qmd.ts

// Load .env before any other import so env-driven singletons (LlamaCpp,
// JinaEmbedder, etc.) see the values on their first read. Safe no-op when
// no .env file is present. Shell env vars still win over .env values.
import { loadDotenv } from "../dotenv.js";
loadDotenv();

import { openDatabase } from "../db.js";
import type { Database } from "../db.js";
import fastGlob from "fast-glob";
import { execSync, spawn as nodeSpawn } from "child_process";
import { fileURLToPath } from "url";
import { dirname, join as pathJoin, relative as relativePath } from "path";
import { parseArgs } from "util";
import { readFileSync, realpathSync, statSync, existsSync, unlinkSync, writeFileSync, openSync, closeSync, mkdirSync, lstatSync, rmSync, symlinkSync, readlinkSync } from "fs";
import { createInterface } from "readline/promises";
import {
  getPwd,
  getRealPath,
  homedir,
  resolve,
  enableProductionMode,
  searchFTS,
  extractSnippet,
  getContextForFile,
  getContextForPath,
  listCollections,
  removeCollection,
  renameCollection,
  findSimilarFiles,
  findDocumentByDocid,
  isDocid,
  matchFilesByGlob,
  getHashesNeedingEmbedding,
  clearAllEmbeddings,
  insertEmbedding,
  getStatus,
  hashContent,
  extractTitle,
  formatDocForEmbedding,
  chunkDocumentByTokens,
  clearCache,
  getCacheKey,
  getCachedResult,
  setCachedResult,
  getIndexHealth,
  parseVirtualPath,
  buildVirtualPath,
  isVirtualPath,
  resolveVirtualPath,
  toVirtualPath,
  insertContent,
  insertDocument,
  findActiveDocument,
  updateDocumentTitle,
  updateDocument,
  deactivateDocument,
  getActiveDocumentPaths,
  cleanupOrphanedContent,
  deleteLLMCache,
  deleteInactiveDocuments,
  cleanupOrphanedVectors,
  vacuumDatabase,
  getCollectionsWithoutContext,
  getTopLevelPathsWithoutContext,
  handelize,
  hybridQuery,
  vectorSearchQuery,
  structuredSearch,
  addLineNumbers,
  type ExpandedQuery,
  type HybridQueryExplain,
  DEFAULT_EMBED_MODEL,
  DEFAULT_EMBED_MAX_BATCH_BYTES,
  DEFAULT_EMBED_MAX_DOCS_PER_BATCH,
  DEFAULT_RERANK_MODEL,
  DEFAULT_GLOB,
  DEFAULT_MULTI_GET_MAX_BYTES,
  createStore,
  getDefaultDbPath,
  reindexCollection,
  generateEmbeddings,
  syncConfigToDb,
  type ReindexResult,
  type ChunkStrategy,
  recordJinaUsage,
  getJinaUsageSummary,
  getJinaUsageTotal,
  clearJinaUsage,
  getUsageSnapshot,
  getDailyUsage,
  type JinaUsageSummary,
  type UsageSnapshot,
  type QuotaWindow,
  type QuotaSeverity,
  type DailyUsageBucket,
} from "../store.js";
import { disposeDefaultLlamaCpp, getDefaultLlamaCpp, setDefaultLlamaCpp, LlamaCpp, withLLMSession, pullModels, DEFAULT_EMBED_MODEL_URI, DEFAULT_GENERATE_MODEL_URI, DEFAULT_RERANK_MODEL_URI, DEFAULT_MODEL_CACHE_DIR } from "../llm.js";
import {
  formatSearchResults,
  formatDocuments,
  escapeXml,
  escapeCSV,
  type OutputFormat,
} from "./formatter.js";
import {
  getCollection as getCollectionFromYaml,
  listCollections as yamlListCollections,
  getDefaultCollectionNames,
  addContext as yamlAddContext,
  removeContext as yamlRemoveContext,
  removeCollection as yamlRemoveCollectionFn,
  renameCollection as yamlRenameCollectionFn,
  setGlobalContext,
  listAllContexts,
  setConfigIndexName,
  loadConfig,
} from "../collections.js";
import { getEmbeddedQmdSkillContent, getEmbeddedQmdSkillFiles } from "../embedded-skills.js";

// Enable production mode - allows using default database path
// Tests must set INDEX_PATH or use createStore() with explicit path
enableProductionMode();

// =============================================================================
// Store/DB lifecycle (no legacy singletons in store.ts)
// =============================================================================

let store: ReturnType<typeof createStore> | null = null;
let storeDbPathOverride: string | undefined;

function getStore(): ReturnType<typeof createStore> {
  if (!store) {
    store = createStore(storeDbPathOverride);
    // Sync YAML config into SQLite store_collections so store.ts reads from DB
    let loadedConfig: ReturnType<typeof loadConfig> | undefined;
    try {
      loadedConfig = loadConfig();
      syncConfigToDb(store.db, loadedConfig);
    } catch {
      // Config may not exist yet — that's fine, DB works without it
    }

    // If config explicitly specifies models, honour them. Bubble up
    // construction errors when a remote provider URI (jina:*) is requested
    // so the user gets a clear error instead of a silent fall-back to local.
    if (loadedConfig?.models) {
      try {
        setDefaultLlamaCpp(new LlamaCpp({
          embedModel: loadedConfig.models.embed,
          generateModel: loadedConfig.models.generate,
          rerankModel: loadedConfig.models.rerank,
        }));
      } catch (err) {
        const wantsRemote =
          loadedConfig.models.embed?.startsWith("jina:") ||
          loadedConfig.models.rerank?.startsWith("jina:");
        if (wantsRemote) {
          process.stderr.write(
            `${c.red}QMD Error: ${err instanceof Error ? err.message : String(err)}${c.reset}\n`,
          );
          process.exit(1);
        }
        // Non-remote errors: keep the legacy silent behaviour.
      }
    }

    // Attach the Jina usage reporter to the active LlamaCpp instance so any
    // remote API calls record their token usage to the local SQLite store.
    // Safe to call even when no remote provider is active — it's a no-op.
    try {
      const dbForReporter = store.db;
      getDefaultLlamaCpp().setRemoteUsageReporter((event) => {
        recordJinaUsage(dbForReporter, {
          operation: event.operation,
          model: event.model,
          totalTokens: event.totalTokens,
          promptTokens: event.promptTokens,
          at: event.at,
        });
      });
    } catch {
      // Reporter wiring is best-effort.
    }
  }
  return store;
}

function getDb(): Database {
  return getStore().db;
}

/** Re-sync YAML config into SQLite after CLI mutations (add/remove/rename collection, context changes) */
function resyncConfig(): void {
  const s = getStore();
  try {
    const config = loadConfig();
    // Clear config hash to force re-sync
    s.db.prepare(`DELETE FROM store_config WHERE key = 'config_hash'`).run();
    syncConfigToDb(s.db, config);
  } catch {
    // Config may not exist — that's fine
  }
}

function closeDb(): void {
  if (store) {
    store.close();
    store = null;
  }
}

function getDbPath(): string {
  return store?.dbPath ?? storeDbPathOverride ?? getDefaultDbPath();
}

function setIndexName(name: string | null): void {
  let normalizedName = name;
  // Normalize relative paths to prevent malformed database paths
  if (name && name.includes('/')) {
    const { resolve } = require('path');
    const { cwd } = require('process');
    const absolutePath = resolve(cwd(), name);
    // Replace path separators with underscores to create a valid filename
    normalizedName = absolutePath.replace(/\//g, '_').replace(/^_/, '');
  }
  storeDbPathOverride = normalizedName ? getDefaultDbPath(normalizedName) : undefined;
  // Reset open handle so next use opens the new index
  closeDb();
}

function ensureVecTable(_db: Database, dimensions: number): void {
  // Store owns the DB; ignore `_db` and ensure vec table on the active store
  getStore().ensureVecTable(dimensions);
}

// Terminal colors (respects NO_COLOR env)
const useColor = !process.env.NO_COLOR && process.stdout.isTTY;
const c = {
  reset: useColor ? "\x1b[0m" : "",
  dim: useColor ? "\x1b[2m" : "",
  bold: useColor ? "\x1b[1m" : "",
  cyan: useColor ? "\x1b[36m" : "",
  yellow: useColor ? "\x1b[33m" : "",
  green: useColor ? "\x1b[32m" : "",
  magenta: useColor ? "\x1b[35m" : "",
  blue: useColor ? "\x1b[34m" : "",
  red: useColor ? "\x1b[31m" : "",
};

// Terminal cursor control
const cursor = {
  hide() { process.stderr.write('\x1b[?25l'); },
  show() { process.stderr.write('\x1b[?25h'); },
};

// Ensure cursor is restored on exit
process.on('SIGINT', () => { cursor.show(); process.exit(130); });
process.on('SIGTERM', () => { cursor.show(); process.exit(143); });

// Terminal progress bar using OSC 9;4 escape sequence (TTY only)
const isTTY = process.stderr.isTTY;
const progress = {
  set(percent: number) {
    if (isTTY) process.stderr.write(`\x1b]9;4;1;${Math.round(percent)}\x07`);
  },
  clear() {
    if (isTTY) process.stderr.write(`\x1b]9;4;0\x07`);
  },
  indeterminate() {
    if (isTTY) process.stderr.write(`\x1b]9;4;3\x07`);
  },
  error() {
    if (isTTY) process.stderr.write(`\x1b]9;4;2\x07`);
  },
};

// Format seconds into human-readable ETA
function formatETA(seconds: number): string {
  if (seconds < 60) return `${Math.round(seconds)}s`;
  if (seconds < 3600) return `${Math.floor(seconds / 60)}m ${Math.round(seconds % 60)}s`;
  return `${Math.floor(seconds / 3600)}h ${Math.floor((seconds % 3600) / 60)}m`;
}


// Check index health and print warnings/tips
function checkIndexHealth(db: Database): void {
  const { needsEmbedding, totalDocs, daysStale } = getIndexHealth(db);

  // Warn if many docs need embedding
  if (needsEmbedding > 0) {
    const pct = Math.round((needsEmbedding / totalDocs) * 100);
    if (pct >= 10) {
      process.stderr.write(`${c.yellow}Warning: ${needsEmbedding} documents (${pct}%) need embeddings. Run 'qmd embed' for better results.${c.reset}\n`);
    } else {
      process.stderr.write(`${c.dim}Tip: ${needsEmbedding} documents need embeddings. Run 'qmd embed' to index them.${c.reset}\n`);
    }
  }

  // Check if most recent document update is older than 2 weeks
  if (daysStale !== null && daysStale >= 14) {
    process.stderr.write(`${c.dim}Tip: Index last updated ${daysStale} days ago. Run 'qmd update' to refresh.${c.reset}\n`);
  }
}

// Compute unique display path for a document
// Always include at least parent folder + filename, add more parent dirs until unique
function computeDisplayPath(
  filepath: string,
  collectionPath: string,
  existingPaths: Set<string>
): string {
  // Get path relative to collection (include collection dir name)
  const collectionDir = collectionPath.replace(/\/$/, '');
  const collectionName = collectionDir.split('/').pop() || '';

  let relativePath: string;
  if (filepath.startsWith(collectionDir + '/')) {
    // filepath is under collection: use collection name + relative path
    relativePath = collectionName + filepath.slice(collectionDir.length);
  } else {
    // Fallback: just use the filepath
    relativePath = filepath;
  }

  const parts = relativePath.split('/').filter(p => p.length > 0);

  // Always include at least parent folder + filename (minimum 2 parts if available)
  // Then add more parent dirs until unique
  const minParts = Math.min(2, parts.length);
  for (let i = parts.length - minParts; i >= 0; i--) {
    const candidate = parts.slice(i).join('/');
    if (!existingPaths.has(candidate)) {
      return candidate;
    }
  }

  // Absolute fallback: use full path (should be unique)
  return filepath;
}


function formatTimeAgo(date: Date): string {
  const seconds = Math.floor((Date.now() - date.getTime()) / 1000);
  if (seconds < 60) return `${seconds}s ago`;
  const minutes = Math.floor(seconds / 60);
  if (minutes < 60) return `${minutes}m ago`;
  const hours = Math.floor(minutes / 60);
  if (hours < 24) return `${hours}h ago`;
  const days = Math.floor(hours / 24);
  return `${days}d ago`;
}

function formatMs(ms: number): string {
  if (ms < 1000) return `${ms}ms`;
  return `${(ms / 1000).toFixed(1)}s`;
}

function formatBytes(bytes: number): string {
  if (bytes < 1024) return `${bytes} B`;
  if (bytes < 1024 * 1024) return `${(bytes / 1024).toFixed(1)} KB`;
  if (bytes < 1024 * 1024 * 1024) return `${(bytes / (1024 * 1024)).toFixed(1)} MB`;
  return `${(bytes / (1024 * 1024 * 1024)).toFixed(1)} GB`;
}

/** Format token counts with K/M/B suffixes for human readability. */
function formatTokens(n: number): string {
  const abs = Math.abs(n);
  const sign = n < 0 ? "-" : "";
  if (abs < 1_000) return `${sign}${abs}`;
  if (abs < 1_000_000) return `${sign}${(abs / 1_000).toFixed(1)}K`;
  if (abs < 1_000_000_000) return `${sign}${(abs / 1_000_000).toFixed(2)}M`;
  return `${sign}${(abs / 1_000_000_000).toFixed(2)}B`;
}

/**
 * Parse a token count from an env var. Accepts plain digits and suffixes:
 *   "1000000000"  → 1_000_000_000
 *   "1b" / "1B"   → 1_000_000_000
 *   "500m"        → 500_000_000
 *   "10k"         → 10_000
 * Returns 0 for invalid / missing values (disables quota tracking).
 */
function parseTokenEnv(raw: string | undefined): number {
  if (!raw) return 0;
  const trimmed = raw.trim().toLowerCase();
  if (!trimmed) return 0;
  const match = trimmed.match(/^([\d.]+)\s*([kmb]?)$/);
  if (!match) return 0;
  const num = Number.parseFloat(match[1]!);
  if (!Number.isFinite(num) || num <= 0) return 0;
  const mul =
    match[2] === "k" ? 1_000 :
    match[2] === "m" ? 1_000_000 :
    match[2] === "b" ? 1_000_000_000 :
    1;
  return Math.floor(num * mul);
}

/** Parse QMD_JINA_QUOTA_WINDOW → QuotaWindow, with default and validation. */
function parseQuotaWindow(raw: string | undefined): QuotaWindow {
  const v = (raw ?? "").trim().toLowerCase();
  if (v === "24h" || v === "7d" || v === "30d" || v === "all") return v;
  return "30d";
}

/**
 * Read quota configuration from environment. Centralised so `qmd usage`,
 * `qmd status`, and any future entrypoints compute the same numbers.
 */
function readQuotaOptionsFromEnv(): {
  quotaLimit: number;
  quotaWindow: QuotaWindow;
  warnFraction: number;
} {
  const quotaLimit = parseTokenEnv(process.env.QMD_JINA_QUOTA);
  const quotaWindow = parseQuotaWindow(process.env.QMD_JINA_QUOTA_WINDOW);
  let warnFraction = 0.8;
  const warnPctRaw = process.env.QMD_JINA_WARN_PCT?.trim();
  if (warnPctRaw) {
    const n = Number.parseFloat(warnPctRaw);
    if (Number.isFinite(n) && n > 0 && n <= 100) {
      warnFraction = n / 100;
    }
  }
  return { quotaLimit, quotaWindow, warnFraction };
}

/** Colour for a quota severity level. */
function severityColor(severity: QuotaSeverity): string {
  if (severity === "over") return c.red;
  if (severity === "critical") return c.red;
  if (severity === "warn") return c.yellow;
  return c.green;
}

/** Human label for a quota severity level. */
function severityLabel(severity: QuotaSeverity): string {
  if (severity === "over") return "OVER QUOTA";
  if (severity === "critical") return "CRITICAL";
  if (severity === "warn") return "WARN";
  return "OK";
}

/**
 * Build the snapshot shape rendered by both `qmd usage` and `qmd status`.
 * The `llmInstance` is captured separately so JSON output can include the
 * active provider names without calling the getter twice.
 */
function collectUsageReport(): {
  snapshot: UsageSnapshot;
  embedProvider: string | null;
  rerankProvider: string | null;
} {
  const db = getDb();
  const llmInstance = getDefaultLlamaCpp();
  const quotaOpts = readQuotaOptionsFromEnv();
  const snapshot = getUsageSnapshot(db, {
    quotaLimit: quotaOpts.quotaLimit,
    quotaWindow: quotaOpts.quotaWindow,
    warnFraction: quotaOpts.warnFraction,
  });
  return {
    snapshot,
    embedProvider: llmInstance.usingRemoteEmbedder ? llmInstance.embedProviderName : null,
    rerankProvider: llmInstance.usingRemoteReranker ? llmInstance.rerankProviderName : null,
  };
}

/**
 * Render a JSON payload describing current Jina usage. Stable shape intended
 * for scripting — wraps `UsageSnapshot` with provider info and a schema tag.
 */
function renderUsageJson(report: ReturnType<typeof collectUsageReport>): string {
  const { snapshot, embedProvider, rerankProvider } = report;
  return JSON.stringify(
    {
      schema: "qmd.usage.v1",
      generated_at: new Date().toISOString(),
      provider: { embed: embedProvider, rerank: rerankProvider },
      totals: {
        last_24h: snapshot.totals.last24h,
        last_7d: snapshot.totals.last7d,
        last_30d: snapshot.totals.last30d,
        all_time: snapshot.totals.allTime,
      },
      by_operation: snapshot.byOperation.map((row) => ({
        operation: row.operation,
        model: row.model,
        calls: row.calls,
        total_tokens: row.totalTokens,
      })),
      quota: snapshot.quota
        ? {
            limit: snapshot.quota.limit,
            window: snapshot.quota.window,
            used: snapshot.quota.used,
            used_fraction: snapshot.quota.usedFraction,
            remaining: snapshot.quota.remaining,
            warn_fraction: snapshot.quota.warnFraction,
            severity: snapshot.quota.severity,
          }
        : null,
    },
    null,
    2,
  );
}

