAU,EAAE,aAAa,MAAM,MAAM,IAAI,EAAE,EAAE;AACnD,OAAI,WAAW,OAAQ,QAAO,SAAS;AACvC,UAAO,EAAE,aAAa,SAAS,EAAE,aAAa;IAC9C;AAGF,SAAO,KAAK;GAAE;GAAW;GAAS,CAAC;;AAGrC,QAAO;;;;;;;;;;;;;;;;;;;;;;;AChBT,SAAgB,WAAW,SAA+B;CACxD,MAAM,QAAsB,EAAE;CAC9B,MAAM,QAAQ,QAAQ,MAAM,KAAK;CAGjC,IAAI,iBAAiB;AACrB,KAAI,QAAQ,WAAW,MAAM,EAAE;EAC7B,MAAM,aAAa,QAAQ,QAAQ,SAAS,EAAE;AAC9C,MAAI,eAAe,GACjB,kBAAiB,QAAQ,MAAM,GAAG,aAAa,EAAE,CAAC,MAAM,KAAK,CAAC,SAAS;;CAK3E,MAAM,aAAa;CAGnB,MAAM,WAAW;AAEjB,MAAK,IAAI,UAAU,GAAG,UAAU,MAAM,QAAQ,WAAW;EACvD,MAAM,OAAO,MAAM;EACnB,MAAM,aAAa,UAAU;EAC7B,MAAM,gBAAgB,UAAU;AAGhC,aAAW,YAAY;EACvB,IAAI;AACJ,UAAQ,QAAQ,WAAW,KAAK,KAAK,MAAM,MAAM;GAC/C,MAAM,UAAU,MAAM,OAAO;GAC7B,MAAM,QAAQ,MAAM;GAGpB,MAAM,UAAU,MAAM,QAAQ,IAAI;GAClC,MAAM,aAAa,YAAY,KAAK,QAAQ,MAAM,MAAM,GAAG,QAAQ;GACnE,MAAM,QAAQ,YAAY,KAAK,OAAO,MAAM,MAAM,UAAU,EAAE;GAG9D,MAAM,UAAU,WAAW,QAAQ,IAAI;GACvC,MAAM,MAAM,YAAY,KAAK,WAAW,MAAM,GAAG,WAAW,MAAM,GAAG,QAAQ,CAAC,MAAM;AAEpF,OAAI,CAAC,IAAK;AAEV,SAAM,KAAK;IACT;IACA,OAAO,OAAO,MAAM,IAAI;IACxB;IACA,SAAS,WAAW,CAAC;IACrB,UAAU;IACX,CAAC;;AAIJ,MAAI,CAAC,eAAe;AAClB,YAAS,YAAY;AACrB,WAAQ,QAAQ,SAAS,KAAK,KAAK,MAAM,MAAM;IAC7C,MAAM,UAAU,MAAM,OAAO;IAC7B,MAAM,cAAc,MAAM;IAC1B,IAAI,SAAS,MAAM,GAAI,MAAM;AAG7B,QAAI,uBAAuB,KAAK,OAAO,CAAE;AAGzC,QAAI,OAAO,WAAW,IAAI,CAAE;IAG5B,MAAM,UAAU,OAAO,QAAQ,IAAI;AACnC,QAAI,YAAY,GAAI,UAAS,OAAO,MAAM,GAAG,QAAQ;AAGrD,QAAI;AACF,cAAS,mBAAmB,OAAO;YAC7B;IAKR,MAAM,MAAM,OAAO,QAAQ,UAAU,GAAG,CAAC,MAAM;AAC/C,QAAI,CAAC,IAAK;AAEV,UAAM,KAAK;KACT;KACA,OAAO,eAAe;KACtB;KACA;KACA,UAAU;KACX,CAAC;;;;AAKR,QAAO;;;;;;;;;;;;;;;ACxGT,SAAgB,eAAe,OAA2C;CACxE,MAAM,wBAAQ,IAAI,KAAuB;AAEzC,MAAK,MAAM,QAAQ,OAAO;EACxB,MAAM,OAAO,SAAS,KAAK,cAAc,MAAM,CAAC,aAAa;EAC7D,MAAM,WAAW,MAAM,IAAI,KAAK;AAChC,MAAI,SACF,UAAS,KAAK,KAAK,aAAa;MAEhC,OAAM,IAAI,MAAM,CAAC,KAAK,aAAa,CAAC;;AAIxC,QAAO;;;;;;;;;;;;;;;;;;;;;;;;;ACJT,SAAgB,gBACd,KACA,WACA,YACe;AACf,KAAI,CAAC,IAAK,QAAO;AAGjB,KAAI