/**
 * Show Jina API token usage broken down by operation + model, plus rolling
 * windows (24h, 7d, 30d, all) and optional quota threshold. Designed to fit
 * in a small terminal.
 *
 * When `json` is true, emit a machine-readable payload instead.
 */
async function showUsage(options: { json?: boolean } = {}): Promise<void> {
  const report = collectUsageReport();

  if (options.json) {
    process.stdout.write(renderUsageJson(report) + "\n");
    return;
  }

  const { snapshot, embedProvider, rerankProvider } = report;

  console.log(`${c.bold}QMD Jina Usage${c.reset}\n`);

  if (!embedProvider && !rerankProvider) {
    console.log(`${c.dim}No remote provider active. Set QMD_EMBED_PROVIDER=jina${c.reset}`);
    console.log(`${c.dim}or QMD_RERANK_PROVIDER=jina to enable Jina API usage tracking.${c.reset}`);
  } else {
    if (embedProvider) {
      console.log(`${c.dim}Embed provider:  ${embedProvider}${c.reset}`);
    }
    if (rerankProvider) {
      console.log(`${c.dim}Rerank provider: ${rerankProvider}${c.reset}`);
    }
    console.log();
  }

  if (snapshot.totals.allTime === 0) {
    console.log(`${c.dim}No usage recorded yet.${c.reset}`);
    return;
  }

  // Quota block (if configured)
  if (snapshot.quota) {
    const q = snapshot.quota;
    const color = severityColor(q.severity);
    const pct = (q.usedFraction * 100).toFixed(1);
    const label = severityLabel(q.severity);
    const warnPct = (q.warnFraction * 100).toFixed(0);
    console.log(`${c.bold}Quota (${q.window} window)${c.reset}`);
    console.log(
      `  Used:       ${color}${formatTokens(q.used)}${c.reset} / ${formatTokens(q.limit)}  ` +
      `(${color}${pct}%${c.reset})  ${color}${label}${c.reset}`,
    );
    if (q.remaining >= 0) {
      console.log(`  Remaining:  ${formatTokens(q.remaining)}`);
    } else {
      console.log(`  Over by:    ${color}${formatTokens(-q.remaining)}${c.reset}`);
    }
    console.log(`  ${c.dim}Warn threshold: ${warnPct}%${c.reset}\n`);
  }

  console.log(`${c.bold}Tokens consumed${c.reset}`);
  console.log(`  Last 24h:  ${formatTokens(snapshot.totals.last24h)}`);
  console.log(`  Last 7d:   ${formatTokens(snapshot.totals.last7d)}`);
  console.log(`  Last 30d:  ${formatTokens(snapshot.totals.last30d)}`);
  console.log(`  All time:  ${formatTokens(snapshot.totals.allTime)}`);

  if (snapshot.byOperation.length > 0) {
    console.log(`\n${c.bold}By operation${c.reset}`);
    // Compute column widths
    const opWidth = Math.max(9, ...snapshot.byOperation.map((s) => s.operation.length));
    const modelWidth = Math.max(5, ...snapshot.byOperation.map((s) => s.model.length));
    for (const row of snapshot.byOperation) {
      const op = row.operation.padEnd(opWidth);
      const model = row.model.padEnd(modelWidth);
      const calls = String(row.calls).padStart(6);
      const tokens = formatTokens(row.totalTokens).padStart(8);
      console.log(`  ${op}  ${c.dim}${model}${c.reset}  ${calls} calls  ${tokens} tok`);
    }
  }

  if (!snapshot.quota) {
    console.log(
      `\n${c.dim}Tip: set QMD_JINA_QUOTA=1B (or 500m, 10k) to enable quota warnings.${c.reset}`,
    );
  }
  console.log(
    `${c.dim}Tip: 'qmd usage reset' to clear history, 'qmd usage --json' for scripting, 'qmd usage chart' for daily histogram.${c.reset}`,
  );
}

/**
 * Render a usage report as CSV for spreadsheet import. Emits one row per
 * (operation, model) pair — the most common shape for tracking consumption
 * in Google Sheets / Excel. Totals and quota live in `--json` output; users
 * who need those can combine `jq` + spreadsheet tools.
 */
function renderUsageCsv(report: ReturnType<typeof collectUsageReport>): string {
  const rows: string[] = [];
  rows.push("operation,model,calls,total_tokens");
  for (const row of report.snapshot.byOperation) {
    // CSV escape: wrap in quotes if contains comma / quote / newline, double-up embedded quotes.
    const esc = (s: string) => {
      if (/[,"\n]/.test(s)) return `"${s.replace(/"/g, '""')}"`;
      return s;
    };
    rows.push(`${esc(row.operation)},${esc(row.model)},${row.calls},${row.totalTokens}`);
  }
  return rows.join("\n") + "\n";
}

/**
 * Render an ASCII histogram of daily token usage. Computes a bar width
 * proportional to the max daily value so the shape scales to any magnitude.
 *
 * Layout:
 *   MM-DD  <bar>  <label>
 *
 * Empty days render as a blank bar so gaps are visible. Uses unicode block
 * characters for partial cells to get finer-grained bars.
 */
function renderUsageChart(
  buckets: DailyUsageBucket[],
  options: { barWidth?: number; useColor?: boolean } = {},
): string {
  if (buckets.length === 0) return `${c.dim}No usage data.${c.reset}\n`;

  const barWidth = options.barWidth ?? 40;
  const max = Math.max(...buckets.map((b) => b.totalTokens));
  if (max === 0) {
    // All zero — show the outline so users know the range was checked.
    const lines: string[] = [];
    lines.push(`${c.dim}No tokens consumed in the last ${buckets.length} days.${c.reset}`);
    for (const b of buckets) {
      const mmdd = b.date.slice(5); // MM-DD
      lines.push(`  ${c.dim}${mmdd}${c.reset}  ${c.dim}${"·".repeat(Math.max(0, barWidth))}${c.reset}  ${c.dim}0${c.reset}`);
    }
    return lines.join("\n") + "\n";
  }

  // Unicode block characters for sub-cell bar width.
  // Each cell = 1 full block; partial cells use 1/8 increments.
  const partials = ["", "▏", "▎", "▍", "▌", "▋", "▊", "▉"];
  const full = "█";

  const lines: string[] = [];
  for (const b of buckets) {
    const mmdd = b.date.slice(5); // MM-DD portion of YYYY-MM-DD
    const fraction = b.totalTokens / max;
    const cells = fraction * barWidth;
    const fullCells = Math.floor(cells);
    const partialIdx = Math.floor((cells - fullCells) * 8);
    const partial = partials[partialIdx] ?? "";
    const bar = full.repeat(fullCells) + partial;
    const padding = " ".repeat(Math.max(0, barWidth - bar.length));

    // Color: dim for empty, green for below 50% of max, yellow above.
    const color =
      b.totalTokens === 0 ? c.dim :
      fraction >= 0.8 ? c.yellow :
      c.green;

    const label = formatTokens(b.totalTokens).padStart(8);
    lines.push(`  ${c.dim}${mmdd}${c.reset}  ${color}${bar}${c.reset}${padding}  ${label}`);
  }
  return lines.join("\n") + "\n";
}

/**
 * Show a daily usage histogram for the last N days. Defaults to 30 days.
 * Can emit JSON with `--json`.
 */
async function showUsageChart(options: { days?: number; json?: boolean } = {}): Promise<void> {
  const db = getDb();
  const days = Math.max(1, Math.min(365, options.days ?? 30));
  const buckets = getDailyUsage(db, { days });

  if (options.json) {
    process.stdout.write(
      JSON.stringify(
        {
          schema: "qmd.usage.chart.v1",
          generated_at: new Date().toISOString(),
          days,
          buckets: buckets.map((b) => ({ date: b.date, total_tokens: b.totalTokens })),
        },
        null,
        2,
      ) + "\n",
    );
    return;
  }

  console.log(`${c.bold}Jina daily usage (last ${days} days, UTC)${c.reset}\n`);
  process.stdout.write(renderUsageChart(buckets));

  const total = buckets.reduce((sum, b) => sum + b.totalTokens, 0);
  const nonZero = buckets.filter((b) => b.totalTokens > 0).length;
  const avg = nonZero > 0 ? total / nonZero : 0;
  const max = Math.max(...buckets.map((b) => b.totalTokens));
  console.log();
  console.log(`${c.dim}Total: ${formatTokens(total)}  |  Active days: ${nonZero}/${days}  |  Peak: ${formatTokens(max)}  |  Avg/active: ${formatTokens(Math.round(avg))}${c.reset}`);
}

/**
 * Render the qmd bench jina report as a human-readable table. Mirrors the
 * JSON shape exactly so scripts and humans see the same numbers. Shows
 * stddev when stage has >1 sample (i.e. --runs > 1 or single-embed).
 */
function renderBenchJinaTable(report: {
  config: { size: number; docLen: number; runs: number; skipRerank: boolean; skipSingle: boolean };
  providers: Array<{
    provider: "local" | "jina";
    label: string;
    modelUri: string;
    stages: Array<{
      stage: "embed_single" | "embed_batch" | "rerank";
      medianMs: number;
      meanMs: number;
      stddevMs: number;
      p95Ms: number;
      minMs: number;
      maxMs: number;
      throughputPerSec: number;
      items: number;
      runs: number;
      latenciesMs: number[];
      ok: boolean;
      error?: string;
    }>;
    totalMs: number;
  }>;
  comparison?: Array<{
    stage: "embed_single" | "embed_batch" | "rerank";
    localMs: number;
    jinaMs: number;
    speedup: number;
    winner: "local" | "jina" | "tie";
  }>;
}): void {
  const stageLabel = (s: string) =>
    s === "embed_single" ? "embed (single)" :
    s === "embed_batch"  ? "embed (batch) " :
    s === "rerank"       ? "rerank        " :
    s;

  const multiSample = report.config.runs > 1;

  for (const prov of report.providers) {
    const provColor = prov.provider === "jina" ? c.green : c.cyan;
    console.log(`${c.bold}${provColor}${prov.label}${c.reset}  ${c.dim}(${prov.modelUri})${c.reset}`);
    if (multiSample) {
      console.log(`  ${c.dim}stage            median       p95      stddev   throughput    items${c.reset}`);
    } else {
      console.log(`  ${c.dim}stage            median       p95      throughput    items${c.reset}`);
    }
    for (const stage of prov.stages) {
      if (!stage.ok) {
        console.log(`  ${stageLabel(stage.stage)}  ${c.red}FAILED: ${stage.error ?? "unknown"}${c.reset}`);
        continue;
      }
      const median = `${stage.medianMs.toFixed(1)}ms`.padStart(10);
      const p95 = `${stage.p95Ms.toFixed(1)}ms`.padStart(9);
      const thru = `${stage.throughputPerSec.toFixed(1)}/s`.padStart(12);
      const items = String(stage.items).padStart(6);
      if (multiSample) {
        const stddev = `±${stage.stddevMs.toFixed(1)}ms`.padStart(10);
        // Highlight noisy stages (stddev > 20% of median) in yellow.
        const noiseColor =
          stage.medianMs > 0 && stage.stddevMs / stage.medianMs > 0.2 ? c.yellow : c.dim;
        console.log(
          `  ${stageLabel(stage.stage)}  ${median}  ${p95}  ${noiseColor}${stddev}${c.reset}  ${thru}  ${items}`,
        );
      } else {
        console.log(`  ${stageLabel(stage.stage)}  ${median}  ${p95}  ${thru}  ${items}`);
      }
    }
    console.log(`  ${c.dim}Total wall time: ${prov.totalMs}ms${c.reset}\n`);
  }

  if (report.comparison && report.comparison.length > 0) {
    console.log(`${c.bold}Comparison (local / jina)${c.reset}`);
    console.log(`  ${c.dim}stage            local       jina       speedup    winner${c.reset}`);
    for (const row of report.comparison) {
      const stage = stageLabel(row.stage);
      const local = `${row.localMs.toFixed(1)}ms`.padStart(10);
      const jina = `${row.jinaMs.toFixed(1)}ms`.padStart(10);
      const speedupRaw = row.speedup.toFixed(2) + "×";
      const speedup = speedupRaw.padStart(9);
      const winnerColor =
        row.winner === "jina" ? c.green :
        row.winner === "local" ? c.cyan :
        c.dim;
      const winner = row.winner.padEnd(5);
      console.log(`  ${stage}  ${local}  ${jina}  ${speedup}    ${winnerColor}${winner}${c.reset}`);
    }
    console.log(
      `\n${c.dim}Speedup = local_median / jina_median. >1 means jina is faster on this stage.${c.reset}`,
    );
  }

  if (multiSample) {
    console.log(
      `${c.dim}Stats computed over ${report.config.runs} runs. Stddev in yellow = high variance (>20% of median).${c.reset}`,
    );
  }
  console.log(`${c.dim}Tip: 'qmd bench jina --runs 5 --json' reduces noise and enables scripting.${c.reset}`);
}

async function showStatus(): Promise<void> {
  const dbPath = getDbPath();
  const db = getDb();

  // Collections are defined in YAML; no duplicate cleanup needed.
  // Collections are defined in YAML; no duplicate cleanup needed.

  // Index size
  let indexSize = 0;
  try {
    const stat = statSync(dbPath).size;
    indexSize = stat;
  } catch { }

  // Collections info (from YAML + database stats)
  const collections = listCollections(db);

  // Overall stats
  const totalDocs = db.prepare(`SELECT COUNT(*) as count FROM documents WHERE active = 1`).get() as { count: number };
  const vectorCount = db.prepare(`SELECT COUNT(*) as count FROM content_vectors`).get() as { count: number };
  const needsEmbedding = getHashesNeedingEmbedding(db);

  // Most recent update across all collections
  const mostRecent = db.prepare(`SELECT MAX(modified_at) as latest FROM documents WHERE active = 1`).get() as { latest: string | null };

  console.log(`${c.bold}QMD Status${c.reset}\n`);
  console.log(`Index: ${dbPath}`);
  console.log(`Size:  ${formatBytes(indexSize)}`);

  // MCP daemon status (check PID file liveness)
  const mcpCacheDir = process.env.XDG_CACHE_HOME
    ? resolve(process.env.XDG_CACHE_HOME, "qmd")
    : resolve(homedir(), ".cache", "qmd");
  const mcpPidPath = resolve(mcpCacheDir, "mcp.pid");
  if (existsSync(mcpPidPath)) {
    const mcpPid = parseInt(readFileSync(mcpPidPath, "utf-8").trim());
    try {
      process.kill(mcpPid, 0);
      console.log(`MCP:   ${c.green}running${c.reset} (PID ${mcpPid})`);
    } catch {
      unlinkSync(mcpPidPath);
      // Stale PID file cleaned up silently
    }
  }
  console.log("");

  console.log(`${c.bold}Documents${c.reset}`);
  console.log(`  Total:    ${totalDocs.count} files indexed`);
  console.log(`  Vectors:  ${vectorCount.count} embedded`);
  if (needsEmbedding > 0) {
    console.log(`  ${c.yellow}Pending:  ${needsEmbedding} need embedding${c.reset} (run 'qmd embed')`);
  }
  if (mostRecent.latest) {
    const lastUpdate = new Date(mostRecent.latest);
    console.log(`  Updated:  ${formatTimeAgo(lastUpdate)}`);
  }

  // Get all contexts grouped by collection (from YAML)
  const allContexts = listAllContexts();
  const contextsByCollection = new Map<string, { path_prefix: string; context: string }[]>();

  for (const ctx of allContexts) {
    // Group contexts by collection name
    if (!contextsByCollection.has(ctx.collection)) {
      contextsByCollection.set(ctx.collection, []);
    }
    contextsByCollection.get(ctx.collection)!.push({
      path_prefix: ctx.path,
      context: ctx.context
    });
  }

  // AST chunking status
  try {
    const { getASTStatus } = await import("../ast.js");
    const ast = await getASTStatus();
    console.log(`\n${c.bold}AST Chunking${c.reset}`);
    if (ast.available) {
      const ok = ast.languages.filter(l => l.available).map(l => l.language);
      const fail = ast.languages.filter(l => !l.available);
      console.log(`  Status:   ${c.green}active${c.reset}`);
      console.log(`  Languages: ${ok.join(", ")}`);
      if (fail.length > 0) {
        for (const f of fail) {
          console.log(`  ${c.yellow}Unavailable: ${f.language} (${f.error})${c.reset}`);
        }
      }
    } else {
      console.log(`  Status:   ${c.yellow}unavailable${c.reset} (falling back to regex chunking)`);
      for (const l of ast.languages) {
        if (l.error) console.log(`  ${c.dim}${l.language}: ${l.error}${c.reset}`);
      }
    }
  } catch {
    console.log(`\n${c.bold}AST Chunking${c.reset}`);
    console.log(`  Status:   ${c.dim}not available${c.reset}`);
  }

  if (collections.length > 0) {
    console.log(`\n${c.bold}Collections${c.reset}`);
    for (const col of collections) {
      const lastMod = col.last_modified ? formatTimeAgo(new Date(col.last_modified)) : "never";
      const contexts = contextsByCollection.get(col.name) || [];

      console.log(`  ${c.cyan}${col.name}${c.reset} ${c.dim}(qmd://${col.name}/)${c.reset}`);
      console.log(`    ${c.dim}Pattern:${c.reset}  ${col.glob_pattern}`);
      console.log(`    ${c.dim}Files:${c.reset}    ${col.active_count} (updated ${lastMod})`);

      if (contexts.length > 0) {
        console.log(`    ${c.dim}Contexts:${c.reset} ${contexts.length}`);
        for (const ctx of contexts) {
          // Handle both empty string and '/' as root context
          const pathDisplay = (ctx.path_prefix === '' || ctx.path_prefix === '/') ? '/' : `/${ctx.path_prefix}`;
          const contextPreview = ctx.context.length > 60
            ? ctx.context.substring(0, 57) + '...'
            : ctx.context;
          console.log(`      ${c.dim}${pathDisplay}:${c.reset} ${contextPreview}`);
        }
      }
    }

    // Show examples of virtual paths
    console.log(`\n${c.bold}Examples${c.reset}`);
    console.log(`  ${c.dim}# List files in a collection${c.reset}`);
    if (collections.length > 0 && collections[0]) {
      console.log(`  qmd ls ${collections[0].name}`);
    }
    console.log(`  ${c.dim}# Get a document${c.reset}`);
    if (collections.length > 0 && collections[0]) {
      console.log(`  qmd get qmd://${collections[0].name}/path/to/file.md`);
    }
    console.log(`  ${c.dim}# Search within a collection${c.reset}`);
    if (collections.length > 0 && collections[0]) {
      console.log(`  qmd search "query" -c ${collections[0].name}`);
    }
  } else {
    console.log(`\n${c.dim}No collections. Run 'qmd collection add .' to index markdown files.${c.reset}`);
  }

  // Models
  {
    // hf:org/repo/file.gguf → https://huggingface.co/org/repo
    const hfLink = (uri: string) => {
      const match = uri.match(/^hf:([^/]+\/[^/]+)\//);
      return match ? `https://huggingface.co/${match[1]}` : uri;
    };
    console.log(`\n${c.bold}Models${c.reset}`);

    // Show active embedding backend. For remote providers (jina:*) we print
    // the provider name + model instead of a HF link. If the user requested
    // a remote provider but init failed (e.g. missing API key), surface the
    // error so they can fix their env instead of silently falling back.
    const requestedRemote = (process.env.QMD_EMBED_PROVIDER ?? "").trim().toLowerCase();
    let activeEmbed: string;
    let embedInitError: string | null = null;
    try {
      activeEmbed = getDefaultLlamaCpp().embedModelName;
    } catch (err) {
      activeEmbed = DEFAULT_EMBED_MODEL_URI;
      embedInitError = err instanceof Error ? err.message : String(err);
    }

    if (embedInitError && requestedRemote) {
      console.log(`  Embedding:   ${c.red}error${c.reset} (${requestedRemote})`);
      console.log(`  ${c.red}${embedInitError}${c.reset}`);
    } else if (activeEmbed.startsWith("jina:")) {
      const model = activeEmbed.slice("jina:".length);
      console.log(`  Embedding:   ${c.green}jina${c.reset} (${model}) ${c.dim}[remote]${c.reset}`);
    } else {
      console.log(`  Embedding:   ${hfLink(activeEmbed)} ${c.dim}[local]${c.reset}`);
    }

    // Active rerank backend (mirrors the embed display).
    const requestedRemoteRerank = (process.env.QMD_RERANK_PROVIDER ?? "").trim().toLowerCase();
    let activeRerank: string;
    let rerankInitError: string | null = null;
    try {
      activeRerank = getDefaultLlamaCpp().rerankModelName;
    } catch (err) {
      activeRerank = DEFAULT_RERANK_MODEL_URI;
      rerankInitError = err instanceof Error ? err.message : String(err);
    }
    if (rerankInitError && requestedRemoteRerank) {
      console.log(`  Reranking:   ${c.red}error${c.reset} (${requestedRemoteRerank})`);
      console.log(`  ${c.red}${rerankInitError}${c.reset}`);
    } else if (activeRerank.startsWith("jina:")) {
      const model = activeRerank.slice("jina:".length);
      console.log(`  Reranking:   ${c.green}jina${c.reset} (${model}) ${c.dim}[remote]${c.reset}`);
    } else {
      console.log(`  Reranking:   ${hfLink(activeRerank)} ${c.dim}[local]${c.reset}`);
    }
    console.log(`  Generation:  ${hfLink(DEFAULT_GENERATE_MODEL_URI)}`);
  }

  // Jina usage summary (only when a remote provider is active and we have data).
  // Quota warnings are color-coded by severity (ok/warn/critical/over).
  try {
    const llmInstance = getDefaultLlamaCpp();
    if (llmInstance.usingRemoteEmbedder || llmInstance.usingRemoteReranker) {
      const quotaOpts = readQuotaOptionsFromEnv();
      const snapshot = getUsageSnapshot(db, {
        quotaLimit: quotaOpts.quotaLimit,
        quotaWindow: quotaOpts.quotaWindow,
        warnFraction: quotaOpts.warnFraction,
      });

      // Only render if we have data or a quota is configured.
      if (snapshot.totals.allTime > 0 || snapshot.quota) {
        console.log(`\n${c.bold}Jina Usage${c.reset}`);

        if (snapshot.quota) {
          const q = snapshot.quota;
          const color = severityColor(q.severity);
          const pct = (q.usedFraction * 100).toFixed(1);
          const label = severityLabel(q.severity);
          console.log(
            `  Quota (${q.window}):  ${color}${formatTokens(q.used)}${c.reset} / ${formatTokens(q.limit)} ` +
            `(${color}${pct}%${c.reset})  ${color}${label}${c.reset}`,
          );
          if (q.severity === "over") {
            console.log(
              `  ${c.red}⚠  Over quota by ${formatTokens(-q.remaining)}. Consider switching to local, waiting for reset, or upgrading.${c.reset}`,
            );
          } else if (q.severity === "critical") {
            console.log(
              `  ${c.red}⚠  Only ${formatTokens(q.remaining)} remaining (${(100 - q.usedFraction * 100).toFixed(1)}%).${c.reset}`,
            );
          } else if (q.severity === "warn") {
            console.log(
              `  ${c.yellow}⚠  ${formatTokens(q.remaining)} remaining.${c.reset}`,
            );
          }
        }

        console.log(`  24h:    ${formatTokens(snapshot.totals.last24h)} tokens`);
        console.log(`  30d:    ${formatTokens(snapshot.totals.last30d)} tokens`);
        console.log(
          `  Total:  ${formatTokens(snapshot.totals.allTime)} tokens  ${c.dim}('qmd usage' for breakdown)${c.reset}`,
        );
      }
    }
  } catch {
    // Don't fail status if usage query fails.
  }

  // Device / GPU info
  try {
    const llm = getDefaultLlamaCpp();
    const device = await llm.getDeviceInfo();
    console.log(`\n${c.bold}Device${c.reset}`);
    if (device.gpu) {
      console.log(`  GPU:      ${c.green}${device.gpu}${c.reset} (offloading: ${device.gpuOffloading ? 'yes' : 'no'})`);
      if (device.gpuDevices.length > 0) {
        // Deduplicate and count GPUs
        const counts = new Map<string, number>();
        for (const name of device.gpuDevices) {
          counts.set(name, (counts.get(name) || 0) + 1);
        }
        const deviceStr = Array.from(counts.entries())
          .map(([name, count]) => count > 1 ? `${count}× ${name}` : name)
          .join(', ');
        console.log(`  Devices:  ${deviceStr}`);
      }
      if (device.vram) {
        console.log(`  VRAM:     ${formatBytes(device.vram.free)} free / ${formatBytes(device.vram.total)} total`);
      }
    } else {
      console.log(`  GPU:      ${c.yellow}none${c.reset} (running on CPU — models will be slow)`);
      console.log(`  ${c.dim}Tip: Install CUDA, Vulkan, or Metal support for GPU acceleration.${c.reset}`);
    }
    console.log(`  CPU:      ${device.cpuCores} math cores`);
  } catch {
    // Don't fail status if LLM init fails
  }

  // Tips section
  const tips: string[] = [];

  // Check for collections without context
  const collectionsWithoutContext = collections.filter(col => {
    const contexts = contextsByCollection.get(col.name) || [];
    return contexts.length === 0;
  });
  if (collectionsWithoutContext.length > 0) {
    const names = collectionsWithoutContext.map(c => c.name).slice(0, 3).join(', ');
    const more = collectionsWithoutContext.length > 3 ? ` +${collectionsWithoutContext.length - 3} more` : '';
    tips.push(`Add context to collections for better search results: ${names}${more}`);
    tips.push(`  ${c.dim}qmd context add qmd://<name>/ "What this collection contains"${c.reset}`);
    tips.push(`  ${c.dim}qmd context add qmd://<name>/meeting-notes "Weekly team meeting notes"${c.reset}`);
  }

  // Check for collections without update commands
  const collectionsWithoutUpdate = collections.filter(col => {
    const yamlCol = getCollectionFromYaml(col.name);
    return !yamlCol?.update;
  });
  if (collectionsWithoutUpdate.length > 0 && collections.length > 1) {
    const names = collectionsWithoutUpdate.map(c => c.name).slice(0, 3).join(', ');
    const more = collectionsWithoutUpdate.length > 3 ? ` +${collectionsWithoutUpdate.length - 3} more` : '';
    tips.push(`Add update commands to keep collections fresh: ${names}${more}`);
    tips.push(`  ${c.dim}qmd collection update-cmd <name> 'git stash && git pull --rebase --ff-only && git stash pop'${c.reset}`);
  }

  if (tips.length > 0) {
    console.log(`\n${c.bold}Tips${c.reset}`);
    for (const tip of tips) {
      console.log(`  ${tip}`);
    }
  }

  closeDb();
}

async function updateCollections(): Promise<void> {
  const db = getDb();
  const storeInstance = getStore();
  // Collections are defined in YAML; no duplicate cleanup needed.

  // Clear Ollama cache on update
  clearCache(db);

  const collections = listCollections(db);

  if (collections.length === 0) {
    console.log(`${c.dim}No collections found. Run 'qmd collection add .' to index markdown files.${c.reset}`);
    closeDb();
    return;
  }

  console.log(`${c.bold}Updating ${collections.length} collection(s)...${c.reset}\n`);

  for (let i = 0; i < collections.length; i++) {
    const col = collections[i];
    if (!col) continue;
    console.log(`${c.cyan}[${i + 1}/${collections.length}]${c.reset} ${c.bold}${col.name}${c.reset} ${c.dim}(${col.glob_pattern})${c.reset}`);

    // Execute custom update command if specified in YAML
    const yamlCol = getCollectionFromYaml(col.name);
    if (yamlCol?.update) {
      console.log(`${c.dim}    Running update command: ${yamlCol.update}${c.reset}`);
      try {
        const proc = nodeSpawn("bash", ["-c", yamlCol.update], {
          cwd: col.pwd,
          stdio: ["ignore", "pipe", "pipe"],
        });

        const [output, errorOutput, exitCode] = await new Promise<[string, string, number]>((resolve, reject) => {
          let out = "";
          let err = "";
          proc.stdout?.on("data", (d: Buffer) => { out += d.toString(); });
          proc.stderr?.on("data", (d: Buffer) => { err += d.toString(); });
          proc.on("error", reject);
          proc.on("close", (code) => resolve([out, err, code ?? 1]));
        });

        if (output.trim()) {
          console.log(output.trim().split('\n').map(l => `    ${l}`).join('\n'));
        }
        if (errorOutput.trim()) {
          console.log(errorOutput.trim().split('\n').map(l => `    ${l}`).join('\n'));
        }

        if (exitCode !== 0) {
          console.log(`${c.yellow}✗ Update command failed with exit code ${exitCode}${c.reset}`);
          process.exit(exitCode);
        }
      } catch (err) {
        console.log(`${c.yellow}✗ Update command failed: ${err}${c.reset}`);
        process.exit(1);
      }
    }

    const startTime = Date.now();
    console.log(`Collection: ${col.pwd} (${col.glob_pattern})`);
    progress.indeterminate();

    const result = await reindexCollection(storeInstance, col.pwd, col.glob_pattern, col.name, {
      ignorePatterns: yamlCol?.ignore,
      onProgress: (info) => {
        progress.set((info.current / info.total) * 100);
        const elapsed = (Date.now() - startTime) / 1000;
        const rate = info.current / elapsed;
        const remaining = (info.total - info.current) / rate;
        const eta = info.current > 2 ? ` ETA: ${formatETA(remaining)}` : "";
        if (isTTY) process.stderr.write(`\rIndexing: ${info.current}/${info.total}${eta}        `);
      },
    });

    progress.clear();
    console.log(`\nIndexed: ${result.indexed} new, ${result.updated} updated, ${result.unchanged} unchanged, ${result.removed} removed`);
    if (result.orphanedCleaned > 0) {
      console.log(`Cleaned up ${result.orphanedCleaned} orphaned content hash(es)`);
    }
    console.log("");
  }

  // Check if any documents need embedding (show once at end)
  const needsEmbedding = getHashesNeedingEmbedding(db);
  closeDb();

  console.log(`${c.green}✓ All collections updated.${c.reset}`);
  if (needsEmbedding > 0) {
    console.log(`\nRun 'qmd embed' to update embeddings (${needsEmbedding} unique hashes need vectors)`);
  }
}

/**
 * Detect which collection (if any) contains the given filesystem path.
 * Returns { collectionId, collectionName, relativePath } or null if not in any collection.
 */
function detectCollectionFromPath(db: Database, fsPath: string): { collectionName: string; relativePath: string } | null {
  const realPath = getRealPath(fsPath);

  // Find collections that this path is under from YAML
  const allCollections = yamlListCollections();

  // Find longest matching path
  let bestMatch: { name: string; path: string } | null = null;
  for (const coll of allCollections) {
    if (realPath.startsWith(coll.path + '/') || realPath === coll.path) {
      if (!bestMatch || coll.path.length > bestMatch.path.length) {
        bestMatch = { name: coll.name, path: coll.path };
      }
    }
  }

  if (!bestMatch) return null;

  // Calculate relative path
  let relativePath = realPath;
  if (relativePath.startsWith(bestMatch.path + '/')) {
    relativePath = relativePath.slice(bestMatch.path.length + 1);
  } else if (relativePath === bestMatch.path) {
    relativePath = '';
  }

  return {
    collectionName: bestMatch.name,
    relativePath
  };
}

async function contextAdd(pathArg: string | undefined, contextText: string): Promise<void> {
  const db = getDb();

  // Handle "/" as global context (applies to all collections)
  if (pathArg === '/') {
    setGlobalContext(contextText);
    resyncConfig();
    console.log(`${c.green}✓${c.reset} Set global context`);
    console.log(`${c.dim}Context: ${contextText}${c.reset}`);
    closeDb();
    return;
  }

  // Resolve path - defaults to current directory if not provided
  let fsPath = pathArg || '.';
  if (fsPath === '.' || fsPath === './') {
    fsPath = getPwd();
  } else if (fsPath.startsWith('~/')) {
    fsPath = homedir() + fsPath.slice(1);
  } else if (!fsPath.startsWith('/') && !fsPath.startsWith('qmd://')) {
    fsPath = resolve(getPwd(), fsPath);
  }

  // Handle virtual paths (qmd://collection/path)
  if (isVirtualPath(fsPath)) {
    const parsed = parseVirtualPath(fsPath);
    if (!parsed) {
      console.error(`${c.yellow}Invalid virtual path: ${fsPath}${c.reset}`);
      process.exit(1);
    }

    const coll = getCollectionFromYaml(parsed.collectionName);
    if (!coll) {
      console.error(`${c.yellow}Collection not found: ${parsed.collectionName}${c.reset}`);
      process.exit(1);
    }

    yamlAddContext(parsed.collectionName, parsed.path, contextText);
    resyncConfig();

    const displayPath = parsed.path
      ? `qmd://${parsed.collectionName}/${parsed.path}`
      : `qmd://${parsed.collectionName}/ (collection root)`;
    console.log(`${c.green}✓${c.reset} Added context for: ${displayPath}`);
    console.log(`${c.dim}Context: ${contextText}${c.reset}`);
    closeDb();
    return;
  }

  // Detect collection from filesystem path
  const detected = detectCollectionFromPath(db, fsPath);
  if (!detected) {
    console.error(`${c.yellow}Path is not in any indexed collection: ${fsPath}${c.reset}`);
    console.error(`${c.dim}Run 'qmd status' to see indexed collections${c.reset}`);
    process.exit(1);
  }

  yamlAddContext(detected.collectionName, detected.relativePath, contextText);
  resyncConfig();

  const displayPath = detected.relativePath ? `qmd://${detected.collectionName}/${detected.relativePath}` : `qmd://${detected.collectionName}/`;
  console.log(`${c.green}✓${c.reset} Added context for: ${displayPath}`);
  console.log(`${c.dim}Context: ${contextText}${c.reset}`);
  closeDb();
}

function contextList(): void {
  const db = getDb();

  const allContexts = listAllContexts();

  if (allContexts.length === 0) {
    console.log(`${c.dim}No contexts configured. Use 'qmd context add' to add one.${c.reset}`);
    closeDb();
    return;
  }

  console.log(`\n${c.bold}Configured Contexts${c.reset}\n`);

  let lastCollection = '';
  for (const ctx of allContexts) {
    if (ctx.collection !== lastCollection) {
      console.log(`${c.cyan}${ctx.collection}${c.reset}`);
      lastCollection = ctx.collection;
    }

    const displayPath = ctx.path ? `  ${ctx.path}` : '  / (root)';
    console.log(`${displayPath}`);
    console.log(`    ${c.dim}${ctx.context}${c.reset}`);
  }

  closeDb();
}

function contextRemove(pathArg: string): void {
  if (pathArg === '/') {
    // Remove global context
    setGlobalContext(undefined);
    // Resync so SQLite store_config is updated
    const s = getStore();
    resyncConfig();
    closeDb();
    console.log(`${c.green}✓${c.reset} Removed global context`);
    return;
  }

  // Handle virtual paths
  if (isVirtualPath(pathArg)) {
    const parsed = parseVirtualPath(pathArg);
    if (!parsed) {
      console.error(`${c.yellow}Invalid virtual path: ${pathArg}${c.reset}`);
      process.exit(1);
    }

    const coll = getCollectionFromYaml(parsed.collectionName);
    if (!coll) {
      console.error(`${c.yellow}Collection not found: ${parsed.collectionName}${c.reset}`);
      process.exit(1);
    }

    const success = yamlRemoveContext(coll.name, parsed.path);

    if (!success) {
      console.error(`${c.yellow}No context found for: ${pathArg}${c.reset}`);
      process.exit(1);
    }

    console.log(`${c.green}✓${c.reset} Removed context for: ${pathArg}`);
    return;
  }

  // Handle filesystem paths
  let fsPath = pathArg;
  if (fsPath === '.' || fsPath === './') {
    fsPath = getPwd();
  } else if (fsPath.startsWith('~/')) {
    fsPath = homedir() + fsPath.slice(1);
  } else if (!fsPath.startsWith('/')) {
    fsPath = resolve(getPwd(), fsPath);
  }

  const db = getDb();
  const detected = detectCollectionFromPath(db, fsPath);
  closeDb();

  if (!detected) {
    console.error(`${c.yellow}Path is not in any indexed collection: ${fsPath}${c.reset}`);
    process.exit(1);
  }

  const success = yamlRemoveContext(detected.collectionName, detected.relativePath);

  if (!success) {
    console.error(`${c.yellow}No context found for: qmd://${detected.collectionName}/${detected.relativePath}${c.reset}`);
    process.exit(1);
  }

  console.log(`${c.green}✓${c.reset} Removed context for: qmd://${detected.collectionName}/${detected.relativePath}`);
}

function getDocument(filename: string, fromLine?: number, maxLines?: number, lineNumbers?: boolean): void {
  const db = getDb();