,IAAI,SAAS,IAAI,EAAE;EACrB,MAAM,aAAa,UAAU,IAAI;EACjC,MAAM,eAAe,WAAW,SAAS,MAAM,GAAG,aAAa,aAAa;AAG5E,OAAK,MAAM,GAAG,UAAU,UACtB,MAAK,MAAM,KAAK,OAAO;AACrB,OAAI,MAAM,gBAAgB,MAAM,WAAY,QAAO;AACnD,OAAI,EAAE,aAAa,KAAK,aAAa,aAAa,CAAE,QAAO;;;CAQjE,MAAM,UAAU,SAAS,IAAI,CAC1B,QAAQ,UAAU,GAAG,CACrB,aAAa,CACb,MAAM;AAET,KAAI,CAAC,QAAS,QAAO;CAErB,MAAM,aAAa,UAAU,IAAI,QAAQ;AAEzC,KAAI,CAAC,cAAc,WAAW,WAAW,EACvC,QAAO;AAGT,KAAI,WAAW,WAAW,EACxB,QAAO,WAAW;CAIpB,MAAM,YAAY,QAAQ,WAAW;CAErC,IAAI,WAA0B;CAC9B,IAAI,gBAAgB;CACpB,IAAI,cAAc;AAElB,MAAK,MAAM,aAAa,YAAY;EAElC,MAAM,YAAY,mBAAmB,WADhB,QAAQ,UAAU,CACsB;EAC7D,MAAM,UAAU,UAAU;AAE1B,MACE,YAAY,iBACX,cAAc,iBAAiB,UAAU,aAC1C;AACA,mBAAgB;AAChB,iBAAc;AACd,cAAW;;;AAIf,QAAO;;;;;;;;AAST,SAAS,mBAAmB,GAAW,GAAmB;AACxD,KAAI,MAAM,OAAO,MAAM,IAAK,QAAO;CACnC,MAAM,SAAS,MAAM,MAAM,EAAE,GAAG,EAAE,MAAM,IAAI;CAC5C,MAAM,SAAS,MAAM,MAAM,EAAE,GAAG,EAAE,MAAM,IAAI;CAC5C,IAAI,QAAQ;CACZ,MAAM,MAAM,KAAK,IAAI,OAAO,QAAQ,OAAO,OAAO;AAClD,MAAK,IAAI,IAAI,GAAG,IAAI,KAAK,IACvB,KAAI,OAAO,OAAO,OAAO,GACvB;KAEA;AAGJ,QAAO;;;;;;;;;;;;;;;;;;;;ACxET,MAAM,oBAAoB;;;;;;;;;;;;;;;;;;;;;;;;;;;AA4B1B,eAAsB,WACpB,SACA,gBACA,WAC2B;CAC3B,MAAM,YAAY,KAAK,KAAK;CAE5B,MAAM,SAA2B;EAC/B,cAAc;EACd,eAAe;EACf,cAAc;EACd,iBAAiB;EACjB,gBAAgB;EAChB,gBAAgB;EAChB,cAAc;EACd,SAAS;EACV;CAGD,MAAM,WAAW,iBAAiB,UAAU;CAG5C,MAAM,cAAc,mBAAmB,SAAS;CAGhD,MAAM,YAAY,eAAe,SAAS;AAG1C,OAAM,kBAAkB;CACxB,IAAI,kBAAkB;AAEtB,MAAK,MAAM,SAAS,aAAa;AAC/B,MAAI,mBAAmB,mBAAmB;AACxC,SAAM,kBAAkB;AACxB,qBAAkB;;AAEpB;EAEA,MAAM,EAAE,cAAc;EAEtB,IAAI;AACJ,MAAI;AACF,aAAU,aAAa,UAAU,SAAS,OAAO;UAC3C;AACN,UAAO;AACP;;EAGF,MAAM,OAAO,WAAW,QAAQ;AAIhC,OADiB,MAAM,QAAQ,aAAa,UAAU,aAAa,GACrD,SAAS,MAAM;AAC3B,UAAO;AACP;;AAIF,QAAM,QAAQ,oBAAoB,gBAAgB,UAAU,aAAa;EAGzE,MAAM,SAAS,cAAc,QAAQ;EACrC,MAAM,YAAY,KAAK,KAAK;EAG5B,MAAM,aAAa,cAAc,KAAK,QAAQ;EAC9C,MAAM,QAAQ,aACV,WAAW,GAAI,MAAM,GACrB,SAAS,UAAU,cAAc,MAAM;EAG3C,MAAM,YAAwB,EAAE;AAChC,OAAK,IAAI,IAAI,GAAG,IAAI,OAAO,QAAQ,KAAK;GACtC,MAAM,QAAQ,OAAO;GACrB,MAAM,KAAK,QACT,gBACA,UAAU,cACV,GACA,MAAM,WACN,MAAM,QACP;AACD,aAAU,KAAK;IACb;IACA,WAAW;IACX,QAAQ;IACR,MAAM;IACN,MAAM,UAAU;IAChB,WAAW,MAAM;IACjB,SAAS,MAAM;IACf,MAAM,MAAM;IACZ,MAAM,MAAM;IACZ;IACD,CAAC;;AAGJ,MAAI,UAAU,SAAS,EACrB,OAAM,QAAQ,aAAa,UAAU;EAIvC,MAAM,eAA6B;GACjC,WAAW,UAAU;GACrB,OAAO,UAAU;GACjB,QAAQ,UAAU;GAClB;GACA;GACA,WAAW;GACZ;AACD,QAAM,QAAQ,gBAAgB,aAAa;AAE3C,SAAO;AACP,SAAO,iBAAiB,OAAO;;AAIjC,OAAM,kBAAkB;CAExB,MAAM,aAA8B,EAAE;AACtC,MAAK,MAAM,SAAS,YAClB,MAAK,MAAM,SAAS,MAAM,SAAS;AACjC,aAAW,KAAK;GACd,WAAW,MAAM;GACjB,eAAe,MAAM,UAAU;GAC/B,OAAO,MAAM;GACb,QAAQ,MAAM;GACf,CAAC;AACF,SAAO;;CAIX,MAAM,iBAAiB,IAAI,IAAI,YAAY,KAAK,MAAM,EAAE,UAAU,aAAa,CAAC;AAChF,MAAK,MAAM,aAAa,eACtB,OAAM,QAAQ,mBAAmB,UAAU;AAE7C,KAAI,WAAW,SAAS,EACtB,OAAM,QAAQ,mBAAmB,WAAW;AAI9C,OAAM,kBAAkB;CAExB,MAAM,cAAgC,EAAE;AACxC,MAAK,MAAM,CAAC,MAAM,UAAU,UAC1B,MAAK,MAAM,QAAQ,MACjB,aAAY,KAAK;EAAE;EAAM,WAAW;EAAM,CAAC;AAG/C,OAAM,QAAQ,iBAAiB,YAAY;AAG3C,OAAM,kBAAkB;CAExB,MAAM,WAA2B,EAAE;CACnC,MAAM,iBAA2B,EAAE;CAEnC,IAAI,iBAAiB;AACrB,MAAK,MAAM,SAAS,aAAa;AAC/B,MAAI,mBAAmB,sBAAsB,EAC3C,OAAM,kBAAkB;EAG1B,MAAM,EAAE,cAAc;AACtB,iBAAe,KAAK,UAAU,aAAa;EAE3C,IAAI;AACJ,MAAI;AACF,aAAU,aAAa,UAAU,SAAS,OAAO;UAC3C;AACN;;EAGF,MAAM,cAAc,WAAW,QAAQ;AACvC,OAAK,MAAM,QAAQ,aAAa;GAC9B,MAAM,SAAS,gBAAgB,KAAK,KAAK,WAAW,UAAU,aAAa;GAC3E,IAAI;AACJ,OAAI,KAAK,SACP,YAAW,KAAK,UAAU,aAAa;OAEvC,YAAW,KAAK,UAAU,UAAU;AAEtC,YAAS,KAAK;IACZ,YAAY,UAAU;IACtB,WAAW,KAAK;IAChB,YAAY;IACZ;IACA,YAAY,KAAK;IACjB,YAAY;IACb,CAAC;;;CAKN,MAAM,kBAAkB;AACxB,MAAK,IAAI,IAAI,GAAG,IAAI,eAAe,QAAQ,KAAK,iBAAiB;EAC/D,MAAM,eAAe,eAAe,MAAM,GAAG,IAAI,gBAAgB;EACjE,MAAM,aAAa,SAAS,QAAQ,MAAM,aAAa,SAAS,EAAE,WAAW,CAAC;AAC9E,QAAM,QAAQ,uBAAuB,cAAc,WAAW;AAC9D,QAAM,kBAAkB;;AAG1B,QAAO,iBAAiB,SAAS;AACjC,QAAO,iBAAiB,SAAS,QAAQ,MAAM,EAAE,eAAe,KAAK,CAAC;AAGtE,OAAM,kBAAkB;CAExB,MAAM,8BAAc,IAAI,KAAqB;CAC7C,MAAM,0BAAU,IAAI,KAAqB;CACzC,MAAM,6BAAa,IAAI,KAAqB;AAE5C,MAAK,MAAM,OAAO,UAAU;AAC1B,cAAY,IAAI,IAAI,aAAa,YAAY,IAAI,IAAI,WAAW,IAAI,KAAK,EAAE;AAC3E,MAAI,IAAI,eAAe,KACrB,SAAQ,IAAI,IAAI,aAAa,QAAQ,IAAI,IAAI,WAAW,IAAI,KAAK,EAAE;MAEnE,YAAW,IAAI,IAAI,aAAa,WAAW,IAAI,IAAI,WAAW,IAAI,KAAK,EAAE;;CAI7E,MAAM,aAAa,KAAK,KAAK;CAC7B,IAAI,cAAc;CAElB,MAAM,oBAAoB;AAC1B,MAAK,IAAI,IAAI,GAAG,IAAI,YAAY,QAAQ,KAAK,mBAAmB;EAE9D,MAAM,aADQ,YAAY,MAAM,GAAG,IAAI,kBAAkB,CACd,KAAK,UAAU;GACxD,MAAM,OAAO,MAAM,UAAU;GAC7B,MAAM,UAAU,WAAW,IAAI,KAAK,IAAI;GACxC,MAAM,WAAW,YAAY,IAAI,KAAK,IAAI;GAC1C,MAAM,OAAO,QAAQ,IAAI,KAAK,IAAI;GAClC,MAAM,WAAW,YAAY;AAC7B,OAAI,SAAU;AACd,UAAO;IACL,WAAW;IACX,cAAc;IACd,eAAe;IACf,eAAe;IACf;IACA;IACD;IACD;AACF,QAAM,QAAQ,kBAAkB,WAAW;AAC3C,QAAM,kBAAkB;;AAG1B,QAAO,eAAe;AACtB,QAAO,UAAU,KAAK,KAAK,GAAG;AAE9B,QAAO"}

package/dist/{work-queue-worker-94INtUiz.mjs → work-queue-worker-BxsBtz6j.mjs}

@@ -1,6 +1,6 @@
 import { t as __exportAll } from "./rolldown-runtime-95iHPtFO.mjs";
 import { n as openFederation } from "./db-CYmBWcjh.mjs";
-import { i as performScan, t as extractAndStoreTriples$1 } from "./kg-extraction-r5KkTmWY.mjs";
+import { i as performScan, t as extractAndStoreTriples$1 } from "./kg-extraction-iR1BKKWK.mjs";
 import { O as storageBackend, t as daemonConfig, u as registryDb } from "./state-BHZAhLou.mjs";
 import { t as detectTopicShift } from "./detector-Q_lVXtj-.mjs";
 import { existsSync, mkdirSync, readFileSync, readdirSync, renameSync, statSync, unlinkSync, writeFileSync } from "node:fs";
@@ -1821,4 +1821,4 @@ function extractWorkFromTranscript(lines) {
 
 //#endregion
 export { enqueue as a, work_queue_worker_exports as i, startWorker as n, getStats as o, stopWorker as r, loadQueue as s, notifyNewWork as t };
-//# sourceMappingURL=work-queue-worker-94INtUiz.mjs.map
+//# sourceMappingURL=work-queue-worker-BxsBtz6j.mjs.map