  // Parse :linenum suffix from filename (e.g., "file.md:100")
  let inputPath = filename;
  const colonMatch = inputPath.match(/:(\d+)$/);
  if (colonMatch && !fromLine) {
    const matched = colonMatch[1];
    if (matched) {
      fromLine = parseInt(matched, 10);
      inputPath = inputPath.slice(0, -colonMatch[0].length);
    }
  }

  // Handle docid lookup (#abc123, abc123, "#abc123", "abc123", etc.)
  if (isDocid(inputPath)) {
    const docidMatch = findDocumentByDocid(db, inputPath);
    if (docidMatch) {
      inputPath = docidMatch.filepath;
    } else {
      console.error(`Document not found: ${filename}`);
      closeDb();
      process.exit(1);
    }
  }

  let doc: { collectionName: string; path: string; body: string } | null = null;
  let virtualPath: string;

  // Handle virtual paths (qmd://collection/path)
  if (isVirtualPath(inputPath)) {
    const parsed = parseVirtualPath(inputPath);
    if (!parsed) {
      console.error(`Invalid virtual path: ${inputPath}`);
      closeDb();
      process.exit(1);
    }

    // Try exact match on collection + path
    doc = db.prepare(`
      SELECT d.collection as collectionName, d.path, content.doc as body
      FROM documents d
      JOIN content ON content.hash = d.hash
      WHERE d.collection = ? AND d.path = ? AND d.active = 1
    `).get(parsed.collectionName, parsed.path) as typeof doc;

    if (!doc) {
      // Try fuzzy match by path ending
      doc = db.prepare(`
        SELECT d.collection as collectionName, d.path, content.doc as body
        FROM documents d
        JOIN content ON content.hash = d.hash
        WHERE d.collection = ? AND d.path LIKE ? AND d.active = 1
        LIMIT 1
      `).get(parsed.collectionName, `%${parsed.path}`) as typeof doc;
    }

    virtualPath = inputPath;
  } else {
    // Try to interpret as collection/path format first (before filesystem path)
    // If path is relative (no / or ~ prefix), check if first component is a collection name
    if (!inputPath.startsWith('/') && !inputPath.startsWith('~')) {
      const parts = inputPath.split('/');
      if (parts.length >= 2) {
        const possibleCollection = parts[0];
        const possiblePath = parts.slice(1).join('/');

        // Check if this collection exists
        const collExists = possibleCollection ? db.prepare(`
          SELECT 1 FROM documents WHERE collection = ? AND active = 1 LIMIT 1
        `).get(possibleCollection) : null;

        if (collExists) {
          // Try exact match on collection + path
          doc = db.prepare(`
            SELECT d.collection as collectionName, d.path, content.doc as body
            FROM documents d
            JOIN content ON content.hash = d.hash
            WHERE d.collection = ? AND d.path = ? AND d.active = 1
          `).get(possibleCollection || "", possiblePath || "") as { collectionName: string; path: string; body: string } | null;

          if (!doc) {
            // Try fuzzy match by path ending
            doc = db.prepare(`
              SELECT d.collection as collectionName, d.path, content.doc as body
              FROM documents d
              JOIN content ON content.hash = d.hash
              WHERE d.collection = ? AND d.path LIKE ? AND d.active = 1
              LIMIT 1
            `).get(possibleCollection || "", `%${possiblePath}`) as { collectionName: string; path: string; body: string } | null;
          }

          if (doc) {
            virtualPath = buildVirtualPath(doc.collectionName, doc.path);
            // Skip the filesystem path handling below
          }
        }
      }
    }

    // If not found as collection/path, handle as filesystem paths
    if (!doc) {
      let fsPath = inputPath;

      // Expand ~ to home directory
      if (fsPath.startsWith('~/')) {
        fsPath = homedir() + fsPath.slice(1);
      } else if (!fsPath.startsWith('/')) {
        // Relative path - resolve from current directory
        fsPath = resolve(getPwd(), fsPath);
      }
      fsPath = getRealPath(fsPath);

      // Try to detect which collection contains this path
      const detected = detectCollectionFromPath(db, fsPath);

      if (detected) {
        // Found collection - query by collection name + relative path
        doc = db.prepare(`
          SELECT d.collection as collectionName, d.path, content.doc as body
          FROM documents d
          JOIN content ON content.hash = d.hash
          WHERE d.collection = ? AND d.path = ? AND d.active = 1
        `).get(detected.collectionName, detected.relativePath) as { collectionName: string; path: string; body: string } | null;
      }

      // Fuzzy match by filename (last component of path)
      if (!doc) {
        const filename = inputPath.split('/').pop() || inputPath;
        doc = db.prepare(`
          SELECT d.collection as collectionName, d.path, content.doc as body
          FROM documents d
          JOIN content ON content.hash = d.hash
          WHERE d.path LIKE ? AND d.active = 1
          LIMIT 1
        `).get(`%${filename}`) as { collectionName: string; path: string; body: string } | null;
      }

      if (doc) {
        virtualPath = buildVirtualPath(doc.collectionName, doc.path);
      } else {
        virtualPath = inputPath;
      }
    }
  }

  // Ensure doc is not null before proceeding
  if (!doc) {
    console.error(`Document not found: ${filename}`);
    closeDb();
    process.exit(1);
  }

  // Get context for this file
  const context = getContextForPath(db, doc.collectionName, doc.path);

  let output = doc.body;
  const startLine = fromLine || 1;

  // Apply line filtering if specified
  if (fromLine !== undefined || maxLines !== undefined) {
    const lines = output.split('\n');
    const start = startLine - 1; // Convert to 0-indexed
    const end = maxLines !== undefined ? start + maxLines : lines.length;
    output = lines.slice(start, end).join('\n');
  }

  // Add line numbers if requested
  if (lineNumbers) {
    output = addLineNumbers(output, startLine);
  }

  // Output context header if exists
  if (context) {
    console.log(`Folder Context: ${context}\n---\n`);
  }
  console.log(output);
  closeDb();
}

// Multi-get: fetch multiple documents by glob pattern or comma-separated list
function multiGet(pattern: string, maxLines?: number, maxBytes: number = DEFAULT_MULTI_GET_MAX_BYTES, format: OutputFormat = "cli"): void {
  const db = getDb();

  // Check if it's a comma-separated list or a glob pattern
  const isCommaSeparated = pattern.includes(',') && !pattern.includes('*') && !pattern.includes('?') && !pattern.includes('{');

  let files: { filepath: string; displayPath: string; bodyLength: number; collection?: string; path?: string }[];

  if (isCommaSeparated) {
    // Comma-separated list of files (can be virtual paths or relative paths)
    const names = pattern.split(',').map(s => s.trim()).filter(Boolean);
    files = [];
    for (const name of names) {
      let doc: { virtual_path: string; body_length: number; collection: string; path: string } | null = null;

      // Handle virtual paths
      if (isVirtualPath(name)) {
        const parsed = parseVirtualPath(name);
        if (parsed) {
          // Try exact match on collection + path
          doc = db.prepare(`
            SELECT
              'qmd://' || d.collection || '/' || d.path as virtual_path,
              LENGTH(content.doc) as body_length,
              d.collection,
              d.path
            FROM documents d
            JOIN content ON content.hash = d.hash
            WHERE d.collection = ? AND d.path = ? AND d.active = 1
          `).get(parsed.collectionName, parsed.path) as typeof doc;
        }
      } else {
        // Try exact match on path
        doc = db.prepare(`
          SELECT
            'qmd://' || d.collection || '/' || d.path as virtual_path,
            LENGTH(content.doc) as body_length,
            d.collection,
            d.path
          FROM documents d
          JOIN content ON content.hash = d.hash
          WHERE d.path = ? AND d.active = 1
          LIMIT 1
        `).get(name) as { virtual_path: string; body_length: number; collection: string; path: string } | null;

        // Try suffix match
        if (!doc) {
          doc = db.prepare(`
            SELECT
              'qmd://' || d.collection || '/' || d.path as virtual_path,
              LENGTH(content.doc) as body_length,
              d.collection,
              d.path
            FROM documents d
            JOIN content ON content.hash = d.hash
            WHERE d.path LIKE ? AND d.active = 1
            LIMIT 1
          `).get(`%${name}`) as { virtual_path: string; body_length: number; collection: string; path: string } | null;
        }
      }

      if (doc) {
        files.push({
          filepath: doc.virtual_path,
          displayPath: doc.virtual_path,
          bodyLength: doc.body_length,
          collection: doc.collection,
          path: doc.path
        });
      } else {
        console.error(`File not found: ${name}`);
      }
    }
  } else {
    // Glob pattern - matchFilesByGlob now returns virtual paths
    files = matchFilesByGlob(db, pattern).map(f => ({
      ...f,
      collection: undefined,  // Will be fetched later if needed
      path: undefined
    }));
    if (files.length === 0) {
      console.error(`No files matched pattern: ${pattern}`);
      closeDb();
      process.exit(1);
    }
  }

  // Collect results for structured output
  const results: { file: string; displayPath: string; title: string; body: string; context: string | null; skipped: boolean; skipReason?: string }[] = [];

  for (const file of files) {
    // Parse virtual path to get collection info if not already available
    let collection = file.collection;
    let path = file.path;

    if (!collection || !path) {
      const parsed = parseVirtualPath(file.filepath);
      if (parsed) {
        collection = parsed.collectionName;
        path = parsed.path;
      }
    }

    // Get context using collection-scoped function
    const context = collection && path ? getContextForPath(db, collection, path) : null;

    // Check size limit
    if (file.bodyLength > maxBytes) {
      results.push({
        file: file.filepath,
        displayPath: file.displayPath,
        title: file.displayPath.split('/').pop() || file.displayPath,
        body: "",
        context,
        skipped: true,
        skipReason: `File too large (${Math.round(file.bodyLength / 1024)}KB > ${Math.round(maxBytes / 1024)}KB). Use 'qmd get ${file.displayPath}' to retrieve.`,
      });
      continue;
    }

    // Fetch document content using collection and path
    if (!collection || !path) continue;

    const doc = db.prepare(`
      SELECT content.doc as body, d.title
      FROM documents d
      JOIN content ON content.hash = d.hash
      WHERE d.collection = ? AND d.path = ? AND d.active = 1
    `).get(collection, path) as { body: string; title: string } | null;

    if (!doc) continue;

    let body = doc.body;

    // Apply line limit if specified
    if (maxLines !== undefined) {
      const lines = body.split('\n');
      body = lines.slice(0, maxLines).join('\n');
      if (lines.length > maxLines) {
        body += `\n\n[... truncated ${lines.length - maxLines} more lines]`;
      }
    }

    results.push({
      file: file.filepath,
      displayPath: file.displayPath,
      title: doc.title || file.displayPath.split('/').pop() || file.displayPath,
      body,
      context,
      skipped: false,
    });
  }

  closeDb();

  // Output based on format
  if (format === "json") {
    const output = results.map(r => ({
      file: r.displayPath,
      title: r.title,
      ...(r.context && { context: r.context }),
      ...(r.skipped ? { skipped: true, reason: r.skipReason } : { body: r.body }),
    }));
    console.log(JSON.stringify(output, null, 2));
  } else if (format === "csv") {
    const escapeField = (val: string | null | undefined): string => {
      if (val === null || val === undefined) return "";
      const str = String(val);
      if (str.includes(",") || str.includes('"') || str.includes("\n")) {
        return `"${str.replace(/"/g, '""')}"`;
      }
      return str;
    };
    console.log("file,title,context,skipped,body");
    for (const r of results) {
      console.log([r.displayPath, r.title, r.context, r.skipped ? "true" : "false", r.skipped ? r.skipReason : r.body].map(escapeField).join(","));
    }
  } else if (format === "files") {
    for (const r of results) {
      const ctx = r.context ? `,"${r.context.replace(/"/g, '""')}"` : "";
      const status = r.skipped ? "[SKIPPED]" : "";
      console.log(`${r.displayPath}${ctx}${status ? `,${status}` : ""}`);
    }
  } else if (format === "md") {
    for (const r of results) {
      console.log(`## ${r.displayPath}\n`);
      if (r.title && r.title !== r.displayPath) console.log(`**Title:** ${r.title}\n`);
      if (r.context) console.log(`**Context:** ${r.context}\n`);
      if (r.skipped) {
        console.log(`> ${r.skipReason}\n`);
      } else {
        console.log("```");
        console.log(r.body);
        console.log("```\n");
      }
    }
  } else if (format === "xml") {
    console.log('<?xml version="1.0" encoding="UTF-8"?>');
    console.log("<documents>");
    for (const r of results) {
      console.log("  <document>");
      console.log(`    <file>${escapeXml(r.displayPath)}</file>`);
      console.log(`    <title>${escapeXml(r.title)}</title>`);
      if (r.context) console.log(`    <context>${escapeXml(r.context)}</context>`);
      if (r.skipped) {
        console.log(`    <skipped>true</skipped>`);
        console.log(`    <reason>${escapeXml(r.skipReason || "")}</reason>`);
      } else {
        console.log(`    <body>${escapeXml(r.body)}</body>`);
      }
      console.log("  </document>");
    }
    console.log("</documents>");
  } else {
    // CLI format (default)
    for (const r of results) {
      console.log(`\n${'='.repeat(60)}`);
      console.log(`File: ${r.displayPath}`);
      console.log(`${'='.repeat(60)}\n`);

      if (r.skipped) {
        console.log(`[SKIPPED: ${r.skipReason}]`);
        continue;
      }

      if (r.context) {
        console.log(`Folder Context: ${r.context}\n---\n`);
      }
      console.log(r.body);
    }
  }
}

// List files in virtual file tree
function listFiles(pathArg?: string): void {
  const db = getDb();

  if (!pathArg) {
    // No argument - list all collections
    const yamlCollections = yamlListCollections();

    if (yamlCollections.length === 0) {
      console.log("No collections found. Run 'qmd collection add .' to index files.");
      closeDb();
      return;
    }

    // Get file counts from database for each collection
    const collections = yamlCollections.map(coll => {
      const stats = db.prepare(`
        SELECT COUNT(*) as file_count
        FROM documents d
        WHERE d.collection = ? AND d.active = 1
      `).get(coll.name) as { file_count: number } | null;

      return {
        name: coll.name,
        file_count: stats?.file_count || 0
      };
    });

    console.log(`${c.bold}Collections:${c.reset}\n`);
    for (const coll of collections) {
      console.log(`  ${c.dim}qmd://${c.reset}${c.cyan}${coll.name}/${c.reset}  ${c.dim}(${coll.file_count} files)${c.reset}`);
    }
    closeDb();
    return;
  }

  // Parse the path argument
  let collectionName: string;
  let pathPrefix: string | null = null;

  if (pathArg.startsWith('qmd://')) {
    // Virtual path format: qmd://collection/path
    const parsed = parseVirtualPath(pathArg);
    if (!parsed) {
      console.error(`Invalid virtual path: ${pathArg}`);
      closeDb();
      process.exit(1);
    }
    collectionName = parsed.collectionName;
    pathPrefix = parsed.path;
  } else {
    // Just collection name or collection/path
    const parts = pathArg.split('/');
    collectionName = parts[0] || '';
    if (parts.length > 1) {
      pathPrefix = parts.slice(1).join('/');
    }
  }

  // Get the collection
  const coll = getCollectionFromYaml(collectionName);
  if (!coll) {
    console.error(`Collection not found: ${collectionName}`);
    console.error(`Run 'qmd ls' to see available collections.`);
    closeDb();
    process.exit(1);
  }

  // List files in the collection with size and modification time
  let query: string;
  let params: any[];

  if (pathPrefix) {
    // List files under a specific path
    query = `
      SELECT d.path, d.title, d.modified_at, LENGTH(ct.doc) as size
      FROM documents d
      JOIN content ct ON d.hash = ct.hash
      WHERE d.collection = ? AND d.path LIKE ? AND d.active = 1
      ORDER BY d.path
    `;
    params = [coll.name, `${pathPrefix}%`];
  } else {
    // List all files in the collection
    query = `
      SELECT d.path, d.title, d.modified_at, LENGTH(ct.doc) as size
      FROM documents d
      JOIN content ct ON d.hash = ct.hash
      WHERE d.collection = ? AND d.active = 1
      ORDER BY d.path
    `;
    params = [coll.name];
  }

  const files = db.prepare(query).all(...params) as { path: string; title: string; modified_at: string; size: number }[];

  if (files.length === 0) {
    if (pathPrefix) {
      console.log(`No files found under qmd://${collectionName}/${pathPrefix}`);
    } else {
      console.log(`No files found in collection: ${collectionName}`);
    }
    closeDb();
    return;
  }

  // Calculate max widths for alignment
  const maxSize = Math.max(...files.map(f => formatBytes(f.size).length));

  // Output in ls -l style
  for (const file of files) {
    const sizeStr = formatBytes(file.size).padStart(maxSize);
    const date = new Date(file.modified_at);
    const timeStr = formatLsTime(date);

    // Dim the qmd:// prefix, highlight the filename
    console.log(`${sizeStr}  ${timeStr}  ${c.dim}qmd://${collectionName}/${c.reset}${c.cyan}${file.path}${c.reset}`);
  }

  closeDb();
}

// Format date/time like ls -l
function formatLsTime(date: Date): string {
  const now = new Date();
  const sixMonthsAgo = new Date(now.getTime() - 6 * 30 * 24 * 60 * 60 * 1000);

  const months = ['Jan', 'Feb', 'Mar', 'Apr', 'May', 'Jun', 'Jul', 'Aug', 'Sep', 'Oct', 'Nov', 'Dec'];
  const month = months[date.getMonth()];
  const day = date.getDate().toString().padStart(2, ' ');

  // If file is older than 6 months, show year instead of time
  if (date < sixMonthsAgo) {
    const year = date.getFullYear();
    return `${month} ${day}  ${year}`;
  } else {
    const hours = date.getHours().toString().padStart(2, '0');
    const minutes = date.getMinutes().toString().padStart(2, '0');
    return `${month} ${day} ${hours}:${minutes}`;
  }
}

// Collection management commands
function collectionList(): void {
  const db = getDb();
  const collections = listCollections(db);

  if (collections.length === 0) {
    console.log("No collections found. Run 'qmd collection add .' to create one.");
    closeDb();
    return;
  }

  console.log(`${c.bold}Collections (${collections.length}):${c.reset}\n`);

  for (const coll of collections) {
    const updatedAt = coll.last_modified ? new Date(coll.last_modified) : new Date();
    const timeAgo = formatTimeAgo(updatedAt);
    
    // Get YAML config to check includeByDefault
    const yamlColl = getCollectionFromYaml(coll.name);
    const excluded = yamlColl?.includeByDefault === false;
    const excludeTag = excluded ? ` ${c.yellow}[excluded]${c.reset}` : '';

    console.log(`${c.cyan}${coll.name}${c.reset} ${c.dim}(qmd://${coll.name}/)${c.reset}${excludeTag}`);
    console.log(`  ${c.dim}Pattern:${c.reset}  ${coll.glob_pattern}`);
    if (yamlColl?.ignore?.length) {
      console.log(`  ${c.dim}Ignore:${c.reset}   ${yamlColl.ignore.join(', ')}`);
    }
    console.log(`  ${c.dim}Files:${c.reset}    ${coll.active_count}`);
    console.log(`  ${c.dim}Updated:${c.reset}  ${timeAgo}`);
    console.log();
  }

  closeDb();
}

async function collectionAdd(pwd: string, globPattern: string, name?: string): Promise<void> {
  // If name not provided, generate from pwd basename
  let collName = name;
  if (!collName) {
    const parts = pwd.split('/').filter(Boolean);
    collName = parts[parts.length - 1] || 'root';
  }

  // Check if collection with this name already exists in YAML
  const existing = getCollectionFromYaml(collName);
  if (existing) {
    console.error(`${c.yellow}Collection '${collName}' already exists.${c.reset}`);
    console.error(`Use a different name with --name <name>`);
    process.exit(1);
  }

  // Check if a collection with this pwd+glob already exists in YAML
  const allCollections = yamlListCollections();
  const existingPwdGlob = allCollections.find(c => c.path === pwd && c.pattern === globPattern);

  if (existingPwdGlob) {
    console.error(`${c.yellow}A collection already exists for this path and pattern:${c.reset}`);
    console.error(`  Name: ${existingPwdGlob.name} (qmd://${existingPwdGlob.name}/)`);
    console.error(`  Pattern: ${globPattern}`);
    console.error(`\nUse 'qmd update' to re-index it, or remove it first with 'qmd collection remove ${existingPwdGlob.name}'`);
    process.exit(1);
  }

  // Add to YAML config + sync to SQLite
  const { addCollection } = await import("../collections.js");
  addCollection(collName, pwd, globPattern);
  resyncConfig();

  // Create the collection and index files
  console.log(`Creating collection '${collName}'...`);
  const newColl = getCollectionFromYaml(collName);
  await indexFiles(pwd, globPattern, collName, false, newColl?.ignore);
  console.log(`${c.green}✓${c.reset} Collection '${collName}' created successfully`);
}

function collectionRemove(name: string): void {
  // Check if collection exists in YAML
  const coll = getCollectionFromYaml(name);
  if (!coll) {
    console.error(`${c.yellow}Collection not found: ${name}${c.reset}`);
    console.error(`Run 'qmd collection list' to see available collections.`);
    process.exit(1);
  }

  const db = getDb();
  const result = removeCollection(db, name);
  // Also remove from YAML config
  yamlRemoveCollectionFn(name);
  closeDb();

  console.log(`${c.green}✓${c.reset} Removed collection '${name}'`);
  console.log(`  Deleted ${result.deletedDocs} documents`);
  if (result.cleanedHashes > 0) {
    console.log(`  Cleaned up ${result.cleanedHashes} orphaned content hashes`);
  }
}

function collectionRename(oldName: string, newName: string): void {
  // Check if old collection exists in YAML
  const coll = getCollectionFromYaml(oldName);
  if (!coll) {
    console.error(`${c.yellow}Collection not found: ${oldName}${c.reset}`);
    console.error(`Run 'qmd collection list' to see available collections.`);
    process.exit(1);
  }

  // Check if new name already exists in YAML
  const existing = getCollectionFromYaml(newName);
  if (existing) {
    console.error(`${c.yellow}Collection name already exists: ${newName}${c.reset}`);
    console.error(`Choose a different name or remove the existing collection first.`);
    process.exit(1);
  }

  const db = getDb();
  renameCollection(db, oldName, newName);
  // Also rename in YAML config
  yamlRenameCollectionFn(oldName, newName);
  closeDb();

  console.log(`${c.green}✓${c.reset} Renamed collection '${oldName}' to '${newName}'`);
  console.log(`  Virtual paths updated: ${c.cyan}qmd://${oldName}/${c.reset} → ${c.cyan}qmd://${newName}/${c.reset}`);
}

async function indexFiles(pwd?: string, globPattern: string = DEFAULT_GLOB, collectionName?: string, suppressEmbedNotice: boolean = false, ignorePatterns?: string[]): Promise<void> {
  const db = getDb();
  const resolvedPwd = pwd || getPwd();
  const now = new Date().toISOString();
  const excludeDirs = ["node_modules", ".git", ".cache", "vendor", "dist", "build"];

  // Clear Ollama cache on index
  clearCache(db);

  // Collection name must be provided (from YAML)
  if (!collectionName) {
    throw new Error("Collection name is required. Collections must be defined in ~/.config/qmd/index.yml");
  }

  console.log(`Collection: ${resolvedPwd} (${globPattern})`);

  progress.indeterminate();
  const allIgnore = [
    ...excludeDirs.map(d => `**/${d}/**`),
    ...(ignorePatterns || []),
  ];
  const allFiles: string[] = await fastGlob(globPattern, {
    cwd: resolvedPwd,
    onlyFiles: true,
    followSymbolicLinks: false,
    dot: false,
    ignore: allIgnore,
  });
  // Filter hidden files/folders (dot: false handles top-level but not nested)
  const files = allFiles.filter(file => {
    const parts = file.split("/");
    return !parts.some(part => part.startsWith("."));
  });

  const total = files.length;
  const hasNoFiles = total === 0;
  if (hasNoFiles) {
    progress.clear();
    console.log("No files found matching pattern.");
    // Continue so the deactivation pass can mark previously indexed docs as inactive.
  }

  let indexed = 0, updated = 0, unchanged = 0, processed = 0;
  const seenPaths = new Set<string>();
  const startTime = Date.now();

  for (const relativeFile of files) {
    const filepath = getRealPath(resolve(resolvedPwd, relativeFile));
    const path = handelize(relativeFile); // Normalize path for token-friendliness
    seenPaths.add(path);

    let content: string;
    try {
      content = readFileSync(filepath, "utf-8");
    } catch (err: any) {
      // Skip files that can't be read (e.g. iCloud evicted files returning EAGAIN)
      processed++;
      progress.set((processed / total) * 100);
      continue;
    }

    // Skip empty files - nothing useful to index
    if (!content.trim()) {
      processed++;
      continue;
    }

    const hash = await hashContent(content);
    const title = extractTitle(content, relativeFile);

    // Check if document exists in this collection with this path
    const existing = findActiveDocument(db, collectionName, path);

    if (existing) {
      if (existing.hash === hash) {
        // Hash unchanged, but check if title needs updating
        if (existing.title !== title) {
          updateDocumentTitle(db, existing.id, title, now);
          updated++;
        } else {
          unchanged++;
        }
      } else {
        // Content changed - insert new content hash and update document
        insertContent(db, hash, content, now);
        const stat = statSync(filepath);
        updateDocument(db, existing.id, title, hash,
          stat ? new Date(stat.mtime).toISOString() : now);
        updated++;
      }
    } else {
      // New document - insert content and document
      indexed++;
      insertContent(db, hash, content, now);
      const stat = statSync(filepath);
      insertDocument(db, collectionName, path, title, hash,
        stat ? new Date(stat.birthtime).toISOString() : now,
        stat ? new Date(stat.mtime).toISOString() : now);
    }

    processed++;
    progress.set((processed / total) * 100);
    const elapsed = (Date.now() - startTime) / 1000;
    const rate = processed / elapsed;
    const remaining = (total - processed) / rate;
    const eta = processed > 2 ? ` ETA: ${formatETA(remaining)}` : "";
    if (isTTY) process.stderr.write(`\rIndexing: ${processed}/${total}${eta}        `);
  }

  // Deactivate documents in this collection that no longer exist
  const allActive = getActiveDocumentPaths(db, collectionName);
  let removed = 0;
  for (const path of allActive) {
    if (!seenPaths.has(path)) {
      deactivateDocument(db, collectionName, path);
      removed++;
    }
  }

  // Clean up orphaned content hashes (content not referenced by any document)
  const orphanedContent = cleanupOrphanedContent(db);

  // Check if vector index needs updating
  const needsEmbedding = getHashesNeedingEmbedding(db);

  progress.clear();
  console.log(`\nIndexed: ${indexed} new, ${updated} updated, ${unchanged} unchanged, ${removed} removed`);
  if (orphanedContent > 0) {
    console.log(`Cleaned up ${orphanedContent} orphaned content hash(es)`);
  }

  if (needsEmbedding > 0 && !suppressEmbedNotice) {
    console.log(`\nRun 'qmd embed' to update embeddings (${needsEmbedding} unique hashes need vectors)`);
  }

  closeDb();
}

function renderProgressBar(percent: number, width: number = 30): string {
  const filled = Math.round((percent / 100) * width);
  const empty = width - filled;
  const bar = "█".repeat(filled) + "░".repeat(empty);
  return bar;
}

function parseEmbedBatchOption(name: string, value: unknown): number | undefined {
  if (value === undefined) return undefined;
  const parsed = Number(value);
  if (!Number.isInteger(parsed) || parsed < 1) {
    throw new Error(`${name} must be a positive integer`);
  }
  return parsed;
}

function parseChunkStrategy(value: unknown): ChunkStrategy | undefined {
  if (value === undefined) return undefined;
  const s = String(value);
  if (s === "auto" || s === "regex") return s;
  throw new Error(`--chunk-strategy must be "auto" or "regex" (got "${s}")`);
}

async function vectorIndex(
  model: string = DEFAULT_EMBED_MODEL_URI,
  force: boolean = false,
  batchOptions?: { maxDocsPerBatch?: number; maxBatchBytes?: number; chunkStrategy?: ChunkStrategy },
): Promise<void> {
  const storeInstance = getStore();
  const db = storeInstance.db;

  if (force) {
    console.log(`${c.yellow}Force re-indexing: clearing all vectors...${c.reset}`);
  }

  // Check if there's work to do before starting
  const hashesToEmbed = getHashesNeedingEmbedding(db);
  if (hashesToEmbed === 0 && !force) {
    console.log(`${c.green}✓ All content hashes already have embeddings.${c.reset}`);
    closeDb();
    return;
  }

  console.log(`${c.dim}Model: ${model}${c.reset}\n`);
  if (batchOptions?.maxDocsPerBatch !== undefined || batchOptions?.maxBatchBytes !== undefined) {
    const maxDocsPerBatch = batchOptions.maxDocsPerBatch ?? DEFAULT_EMBED_MAX_DOCS_PER_BATCH;
    const maxBatchBytes = batchOptions.maxBatchBytes ?? DEFAULT_EMBED_MAX_BATCH_BYTES;
    console.log(`${c.dim}Batch: ${maxDocsPerBatch} docs / ${formatBytes(maxBatchBytes)}${c.reset}\n`);
  }
  cursor.hide();
  progress.indeterminate();

  const startTime = Date.now();

  const result = await generateEmbeddings(storeInstance, {
    force,
    model,
    maxDocsPerBatch: batchOptions?.maxDocsPerBatch,
    maxBatchBytes: batchOptions?.maxBatchBytes,
    chunkStrategy: batchOptions?.chunkStrategy,
    onProgress: (info) => {
      if (info.totalBytes === 0) return;
      const percent = (info.bytesProcessed / info.totalBytes) * 100;
      progress.set(percent);

      const elapsed = (Date.now() - startTime) / 1000;
      const bytesPerSec = info.bytesProcessed / elapsed;
      const remainingBytes = info.totalBytes - info.bytesProcessed;
      const etaSec = remainingBytes / bytesPerSec;

      const bar = renderProgressBar(percent);
      const percentStr = percent.toFixed(0).padStart(3);
      const throughput = `${formatBytes(bytesPerSec)}/s`;
      const eta = elapsed > 2 ? formatETA(etaSec) : "...";
      const errStr = info.errors > 0 ? ` ${c.yellow}${info.errors} err${c.reset}` : "";

      if (isTTY) process.stderr.write(`\r${c.cyan}${bar}${c.reset} ${c.bold}${percentStr}%${c.reset} ${c.dim}${info.chunksEmbedded}/${info.totalChunks}${c.reset}${errStr} ${c.dim}${throughput} ETA ${eta}${c.reset}   `);
    },
  });

  progress.clear();
  cursor.show();

  const totalTimeSec = result.durationMs / 1000;

  if (result.chunksEmbedded === 0 && result.docsProcessed === 0) {
    console.log(`${c.green}✓ No non-empty documents to embed.${c.reset}`);
  } else {
    console.log(`\r${c.green}${renderProgressBar(100)}${c.reset} ${c.bold}100%${c.reset}                                    `);
    console.log(`\n${c.green}✓ Done!${c.reset} Embedded ${c.bold}${result.chunksEmbedded}${c.reset} chunks from ${c.bold}${result.docsProcessed}${c.reset} documents in ${c.bold}${formatETA(totalTimeSec)}${c.reset}`);
    if (result.errors > 0) {
      console.log(`${c.yellow}⚠ ${result.errors} chunks failed${c.reset}`);
    }
  }

  closeDb();
}

// Sanitize a term for FTS5: remove punctuation except apostrophes
function sanitizeFTS5Term(term: string): string {
  // Remove all non-alphanumeric except apostrophes (for contractions like "don't")
  return term.replace(/[^\w']/g, '').trim();
}

// Build FTS5 query: phrase-aware with fallback to individual terms
function buildFTS5Query(query: string): string {
  // Sanitize the full query for phrase matching
  const sanitizedQuery = query.replace(/[^\w\s']/g, '').trim();

  const terms = query
    .split(/\s+/)
    .map(sanitizeFTS5Term)
    .filter(term => term.length >= 2); // Skip single chars and empty

  if (terms.length === 0) return "";
  if (terms.length === 1) return `"${terms[0]!.replace(/"/g, '""')}"`;

  // Strategy: exact phrase OR proximity match OR individual terms
  // Exact phrase matches rank highest, then close proximity, then any term
  const phrase = `"${sanitizedQuery.replace(/"/g, '""')}"`;
  const quotedTerms = terms.map(t => `"${t.replace(/"/g, '""')}"`);

  // FTS5 NEAR syntax: NEAR(term1 term2, distance)
  const nearPhrase = `NEAR(${quotedTerms.join(' ')}, 10)`;
  const orTerms = quotedTerms.join(' OR ');

  // Exact phrase > proximity > any term
  return `(${phrase}) OR (${nearPhrase}) OR (${orTerms})`;
}

// Normalize BM25 score to 0-1 range using sigmoid
function normalizeBM25(score: number): number {
  // BM25 scores are negative in SQLite (lower = better)
  // Typical range: -15 (excellent) to -2 (weak match)
  // Map to 0-1 where higher is better
  const absScore = Math.abs(score);
  // Sigmoid-ish normalization: maps ~2-15 range to ~0.1-0.95
  return 1 / (1 + Math.exp(-(absScore - 5) / 3));
}

type OutputOptions = {
  format: OutputFormat;
  full: boolean;
  limit: number;
  minScore: number;
  all?: boolean;
  collection?: string | string[];  // Filter by collection name(s)
  lineNumbers?: boolean; // Add line numbers to output
  explain?: boolean;     // Include retrieval score traces (query only)
  context?: string;      // Optional context for query expansion
  candidateLimit?: number;  // Max candidates to rerank (default: 40)
  intent?: string;       // Domain intent for disambiguation
  skipRerank?: boolean;  // Skip LLM reranking, use RRF scores only
  chunkStrategy?: ChunkStrategy;  // "auto" (default) or "regex"
};

// Highlight query terms in text (skip short words < 3 chars)
function highlightTerms(text: string, query: string): string {
  if (!useColor) return text;
  const terms = query.toLowerCase().split(/\s+/).filter(t => t.length >= 3);
  let result = text;
  for (const term of terms) {
    const regex = new RegExp(`(${term.replace(/[.*+?^${}()|[\]\\]/g, '\\$&')})`, 'gi');
    result = result.replace(regex, `${c.yellow}${c.bold}$1${c.reset}`);
  }
  return result;
}

// Format score with color based on value
function formatScore(score: number): string {
  const pct = (score * 100).toFixed(0).padStart(3);
  if (!useColor) return `${pct}%`;
  if (score >= 0.7) return `${c.green}${pct}%${c.reset}`;
  if (score >= 0.4) return `${c.yellow}${pct}%${c.reset}`;
  return `${c.dim}${pct}%${c.reset}`;
}

function formatExplainNumber(value: number): string {
  return value.toFixed(4);
}

// Shorten directory path for display - relative to $HOME (used for context paths, not documents)
function shortPath(dirpath: string): string {
  const home = homedir();
  if (dirpath.startsWith(home)) {
    return '~' + dirpath.slice(home.length);
  }
  return dirpath;
}

type EmptySearchReason = "no_results" | "min_score";

// Emit format-safe empty output for search commands.
function printEmptySearchResults(format: OutputFormat, reason: EmptySearchReason = "no_results"): void {
  if (format === "json") {
    console.log("[]");
    return;
  }
  if (format === "csv") {
    console.log("docid,score,file,title,context,line,snippet");
    return;
  }
  if (format === "xml") {
    console.log("<results></results>");
    return;
  }
  if (format === "md" || format === "files") {
    return;
  }

  if (reason === "min_score") {
    console.log("No results found above minimum score threshold.");
    return;
  }
  console.log("No results found.");
}

type OutputRow = {
  file: string;
  displayPath: string;
  title: string;
  body: string;
  score: number;
  context?: string | null;
  chunkPos?: number;
  hash?: string;
  docid?: string;
  explain?: HybridQueryExplain;
};

const DEFAULT_EDITOR_URI_TEMPLATE = "vscode://file/{path}:{line}:{col}";

function encodePathForEditorUri(absolutePath: string): string {
  return encodeURI(absolutePath)
    .replace(/\?/g, "%3F")
    .replace(/#/g, "%23");
}

function getEditorUriTemplate(): string {
  const envTemplate = process.env.QMD_EDITOR_URI?.trim();
  if (envTemplate) return envTemplate;

  try {
    const config = loadConfig() as unknown as {
      editor_uri?: string;
      editor_uri_template?: string;
      editorUri?: string;
      [key: string]: unknown;
    };
    const configTemplate = (
      config.editor_uri
      || config.editor_uri_template
      || config.editorUri
      || (typeof config["editor-uri"] === "string" ? config["editor-uri"] : undefined)
    )?.trim();

    if (configTemplate) return configTemplate;
  } catch {
    // Ignore config parsing issues and use default template.
  }

  return DEFAULT_EDITOR_URI_TEMPLATE;
}

export function buildEditorUri(template: string, absolutePath: string, line: number, col: number): string {
  const safeLine = Number.isFinite(line) && line > 0 ? Math.floor(line) : 1;
  const safeCol = Number.isFinite(col) && col > 0 ? Math.floor(col) : 1;
  const encodedPath = encodePathForEditorUri(absolutePath);

  return template
    .replace(/\{path\}/g, encodedPath)
    .replace(/\{line\}/g, String(safeLine))
    .replace(/\{col\}/g, String(safeCol))
    .replace(/\{column\}/g, String(safeCol));
}

export function termLink(text: string, url: string, isTTY: boolean = !!process.stdout.isTTY): string {
  if (!isTTY) return text;
  return `\x1b]8;;${url}\x07${text}\x1b]8;;\x07`;
}

function outputResults(results: OutputRow[], query: string, opts: OutputOptions): void {
  const filtered = results.filter(r => r.score >= opts.minScore).slice(0, opts.limit);

  if (filtered.length === 0) {
    printEmptySearchResults(opts.format, "min_score");
    return;
  }

  // Helper to create qmd:// URI from displayPath
  const toQmdPath = (displayPath: string) => `qmd://${displayPath}`;

  if (opts.format === "json") {
    // JSON output for LLM consumption
    const output = filtered.map(row => {
      const docid = row.docid || (row.hash ? row.hash.slice(0, 6) : undefined);
      let body = opts.full ? row.body : undefined;
      let snippet = !opts.full ? extractSnippet(row.body, query, 300, row.chunkPos, undefined, opts.intent).snippet : undefined;
      if (opts.lineNumbers) {
        if (body) body = addLineNumbers(body);
        if (snippet) snippet = addLineNumbers(snippet);
      }
      return {
        ...(docid && { docid: `#${docid}` }),
        score: Math.round(row.score * 100) / 100,
        file: toQmdPath(row.displayPath),
        title: row.title,
        ...(row.context && { context: row.context }),
        ...(body && { body }),
        ...(snippet && { snippet }),
        ...(opts.explain && row.explain && { explain: row.explain }),
      };
    });
    console.log(JSON.stringify(output, null, 2));
  } else if (opts.format === "files") {
    // Simple docid,score,filepath,context output
    for (const row of filtered) {
      const docid = row.docid || (row.hash ? row.hash.slice(0, 6) : "");
      const ctx = row.context ? `,"${row.context.replace(/"/g, '""')}"` : "";
      console.log(`#${docid},${row.score.toFixed(2)},${toQmdPath(row.displayPath)}${ctx}`);
    }
  } else if (opts.format === "cli") {
    const editorUriTemplate = getEditorUriTemplate();
    const linkDb = getDb();

    for (let i = 0; i < filtered.length; i++) {
      const row = filtered[i];
      if (!row) continue;
      const { line, snippet } = extractSnippet(row.body, query, 500, row.chunkPos, undefined, opts.intent);
      const docid = row.docid || (row.hash ? row.hash.slice(0, 6) : undefined);

      // Line 1: filepath with docid
      const virtualPath = row.file.startsWith("qmd://") ? row.file : toQmdPath(row.displayPath);
      const parsed = parseVirtualPath(virtualPath);
      const absolutePath = resolveVirtualPath(linkDb, virtualPath);

      const legacyPath = toQmdPath(row.displayPath);
      const displayPath = parsed?.path || row.displayPath;

      // Only show :line if we actually found a term match in the snippet body (exclude header line).
      const snippetBody = snippet.split("\n").slice(1).join("\n").toLowerCase();
      const hasMatch = query.toLowerCase().split(/\s+/).some(t => t.length > 0 && snippetBody.includes(t));
      const lineInfo = hasMatch ? `:${line}` : "";
      const docidStr = docid ? ` ${c.dim}#${docid}${c.reset}` : "";

      if (process.stdout.isTTY && absolutePath && parsed?.path) {
        const linkLine = hasMatch ? line : 1;
        const linkTarget = buildEditorUri(editorUriTemplate, absolutePath, linkLine, 1);
        const clickable = termLink(`${displayPath}${lineInfo}`, linkTarget);
        console.log(`${c.cyan}${clickable}${c.reset}${docidStr}`);
      } else {
        console.log(`${c.cyan}${legacyPath}${c.dim}${lineInfo}${c.reset}${docidStr}`);
      }

      // Line 2: Title (if available)
      if (row.title) {
        console.log(`${c.bold}Title: ${row.title}${c.reset}`);
      }

      // Line 3: Context (if available)
      if (row.context) {
        console.log(`${c.dim}Context: ${row.context}${c.reset}`);
      }

      // Line 4: Score
      const score = formatScore(row.score);
      console.log(`Score: ${c.bold}${score}${c.reset}`);
      if (opts.explain && row.explain) {
        const explain = row.explain;
        const ftsScores = explain.ftsScores.length > 0
          ? explain.ftsScores.map(formatExplainNumber).join(", ")
          : "none";
        const vecScores = explain.vectorScores.length > 0
          ? explain.vectorScores.map(formatExplainNumber).join(", ")
          : "none";
        const contribSummary = explain.rrf.contributions
          .slice()
          .sort((a, b) => b.rrfContribution - a.rrfContribution)
          .slice(0, 3)
          .map(c => `${c.source}/${c.queryType}#${c.rank}:${formatExplainNumber(c.rrfContribution)}`)
          .join(" | ");

        console.log(`${c.dim}Explain: fts=[${ftsScores}] vec=[${vecScores}]${c.reset}`);
        console.log(`${c.dim}  RRF: total=${formatExplainNumber(explain.rrf.totalScore)} base=${formatExplainNumber(explain.rrf.baseScore)} bonus=${formatExplainNumber(explain.rrf.topRankBonus)} rank=${explain.rrf.rank}${c.reset}`);
        console.log(`${c.dim}  Blend: ${Math.round(explain.rrf.weight * 100)}%*${formatExplainNumber(explain.rrf.positionScore)} + ${Math.round((1 - explain.rrf.weight) * 100)}%*${formatExplainNumber(explain.rerankScore)} = ${formatExplainNumber(explain.blendedScore)}${c.reset}`);
        if (contribSummary.length > 0) {
          console.log(`${c.dim}  Top RRF contributions: ${contribSummary}${c.reset}`);
        }
      }
      console.log();

      // Snippet with highlighting (diff-style header included)
      let displaySnippet = opts.lineNumbers ? addLineNumbers(snippet, line) : snippet;
      const highlighted = highlightTerms(displaySnippet, query);
      console.log(highlighted);

      // Double empty line between results
      if (i < filtered.length - 1) console.log('\n');
    }
  } else if (opts.format === "md") {
    for (let i = 0; i < filtered.length; i++) {
      const row = filtered[i];
      if (!row) continue;
      const heading = row.title || row.displayPath;
      const docid = row.docid || (row.hash ? row.hash.slice(0, 6) : undefined);
      let content = opts.full ? row.body : extractSnippet(row.body, query, 500, row.chunkPos, undefined, opts.intent).snippet;
      if (opts.lineNumbers) {
        content = addLineNumbers(content);
      }
      const docidLine = docid ? `**docid:** \`#${docid}\`\n` : "";
      const contextLine = row.context ? `**context:** ${row.context}\n` : "";
      console.log(`---\n# ${heading}\n${docidLine}${contextLine}\n${content}\n`);
    }
  } else if (opts.format === "xml") {
    for (const row of filtered) {
      const titleAttr = row.title ? ` title="${row.title.replace(/"/g, '&quot;')}"` : "";
      const contextAttr = row.context ? ` context="${row.context.replace(/"/g, '&quot;')}"` : "";
      const docid = row.docid || (row.hash ? row.hash.slice(0, 6) : "");
      let content = opts.full ? row.body : extractSnippet(row.body, query, 500, row.chunkPos, undefined, opts.intent).snippet;
      if (opts.lineNumbers) {
        content = addLineNumbers(content);
      }
      console.log(`<file docid="#${docid}" name="${toQmdPath(row.displayPath)}"${titleAttr}${contextAttr}>\n${content}\n</file>\n`);
    }
  } else {
    // CSV format
    console.log("docid,score,file,title,context,line,snippet");
    for (const row of filtered) {
      const { line, snippet } = extractSnippet(row.body, query, 500, row.chunkPos, undefined, opts.intent);
      let content = opts.full ? row.body : snippet;
      if (opts.lineNumbers) {
        content = addLineNumbers(content, line);
      }
      const docid = row.docid || (row.hash ? row.hash.slice(0, 6) : "");
      const snippetText = content || "";
      console.log(`#${docid},${row.score.toFixed(4)},${escapeCSV(toQmdPath(row.displayPath))},${escapeCSV(row.title || "")},${escapeCSV(row.context || "")},${line},${escapeCSV(snippetText)}`);
    }
  }
}

// Resolve -c collection filter: supports single string, array, or undefined.
// Returns validated collection names (exits on unknown collection).
function resolveCollectionFilter(raw: string | string[] | undefined, useDefaults: boolean = false): string[] {
  // If no filter specified and useDefaults is true, use default collections
  if (!raw && useDefaults) {
    return getDefaultCollectionNames();
  }
  if (!raw) return [];
  const names = Array.isArray(raw) ? raw : [raw];
  const validated: string[] = [];
  for (const name of names) {
    const coll = getCollectionFromYaml(name);
    if (!coll) {
      console.error(`Collection not found: ${name}`);
      closeDb();
      process.exit(1);
    }
    validated.push(name);
  }
  return validated;
}

// Post-filter results to only include files from specified collections.
function filterByCollections<T extends { filepath?: string; file?: string }>(results: T[], collectionNames: string[]): T[] {
  if (collectionNames.length <= 1) return results;
  const prefixes = collectionNames.map(n => `qmd://${n}/`);
  return results.filter(r => {
    const path = r.filepath || r.file || '';
    return prefixes.some(p => path.startsWith(p));
  });
}

/**
 * Parse structured search query syntax.
 * Lines starting with lex:, vec:, or hyde: are routed directly.
 * Plain lines without prefix go through query expansion.
 * 
 * Returns null if this is a plain query (single line, no prefix).
 * Returns ExpandedQuery[] if structured syntax detected.
 * Throws if multiple plain lines (ambiguous).
 * 
 * Examples:
 *   "CAP theorem"                    -> null (plain query, use expansion)
 *   "lex: CAP theorem"               -> [{ type: 'lex', query: 'CAP theorem' }]
 *   "lex: CAP\nvec: consistency"     -> [{ type: 'lex', ... }, { type: 'vec', ... }]
 *   "CAP\nconsistency"               -> throws (multiple plain lines)
 */
interface ParsedStructuredQuery {
  searches: ExpandedQuery[];
  intent?: string;
}

function parseStructuredQuery(query: string): ParsedStructuredQuery | null {
  const rawLines = query.split('\n').map((line, idx) => ({
    raw: line,
    trimmed: line.trim(),
    number: idx + 1,
  })).filter(line => line.trimmed.length > 0);

  if (rawLines.length === 0) return null;

  const prefixRe = /^(lex|vec|hyde):\s*/i;
  const expandRe = /^expand:\s*/i;
  const intentRe = /^intent:\s*/i;
  const typed: ExpandedQuery[] = [];
  let intent: string | undefined;

  for (const line of rawLines) {
    if (expandRe.test(line.trimmed)) {
      if (rawLines.length > 1) {
        throw new Error(`Line ${line.number} starts with expand:, but query documents cannot mix expand with typed lines. Submit a single expand query instead.`);
      }
      const text = line.trimmed.replace(expandRe, '').trim();
      if (!text) {
        throw new Error('expand: query must include text.');
      }
      return null; // treat as standalone expand query
    }

    // Parse intent: lines
    if (intentRe.test(line.trimmed)) {
      if (intent !== undefined) {
        throw new Error(`Line ${line.number}: only one intent: line is allowed per query document.`);
      }
      const text = line.trimmed.replace(intentRe, '').trim();
      if (!text) {
        throw new Error(`Line ${line.number}: intent: must include text.`);
      }
      intent = text;
      continue;
    }

    const match = line.trimmed.match(prefixRe);
    if (match) {
      const type = match[1]!.toLowerCase() as 'lex' | 'vec' | 'hyde';
      const text = line.trimmed.slice(match[0].length).trim();
      if (!text) {
        throw new Error(`Line ${line.number} (${type}:) must include text.`);
      }
      if (/\r|\n/.test(text)) {
        throw new Error(`Line ${line.number} (${type}:) contains a newline. Keep each query on a single line.`);
      }
      typed.push({ type, query: text, line: line.number });
      continue;
    }

    if (rawLines.length === 1) {
      // Single plain line -> implicit expand
      return null;
    }

    throw new Error(`Line ${line.number} is missing a lex:/vec:/hyde:/intent: prefix. Each line in a query document must start with one.`);
  }

  // intent: alone is not a valid query — must have at least one search
  if (intent && typed.length === 0) {
    throw new Error('intent: cannot appear alone. Add at least one lex:, vec:, or hyde: line.');
  }

  return typed.length > 0 ? { searches: typed, intent } : null;
}

function search(query: string, opts: OutputOptions): void {
  const db = getDb();

  // Validate collection filter (supports multiple -c flags)
  // Use default collections if none specified
  const collectionNames = resolveCollectionFilter(opts.collection, true);
  const singleCollection = collectionNames.length === 1 ? collectionNames[0] : undefined;

  // Use large limit for --all, otherwise fetch more than needed and let outputResults filter
  const fetchLimit = opts.all ? 100000 : Math.max(50, opts.limit * 2);
  const results = filterByCollections(
    searchFTS(db, query, fetchLimit, singleCollection),
    collectionNames
  );

  // Add context to results
  const resultsWithContext = results.map(r => ({
    file: r.filepath,
    displayPath: r.displayPath,
    title: r.title,
    body: r.body || "",
    score: r.score,
    context: getContextForFile(db, r.filepath),
    hash: r.hash,
    docid: r.docid,
  }));

  closeDb();

  if (resultsWithContext.length === 0) {
    printEmptySearchResults(opts.format);
    return;
  }
  outputResults(resultsWithContext, query, opts);
}

// Log query expansion as a tree to stderr (CLI progress feedback)
function logExpansionTree(originalQuery: string, expanded: ExpandedQuery[]): void {
  const lines: string[] = [];
  lines.push(`${c.dim}├─ ${originalQuery}${c.reset}`);
  for (const q of expanded) {
    let preview = q.query.replace(/\n/g, ' ');
    if (preview.length > 72) preview = preview.substring(0, 69) + '...';
    lines.push(`${c.dim}├─ ${q.type}: ${preview}${c.reset}`);
  }
  if (lines.length > 0) {
    lines[lines.length - 1] = lines[lines.length - 1]!.replace('├─', '└─');
  }
  for (const line of lines) process.stderr.write(line + '\n');
}

async function vectorSearch(query: string, opts: OutputOptions, _model: string = DEFAULT_EMBED_MODEL): Promise<void> {
  const store = getStore();

  // Validate collection filter (supports multiple -c flags)
  // Use default collections if none specified
  const collectionNames = resolveCollectionFilter(opts.collection, true);
  const singleCollection = collectionNames.length === 1 ? collectionNames[0] : undefined;

  checkIndexHealth(store.db);

  await withLLMSession(async () => {
    let results = await vectorSearchQuery(store, query, {
      collection: singleCollection,
      limit: opts.all ? 500 : (opts.limit || 10),
      minScore: opts.minScore || 0.3,
      intent: opts.intent,
      hooks: {
        onExpand: (original, expanded) => {
          logExpansionTree(original, expanded);
          process.stderr.write(`${c.dim}Searching ${expanded.length + 1} vector queries...${c.reset}\n`);
        },
      },
    });

    // Post-filter for multi-collection
    if (collectionNames.length > 1) {
      results = results.filter(r => {
        const prefixes = collectionNames.map(n => `qmd://${n}/`);
        return prefixes.some(p => r.file.startsWith(p));
      });
    }

    closeDb();

    if (results.length === 0) {
      printEmptySearchResults(opts.format);
      return;
    }

    outputResults(results.map(r => ({
      file: r.file,
      displayPath: r.displayPath,
      title: r.title,
      body: r.body,
      score: r.score,
      context: r.context,
      docid: r.docid,
    })), query, { ...opts, limit: results.length });
  }, { maxDuration: 10 * 60 * 1000, name: 'vectorSearch' });
}

async function querySearch(query: string, opts: OutputOptions, _embedModel: string = DEFAULT_EMBED_MODEL, _rerankModel: string = DEFAULT_RERANK_MODEL): Promise<void> {
  const store = getStore();

  // Validate collection filter (supports multiple -c flags)
  // Use default collections if none specified
  const collectionNames = resolveCollectionFilter(opts.collection, true);
  const singleCollection = collectionNames.length === 1 ? collectionNames[0] : undefined;

  checkIndexHealth(store.db);

  // Check for structured query syntax (lex:/vec:/hyde:/intent: prefixes)
  const parsed = parseStructuredQuery(query);
  // Intent can come from --intent flag or from intent: line in query document
  const intent = opts.intent || parsed?.intent;

  await withLLMSession(async () => {
    let results;

    if (parsed) {
      const structuredQueries = parsed.searches;
      // Structured search — user provided their own query expansions
      const typeLabels = structuredQueries.map(s => s.type).join('+');
      process.stderr.write(`${c.dim}Structured search: ${structuredQueries.length} queries (${typeLabels})${c.reset}\n`);
      if (intent) {
        process.stderr.write(`${c.dim}├─ intent: ${intent}${c.reset}\n`);
      }

      // Log each sub-query
      for (const s of structuredQueries) {
        let preview = s.query.replace(/\n/g, ' ');
        if (preview.length > 72) preview = preview.substring(0, 69) + '...';
        process.stderr.write(`${c.dim}├─ ${s.type}: ${preview}${c.reset}\n`);
      }
      process.stderr.write(`${c.dim}└─ Searching...${c.reset}\n`);

      results = await structuredSearch(store, structuredQueries, {
        collections: singleCollection ? [singleCollection] : undefined,
        limit: opts.all ? 500 : (opts.limit || 10),
        minScore: opts.minScore || 0,
        candidateLimit: opts.candidateLimit,
        skipRerank: opts.skipRerank,
        explain: !!opts.explain,
        intent,
        chunkStrategy: opts.chunkStrategy,
        hooks: {
          onEmbedStart: (count) => {
            process.stderr.write(`${c.dim}Embedding ${count} ${count === 1 ? 'query' : 'queries'}...${c.reset}`);
          },
          onEmbedDone: (ms) => {
            process.stderr.write(`${c.dim} (${formatMs(ms)})${c.reset}\n`);
          },
          onRerankStart: (chunkCount) => {
            process.stderr.write(`${c.dim}Reranking ${chunkCount} chunks...${c.reset}`);
            progress.indeterminate();
          },
          onRerankDone: (ms) => {
            progress.clear();
            process.stderr.write(`${c.dim} (${formatMs(ms)})${c.reset}\n`);
          },
        },
      });
    } else {
      // Standard hybrid query with automatic expansion
      results = await hybridQuery(store, query, {
        collection: singleCollection,
        limit: opts.all ? 500 : (opts.limit || 10),
        minScore: opts.minScore || 0,
        candidateLimit: opts.candidateLimit,
        skipRerank: opts.skipRerank,
        explain: !!opts.explain,
        intent,
        chunkStrategy: opts.chunkStrategy,
        hooks: {
          onStrongSignal: (score) => {
            process.stderr.write(`${c.dim}Strong BM25 signal (${score.toFixed(2)}) — skipping expansion${c.reset}\n`);
          },
          onExpandStart: () => {
            process.stderr.write(`${c.dim}Expanding query...${c.reset}`);
          },
          onExpand: (original, expanded, ms) => {
            process.stderr.write(`${c.dim} (${formatMs(ms)})${c.reset}\n`);
            logExpansionTree(original, expanded);
            process.stderr.write(`${c.dim}Searching ${expanded.length + 1} queries...${c.reset}\n`);
          },
          onEmbedStart: (count) => {
            process.stderr.write(`${c.dim}Embedding ${count} ${count === 1 ? 'query' : 'queries'}...${c.reset}`);
          },
          onEmbedDone: (ms) => {
            process.stderr.write(`${c.dim} (${formatMs(ms)})${c.reset}\n`);
          },
          onRerankStart: (chunkCount) => {
            process.stderr.write(`${c.dim}Reranking ${chunkCount} chunks...${c.reset}`);
            progress.indeterminate();
          },
          onRerankDone: (ms) => {
            progress.clear();
            process.stderr.write(`${c.dim} (${formatMs(ms)})${c.reset}\n`);
          },
        },
      });
    }

    // Post-filter for multi-collection
    if (collectionNames.length > 1) {
      results = results.filter(r => {
        const prefixes = collectionNames.map(n => `qmd://${n}/`);
        return prefixes.some(p => r.file.startsWith(p));
      });
    }

    closeDb();

    if (results.length === 0) {
      printEmptySearchResults(opts.format);
      return;
    }

    // Use first lex/vec query for output context, or original query
    const structuredQueries = parsed?.searches;
    const displayQuery = structuredQueries
      ? (structuredQueries.find(s => s.type === 'lex')?.query || structuredQueries.find(s => s.type === 'vec')?.query || query)
      : query;

    // Map to CLI output format — use bestChunk for snippet display
    outputResults(results.map(r => ({
      file: r.file,
      displayPath: r.displayPath,
      title: r.title,
      body: r.bestChunk,
      chunkPos: r.bestChunkPos,
      score: r.score,
      context: r.context,
      docid: r.docid,
      explain: r.explain,
    })), displayQuery, { ...opts, limit: results.length });
  }, { maxDuration: 10 * 60 * 1000, name: 'querySearch' });
}

// Parse CLI arguments using util.parseArgs
function parseCLI() {
  const { values, positionals } = parseArgs({
    args: process.argv.slice(2), // Skip node and script path
    options: {
      // Global options
      index: {
        type: "string",
      },
      context: {
        type: "string",
      },
      help: { type: "boolean", short: "h" },
      version: { type: "boolean", short: "v" },
      skill: { type: "boolean" },
      global: { type: "boolean" },
      yes: { type: "boolean" },
      // Search options
      n: { type: "string" },
      "min-score": { type: "string" },
      all: { type: "boolean" },
      full: { type: "boolean" },
      csv: { type: "boolean" },
      md: { type: "boolean" },
      xml: { type: "boolean" },
      files: { type: "boolean" },
      json: { type: "boolean" },
      explain: { type: "boolean" },
      collection: { type: "string", short: "c", multiple: true },  // Filter by collection(s)
      // Collection options
      name: { type: "string" },  // collection name
      mask: { type: "string" },  // glob pattern
      // Embed options
      force: { type: "boolean", short: "f" },
      "max-docs-per-batch": { type: "string" },
      "max-batch-mb": { type: "string" },
      // Update options
      pull: { type: "boolean" },  // git pull before update
      refresh: { type: "boolean" },
      // Get options
      l: { type: "string" },  // max lines
      from: { type: "string" },  // start line
      "max-bytes": { type: "string" },  // max bytes for multi-get
      "line-numbers": { type: "boolean" },  // add line numbers to output
      // Query options
      "candidate-limit": { type: "string", short: "C" },
      "no-rerank": { type: "boolean", default: false },
      intent: { type: "string" },
      // Chunking options
      "chunk-strategy": { type: "string" },  // "regex" (default) or "auto" (AST for code files)
      // MCP HTTP transport options
      http: { type: "boolean" },
      daemon: { type: "boolean" },
      port: { type: "string" },
      // bench jina options
      size: { type: "string" },
      "doc-len": { type: "string" },
      provider: { type: "string" },
      "skip-rerank": { type: "boolean" },
      "skip-single": { type: "boolean" },
      runs: { type: "string" },
      // usage chart options
      days: { type: "string" },
    },
    allowPositionals: true,
    strict: false, // Allow unknown options to pass through
  });

  // Select index name (default: "index")
  const indexName = values.index as string | undefined;
  if (indexName) {
    setIndexName(indexName);
    setConfigIndexName(indexName);
  }

  // Determine output format
  let format: OutputFormat = "cli";
  if (values.csv) format = "csv";
  else if (values.md) format = "md";
  else if (values.xml) format = "xml";
  else if (values.files) format = "files";
  else if (values.json) format = "json";

  // Default limit: 20 for --files/--json, 5 otherwise
  // --all means return all results (use very large limit)
  const defaultLimit = (format === "files" || format === "json") ? 20 : 5;
  const isAll = !!values.all;

  const opts: OutputOptions = {
    format,
    full: !!values.full,
    limit: isAll ? 100000 : (values.n ? parseInt(String(values.n), 10) || defaultLimit : defaultLimit),
    minScore: values["min-score"] ? parseFloat(String(values["min-score"])) || 0 : 0,
    all: isAll,
    collection: values.collection as string[] | undefined,
    lineNumbers: !!values["line-numbers"],
    candidateLimit: values["candidate-limit"] ? parseInt(String(values["candidate-limit"]), 10) : undefined,
    skipRerank: !!values["no-rerank"],
    explain: !!values.explain,
    intent: values.intent as string | undefined,
    chunkStrategy: parseChunkStrategy(values["chunk-strategy"]),
  };

  return {
    command: positionals[0] || "",
    args: positionals.slice(1),
    query: positionals.slice(1).join(" "),
    opts,
    values,
  };
}

function getSkillInstallDir(globalInstall: boolean): string {
  return globalInstall
    ? resolve(homedir(), ".agents", "skills", "qmd")
    : resolve(getPwd(), ".agents", "skills", "qmd");
}

function getClaudeSkillLinkPath(globalInstall: boolean): string {
  return globalInstall
    ? resolve(homedir(), ".claude", "skills", "qmd")
    : resolve(getPwd(), ".claude", "skills", "qmd");
}

function pathExists(path: string): boolean {
  try {
    lstatSync(path);
    return true;
  } catch {
    return false;
  }
}

function removePath(path: string): void {
  const stat = lstatSync(path);
  if (stat.isDirectory() && !stat.isSymbolicLink()) {
    rmSync(path, { recursive: true, force: true });
  } else {
    unlinkSync(path);
  }
}

function showSkill(): void {
  console.log("QMD Skill (embedded)");
  console.log("");
  const content = getEmbeddedQmdSkillContent();
  process.stdout.write(content.endsWith("\n") ? content : content + "\n");
}

function writeEmbeddedSkill(targetDir: string, force: boolean): void {
  if (pathExists(targetDir)) {
    if (!force) {
      throw new Error(`Skill already exists: ${targetDir} (use --force to replace it)`);
    }
    removePath(targetDir);
  }

  mkdirSync(targetDir, { recursive: true });
  for (const file of getEmbeddedQmdSkillFiles()) {
    const destination = resolve(targetDir, file.relativePath);
    mkdirSync(dirname(destination), { recursive: true });
    writeFileSync(destination, file.content, "utf-8");
  }
}

function ensureClaudeSymlink(linkPath: string, targetDir: string, force: boolean): boolean {
  const parentDir = dirname(linkPath);
  if (pathExists(parentDir)) {
    const resolvedTargetDir = realpathSync(dirname(targetDir));
    const resolvedLinkParent = realpathSync(parentDir);

    // If .claude/skills already resolves to the same directory as .agents/skills,
    // the skill is already visible to Claude and creating qmd -> qmd would loop.
    if (resolvedTargetDir === resolvedLinkParent) {
      return false;
    }
  }

  const linkTarget = relativePath(parentDir, targetDir) || ".";

  mkdirSync(parentDir, { recursive: true });

  if (pathExists(linkPath)) {
    const stat = lstatSync(linkPath);
    if (stat.isSymbolicLink() && readlinkSync(linkPath) === linkTarget) {
      return true;
    }
    if (!force) {
      throw new Error(`Claude skill path already exists: ${linkPath} (use --force to replace it)`);
    }
    removePath(linkPath);
  }

  symlinkSync(linkTarget, linkPath, "dir");
  return true;
}

async function shouldCreateClaudeSymlink(linkPath: string, autoYes: boolean): Promise<boolean> {
  if (autoYes) {
    return true;
  }
  if (!process.stdin.isTTY || !process.stdout.isTTY) {
    console.log(`Tip: create a Claude symlink manually at ${linkPath}`);
    return false;
  }

  const rl = createInterface({
    input: process.stdin,
    output: process.stdout,
  });

  try {
    const answer = await rl.question(`Create a symlink in ${linkPath}? [y/N] `);
    const normalized = answer.trim().toLowerCase();
    return normalized === "y" || normalized === "yes";
  } finally {
    rl.close();
  }
}

async function installSkill(globalInstall: boolean, force: boolean, autoYes: boolean): Promise<void> {
  const installDir = getSkillInstallDir(globalInstall);
  writeEmbeddedSkill(installDir, force);
  console.log(`✓ Installed QMD skill to ${installDir}`);

  const claudeLinkPath = getClaudeSkillLinkPath(globalInstall);
  if (!(await shouldCreateClaudeSymlink(claudeLinkPath, autoYes))) {
    return;
  }

  const linked = ensureClaudeSymlink(claudeLinkPath, installDir, force);
  if (linked) {
    console.log(`✓ Linked Claude skill at ${claudeLinkPath}`);
  } else {
    console.log(`✓ Claude already sees the skill via ${dirname(claudeLinkPath)}`);
  }
}

function showHelp(): void {
  console.log("qmd — Quick Markdown Search");
  console.log("");
  console.log("Usage:");
  console.log("  qmd <command> [options]");
  console.log("");
  console.log("Primary commands:");
  console.log("  qmd query <query>             - Hybrid search with auto expansion + reranking (recommended)");
  console.log("  qmd query 'lex:..\\nvec:...'   - Structured query document (you provide lex/vec/hyde lines)");
  console.log("  qmd search <query>            - Full-text BM25 keywords (no LLM)");
  console.log("  qmd vsearch <query>           - Vector similarity only");
  console.log("  qmd get <file>[:line] [-l N]  - Show a single document, optional line slice");
  console.log("  qmd multi-get <pattern>       - Batch fetch via glob or comma-separated list");
  console.log("  qmd skill show/install        - Show or install the packaged QMD skill");
  console.log("  qmd mcp                       - Start the MCP server (stdio transport for AI agents)");
  console.log("  qmd bench <fixture.json>      - Run search quality benchmarks against a fixture file");
  console.log("");
  console.log("Collections & context:");
  console.log("  qmd collection add/list/remove/rename/show   - Manage indexed folders");
  console.log("  qmd context add/list/rm                      - Attach human-written summaries");
  console.log("  qmd ls [collection[/path]]                   - Inspect indexed files");
  console.log("");
  console.log("Maintenance:");
  console.log("  qmd status                    - View index + collection health");
  console.log("  qmd update [--pull]           - Re-index collections (optionally git pull first)");
  console.log("  qmd embed [-f]                - Generate/refresh vector embeddings");
  console.log("    --max-docs-per-batch <n>    - Cap docs loaded into memory per embedding batch");
  console.log("    --max-batch-mb <n>          - Cap UTF-8 MB loaded into memory per embedding batch");
  console.log("  qmd usage [--json|--csv]      - Show Jina API token usage");
  console.log("  qmd usage chart [--days N]    - ASCII histogram of daily Jina usage");
  console.log("  qmd usage reset               - Clear Jina usage history");
  console.log("  qmd bench jina [options]      - Benchmark local vs Jina embed/rerank latency");
  console.log("  qmd cleanup                   - Clear caches, vacuum DB");
  console.log("");
  console.log("Query syntax (qmd query):");
  console.log("  QMD queries are either a single expand query (no prefix) or a multi-line");
  console.log("  document where every line is typed with lex:, vec:, or hyde:. This grammar");
  console.log("  matches the docs in docs/SYNTAX.md and is enforced in the CLI.");
  console.log("");
  const grammar = [
    `query          = expand_query | query_document ;`,
    `expand_query   = text | explicit_expand ;`,
    `explicit_expand= "expand:" text ;`,
    `query_document = [ intent_line ] { typed_line } ;`,
    `intent_line    = "intent:" text newline ;`,
    `typed_line     = type ":" text newline ;`,
    `type           = "lex" | "vec" | "hyde" ;`,
    `text           = quoted_phrase | plain_text ;`,
    `quoted_phrase  = '"' { character } '"' ;`,
    `plain_text     = { character } ;`,
    `newline        = "\\n" ;`,
  ];
  console.log("  Grammar:");
  for (const line of grammar) {
    console.log(`    ${line}`);
  }
  console.log("");
  console.log("  Examples:");
  console.log("    qmd query \"how does auth work\"                # single-line → implicit expand");
  console.log("    qmd query $'lex: CAP theorem\\nvec: consistency'  # typed query document");
  console.log("    qmd query $'lex: \"exact matches\" sports -baseball'  # phrase + negation lex search");
  console.log("    qmd query $'hyde: Hypothetical answer text'       # hyde-only document");
  console.log("");
  console.log("  Constraints:");
  console.log("    - Standalone expand queries cannot mix with typed lines.");
  console.log("    - Query documents allow only lex:, vec:, or hyde: prefixes.");
  console.log("    - Each typed line must be single-line text with balanced quotes.");
  console.log("");
  console.log("AI agents & integrations:");
  console.log("  - Run `qmd mcp` to expose the MCP server (stdio) to agents/IDEs.");
  console.log("  - `qmd skill install` installs the QMD skill into ./.agents/skills/qmd.");
  console.log("  - Use `qmd skill install --global` for ~/.agents/skills/qmd.");
  console.log("  - `qmd --skill` is kept as an alias for `qmd skill show`.");
  console.log("  - Advanced: `qmd mcp --http ...` and `qmd mcp --http --daemon` are optional for custom transports.");
  console.log("");
  console.log("Global options:");
  console.log("  --index <name>             - Use a named index (default: index)");
  console.log("  QMD_EDITOR_URI             - Editor link template for clickable TTY search output");
  console.log("");
  console.log("Search options:");
  console.log("  -n <num>                   - Max results (default 5, or 20 for --files/--json)");
  console.log("  --all                      - Return all matches (pair with --min-score)");
  console.log("  --min-score <num>          - Minimum similarity score");
  console.log("  --full                     - Output full document instead of snippet");
  console.log("  -C, --candidate-limit <n>  - Max candidates to rerank (default 40, lower = faster)");
  console.log("  --no-rerank                - Skip LLM reranking (use RRF scores only, much faster on CPU)");
  console.log("  --line-numbers             - Include line numbers in output");
  console.log("  --explain                  - Include retrieval score traces (query --json/CLI)");
  console.log("  --files | --json | --csv | --md | --xml  - Output format");
  console.log("  -c, --collection <name>    - Filter by one or more collections");
  console.log("");
  console.log("Embed/query options:");
  console.log("  --chunk-strategy <auto|regex> - Chunking mode (default: regex; auto uses AST for code files)");
  console.log("");
  console.log("Multi-get options:");
  console.log("  -l <num>                   - Maximum lines per file");
  console.log("  --max-bytes <num>          - Skip files larger than N bytes (default 10240)");
  console.log("  --json/--csv/--md/--xml/--files - Same formats as search");
  console.log("");
  console.log(`Index: ${getDbPath()}`);
}

async function showVersion(): Promise<void> {
  const scriptDir = dirname(fileURLToPath(import.meta.url));
  const pkgPath = resolve(scriptDir, "..", "..", "package.json");
  const pkg = JSON.parse(readFileSync(pkgPath, "utf-8"));

  let commit = "";
  try {
    commit = execSync(`git -C ${scriptDir} rev-parse --short HEAD`, { encoding: "utf-8", stdio: ["pipe", "pipe", "pipe"] }).trim();
  } catch {
    // Not a git repo or git not available
  }

  const versionStr = commit ? `${pkg.version} (${commit})` : pkg.version;
  console.log(`qmd ${versionStr}`);
}

// Main CLI - only run if this is the main module
const __filename = fileURLToPath(import.meta.url);
const argv1 = process.argv[1];
const isMain = argv1 === __filename
  || argv1?.endsWith("/qmd.ts")
  || argv1?.endsWith("/qmd.js")
  || (argv1 != null && realpathSync(argv1) === __filename);
if (isMain) {
  const cli = parseCLI();

  if (cli.values.version) {
    await showVersion();
    process.exit(0);
  }

  if (cli.values.skill) {
    showSkill();
    process.exit(0);
  }

  if (cli.values.help && cli.command === "skill") {
    console.log("Usage: qmd skill <show|install> [options]");
    console.log("");
    console.log("Commands:");
    console.log("  show                 Print the packaged QMD skill");
    console.log("  install              Install into ./.agents/skills/qmd");
    console.log("");
    console.log("Options:");
    console.log("  --global             Install into ~/.agents/skills/qmd");
    console.log("  --yes                Also create the .claude/skills/qmd symlink");
    console.log("  -f, --force          Replace existing install or symlink");
    process.exit(0);
  }

  if (!cli.command || cli.values.help) {
    showHelp();
    process.exit(cli.values.help ? 0 : 1);
  }

  switch (cli.command) {
    case "context": {
      const subcommand = cli.args[0];
      if (!subcommand) {
        console.error("Usage: qmd context <add|list|rm>");
        console.error("");
        console.error("Commands:");
        console.error("  qmd context add [path] \"text\"  - Add context (defaults to current dir)");
        console.error("  qmd context add / \"text\"       - Add global context to all collections");
        console.error("  qmd context list                - List all contexts");
        console.error("  qmd context rm <path>           - Remove context");
        process.exit(1);
      }

      switch (subcommand) {
        case "add": {
          if (cli.args.length < 2) {
            console.error("Usage: qmd context add [path] \"text\"");
            console.error("");
            console.error("Examples:");
            console.error("  qmd context add \"Context for current directory\"");
            console.error("  qmd context add . \"Context for current directory\"");
            console.error("  qmd context add /subfolder \"Context for subfolder\"");
            console.error("  qmd context add / \"Global context for all collections\"");
            console.error("");
            console.error("  Using virtual paths:");
            console.error("  qmd context add qmd://journals/ \"Context for entire journals collection\"");
            console.error("  qmd context add qmd://journals/2024 \"Context for 2024 journals\"");
            process.exit(1);
          }

          let pathArg: string | undefined;
          let contextText: string;

          // Check if first arg looks like a path or if it's the context text
          const firstArg = cli.args[1] || '';
          const secondArg = cli.args[2];

          if (secondArg) {
            // Two args: path + context
            pathArg = firstArg;
            contextText = cli.args.slice(2).join(" ");
          } else {
            // One arg: context only (use current directory)
            pathArg = undefined;
            contextText = firstArg;
          }

          await contextAdd(pathArg, contextText);
          break;
        }

        case "list": {
          contextList();
          break;
        }

        case "rm":
        case "remove": {
          if (cli.args.length < 2 || !cli.args[1]) {
            console.error("Usage: qmd context rm <path>");
            console.error("Examples:");
            console.error("  qmd context rm /");
            console.error("  qmd context rm qmd://journals/2024");
            process.exit(1);
          }
          contextRemove(cli.args[1]);
          break;
        }

        default:
          console.error(`Unknown subcommand: ${subcommand}`);
          console.error("Available: add, list, rm");
          process.exit(1);
      }
      break;
    }

    case "get": {
      if (!cli.args[0]) {
        console.error("Usage: qmd get <filepath>[:line] [--from <line>] [-l <lines>] [--line-numbers]");
        process.exit(1);
      }
      const fromLine = cli.values.from ? parseInt(cli.values.from as string, 10) : undefined;
      const maxLines = cli.values.l ? parseInt(cli.values.l as string, 10) : undefined;
      getDocument(cli.args[0], fromLine, maxLines, cli.opts.lineNumbers);
      break;
    }

    case "multi-get": {
      if (!cli.args[0]) {
        console.error("Usage: qmd multi-get <pattern> [-l <lines>] [--max-bytes <bytes>] [--json|--csv|--md|--xml|--files]");
        console.error("  pattern: glob (e.g., 'journals/2025-05*.md') or comma-separated list");
        process.exit(1);
      }
      const maxLinesMulti = cli.values.l ? parseInt(cli.values.l as string, 10) : undefined;
      const maxBytes = cli.values["max-bytes"] ? parseInt(cli.values["max-bytes"] as string, 10) : DEFAULT_MULTI_GET_MAX_BYTES;
      multiGet(cli.args[0], maxLinesMulti, maxBytes, cli.opts.format);
      break;
    }

    case "ls": {
      listFiles(cli.args[0]);
      break;
    }

    case "collection": {
      const subcommand = cli.args[0];
      switch (subcommand) {
        case "list": {
          collectionList();
          break;
        }

        case "add": {
          const pwd = cli.args[1] || getPwd();
          const resolvedPwd = pwd === '.' ? getPwd() : getRealPath(resolve(pwd));
          const globPattern = cli.values.mask as string || DEFAULT_GLOB;
          const name = cli.values.name as string | undefined;

          await collectionAdd(resolvedPwd, globPattern, name);
          break;
        }

        case "remove":
        case "rm": {
          if (!cli.args[1]) {
            console.error("Usage: qmd collection remove <name>");
            console.error("  Use 'qmd collection list' to see available collections");
            process.exit(1);
          }
          collectionRemove(cli.args[1]);
          break;
        }

        case "rename":
        case "mv": {
          if (!cli.args[1] || !cli.args[2]) {
            console.error("Usage: qmd collection rename <old-name> <new-name>");
            console.error("  Use 'qmd collection list' to see available collections");
            process.exit(1);
          }
          collectionRename(cli.args[1], cli.args[2]);
          break;
        }

        case "set-update":
        case "update-cmd": {
          const name = cli.args[1];
          const cmd = cli.args.slice(2).join(' ') || null;
          if (!name) {
            console.error("Usage: qmd collection update-cmd <name> [command]");
            console.error("  Set the command to run before indexing (e.g., 'git pull')");
            console.error("  Omit command to clear it");
            process.exit(1);
          }
          const { updateCollectionSettings, getCollection } = await import("../collections.js");
          const col = getCollection(name);
          if (!col) {
            console.error(`Collection not found: ${name}`);
            process.exit(1);
          }
          updateCollectionSettings(name, { update: cmd });
          if (cmd) {
            console.log(`✓ Set update command for '${name}': ${cmd}`);
          } else {
            console.log(`✓ Cleared update command for '${name}'`);
          }
          break;
        }

        case "include":
        case "exclude": {
          const name = cli.args[1];
          if (!name) {
            console.error(`Usage: qmd collection ${subcommand} <name>`);
            console.error(`  ${subcommand === 'include' ? 'Include' : 'Exclude'} collection in default queries`);
            process.exit(1);
          }
          const { updateCollectionSettings, getCollection } = await import("../collections.js");
          const col = getCollection(name);
          if (!col) {
            console.error(`Collection not found: ${name}`);
            process.exit(1);
          }
          const include = subcommand === 'include';
          updateCollectionSettings(name, { includeByDefault: include });
          console.log(`✓ Collection '${name}' ${include ? 'included in' : 'excluded from'} default queries`);
          break;
        }

        case "show":
        case "info": {
          const name = cli.args[1];
          if (!name) {
            console.error("Usage: qmd collection show <name>");
            process.exit(1);
          }
          const { getCollection } = await import("../collections.js");
          const col = getCollection(name);
          if (!col) {
            console.error(`Collection not found: ${name}`);
            process.exit(1);
          }
          console.log(`Collection: ${name}`);
          console.log(`  Path:     ${col.path}`);
          console.log(`  Pattern:  ${col.pattern}`);
          console.log(`  Include:  ${col.includeByDefault !== false ? 'yes (default)' : 'no'}`);
          if (col.update) {
            console.log(`  Update:   ${col.update}`);
          }
          if (col.context) {
            const ctxCount = Object.keys(col.context).length;
            console.log(`  Contexts: ${ctxCount}`);
          }
          break;
        }

        case "help":
        case undefined: {
          console.log("Usage: qmd collection <command> [options]");
          console.log("");
          console.log("Commands:");
          console.log("  list                      List all collections");
          console.log("  add <path> [--name NAME]  Add a collection");
          console.log("  remove <name>             Remove a collection");
          console.log("  rename <old> <new>        Rename a collection");
          console.log("  show <name>               Show collection details");
          console.log("  update-cmd <name> [cmd]   Set pre-update command (e.g., 'git pull')");
          console.log("  include <name>            Include in default queries");
          console.log("  exclude <name>            Exclude from default queries");
          console.log("");
          console.log("Examples:");
          console.log("  qmd collection add ~/notes --name notes");
          console.log("  qmd collection update-cmd brain 'git pull'");
          console.log("  qmd collection exclude archive");
          process.exit(0);
        }

        default:
          console.error(`Unknown subcommand: ${subcommand}`);
          console.error("Run 'qmd collection help' for usage");
          process.exit(1);
      }
      break;
    }

    case "status":
      await showStatus();
      break;

    case "update":
      await updateCollections();
      break;

    case "embed":
      try {
        const maxDocsPerBatch = parseEmbedBatchOption("maxDocsPerBatch", cli.values["max-docs-per-batch"]);
        const maxBatchMb = parseEmbedBatchOption("maxBatchBytes", cli.values["max-batch-mb"]);
        const embedChunkStrategy = parseChunkStrategy(cli.values["chunk-strategy"]);
        // Use the active embed backend's model URI so batch output labels are accurate.
        // This resolves to `jina:<model>` when QMD_EMBED_PROVIDER=jina.
        const activeEmbedModel = getDefaultLlamaCpp().embedModelName;
        await vectorIndex(activeEmbedModel, !!cli.values.force, {
          maxDocsPerBatch,
          maxBatchBytes: maxBatchMb === undefined ? undefined : maxBatchMb * 1024 * 1024,
          chunkStrategy: embedChunkStrategy,
        });
      } catch (error) {
        console.error(error instanceof Error ? error.message : String(error));
        process.exit(1);
      }
      break;

    case "pull": {
      const refresh = cli.values.refresh === undefined ? false : Boolean(cli.values.refresh);
      // Skip models served by a remote provider — there's no local GGUF file
      // to fetch.
      const llmInstance = getDefaultLlamaCpp();
      const usingRemoteEmbed = llmInstance.usingRemoteEmbedder;
      const usingRemoteRerank = llmInstance.usingRemoteReranker;
      const models = [
        ...(usingRemoteEmbed ? [] : [DEFAULT_EMBED_MODEL_URI]),
        DEFAULT_GENERATE_MODEL_URI,
        ...(usingRemoteRerank ? [] : [DEFAULT_RERANK_MODEL_URI]),
      ];
      if (usingRemoteEmbed) {
        console.log(`${c.dim}Skipping embedding model (using remote provider: ${llmInstance.embedProviderName})${c.reset}`);
      }
      if (usingRemoteRerank) {
        console.log(`${c.dim}Skipping rerank model (using remote provider: ${llmInstance.rerankProviderName})${c.reset}`);
      }
      console.log(`${c.bold}Pulling models${c.reset}`);
      const results = await pullModels(models, {
        refresh,
        cacheDir: DEFAULT_MODEL_CACHE_DIR,
      });
      for (const result of results) {
        const size = formatBytes(result.sizeBytes);
        const note = result.refreshed ? "refreshed" : "cached/checked";
        console.log(`- ${result.model} -> ${result.path} (${size}, ${note})`);
      }
      break;
    }

    case "usage": {
      // Subcommands:
      //   qmd usage                 → summary (text / --json / --csv)
      //   qmd usage reset           → clear history
      //   qmd usage chart [--days]  → ASCII histogram of daily totals
      const sub = cli.args[0];

      if (sub === "reset") {
        const removed = clearJinaUsage(getDb());
        if (cli.values.json) {
          process.stdout.write(
            JSON.stringify({ schema: "qmd.usage.v1", action: "reset", cleared: removed }, null, 2) +
              "\n",
          );
        } else {
          console.log(`${c.dim}Cleared ${removed} jina_usage entries.${c.reset}`);
        }
        break;
      }

      if (sub === "chart") {
        const daysRaw = cli.values["days"];
        const days = typeof daysRaw === "string" ? parseInt(daysRaw, 10) : undefined;
        await showUsageChart({
          days: Number.isFinite(days) && (days ?? 0) > 0 ? days : undefined,
          json: !!cli.values.json,
        });
        break;
      }

      // Default: summary. Honour --csv and --json (--json takes precedence).
      if (cli.values.csv) {
        const report = collectUsageReport();
        process.stdout.write(renderUsageCsv(report));
        break;
      }
      await showUsage({ json: !!cli.values.json });
      break;
    }

    case "search":
      if (!cli.query) {
        console.error("Usage: qmd search [options] <query>");
        process.exit(1);
      }
      search(cli.query, cli.opts);
      break;

    case "vsearch":
    case "vector-search": // undocumented alias
      if (!cli.query) {
        console.error("Usage: qmd vsearch [options] <query>");
        process.exit(1);
      }
      // Default min-score for vector search is 0.3
      if (!cli.values["min-score"]) {
        cli.opts.minScore = 0.3;
      }
      await vectorSearch(cli.query, cli.opts);
      break;

    case "query":
    case "deep-search": // undocumented alias
      if (!cli.query) {
        console.error("Usage: qmd query [options] <query>");
        process.exit(1);
      }
      await querySearch(cli.query, cli.opts);
      break;

    case "bench": {
      // Subcommand dispatch:
      //   qmd bench jina [...]         → Jina latency/throughput benchmark
      //   qmd bench <fixture.json>     → search quality benchmark (legacy)
      const firstArg = cli.args[0];

      if (firstArg === "jina") {
        const { runJinaBench } = await import("../bench/bench-jina.js");
        const sizeRaw = cli.values["size"];
        const docLenRaw = cli.values["doc-len"];
        const providerRaw = cli.values["provider"];
        const runsRaw = cli.values["runs"];
        const size = typeof sizeRaw === "string" ? Math.max(1, parseInt(sizeRaw, 10) || 50) : 50;
        const docLen =
          typeof docLenRaw === "string" ? Math.max(10, parseInt(docLenRaw, 10) || 500) : 500;
        const runs =
          typeof runsRaw === "string" ? Math.max(1, Math.min(100, parseInt(runsRaw, 10) || 1)) : 1;
        let providerChoice: "local" | "jina" | "both" = "both";
        if (providerRaw === "local" || providerRaw === "jina" || providerRaw === "both") {
          providerChoice = providerRaw;
        }

        const isJson = !!cli.values.json;
        if (!isJson) {
          console.log(`${c.bold}QMD Jina Benchmark${c.reset}`);
          console.log(
            `${c.dim}docs=${size}, doc_len=${docLen}, provider=${providerChoice}, runs=${runs}${c.reset}`,
          );
          if (providerChoice !== "local") {
            console.log(
              `${c.dim}⚠  This run calls the Jina API and consumes tokens from your quota.${c.reset}`,
            );
          }
          console.log();
        }

        try {
          const report = await runJinaBench({
            size,
            docLen,
            runs,
            provider: providerChoice,
            skipRerank: !!cli.values["skip-rerank"],
            skipSingle: !!cli.values["skip-single"],
            json: isJson,
          });

          if (isJson) {
            process.stdout.write(JSON.stringify(report, null, 2) + "\n");
          } else {
            renderBenchJinaTable(report);
          }
        } catch (err) {
          console.error(
            `${c.red}Bench failed: ${err instanceof Error ? err.message : String(err)}${c.reset}`,
          );
          process.exit(1);
        }
        break;
      }

      if (!firstArg) {
        console.error("Usage:");
        console.error("  qmd bench <fixture.json> [--json] [-c collection]   Search quality benchmark");
        console.error("  qmd bench jina [options]                            Latency/throughput benchmark");
        console.error("");
        console.error("qmd bench jina options:");
        console.error("  --size <n>         Number of synthetic docs (default 50)");
        console.error("  --doc-len <n>      Characters per doc (default 500)");
        console.error("  --provider <p>     local | jina | both (default both)");
        console.error("  --runs <n>         Repeat the full bench N times to reduce noise (default 1, max 100)");
        console.error("  --skip-rerank      Skip the rerank stage");
        console.error("  --skip-single      Skip the single-embed latency stage");
        console.error("  --json             JSON output");
        process.exit(1);
      }

      const { runBenchmark } = await import("../bench/bench.js");
      const benchCollection = cli.opts.collection;
      await runBenchmark(firstArg, {
        json: !!(cli.opts as { json?: boolean }).json,
        collection: Array.isArray(benchCollection) ? benchCollection[0] : benchCollection,
      });
      break;
    }

    case "mcp": {
      const sub = cli.args[0]; // stop | status | undefined

      // Cache dir for PID/log files — same dir as the index
      const cacheDir = process.env.XDG_CACHE_HOME
        ? resolve(process.env.XDG_CACHE_HOME, "qmd")
        : resolve(homedir(), ".cache", "qmd");
      const pidPath = resolve(cacheDir, "mcp.pid");

      // Subcommands take priority over flags
      if (sub === "stop") {
        if (!existsSync(pidPath)) {
          console.log("Not running (no PID file).");
          process.exit(0);
        }
        const pid = parseInt(readFileSync(pidPath, "utf-8").trim());
        try {
          process.kill(pid, 0); // alive?
          process.kill(pid, "SIGTERM");
          unlinkSync(pidPath);
          console.log(`Stopped QMD MCP server (PID ${pid}).`);
        } catch {
          unlinkSync(pidPath);
          console.log("Cleaned up stale PID file (server was not running).");
        }
        process.exit(0);
      }

      if (cli.values.http) {
        const port = Number(cli.values.port) || 8181;

        if (cli.values.daemon) {
          // Guard: check if already running
          if (existsSync(pidPath)) {
            const existingPid = parseInt(readFileSync(pidPath, "utf-8").trim());
            try {
              process.kill(existingPid, 0); // alive?
              console.error(`Already running (PID ${existingPid}). Run 'qmd mcp stop' first.`);
              process.exit(1);
            } catch {
              // Stale PID file — continue
            }
          }

          mkdirSync(cacheDir, { recursive: true });
          const logPath = resolve(cacheDir, "mcp.log");
          const logFd = openSync(logPath, "w"); // truncate — fresh log per daemon run
          const selfPath = fileURLToPath(import.meta.url);
          const spawnArgs = selfPath.endsWith(".ts")
            ? ["--import", pathJoin(dirname(selfPath), "..", "..", "node_modules", "tsx", "dist", "esm", "index.mjs"), selfPath, "mcp", "--http", "--port", String(port)]
            : [selfPath, "mcp", "--http", "--port", String(port)];
          const child = nodeSpawn(process.execPath, spawnArgs, {
            stdio: ["ignore", logFd, logFd],
            detached: true,
          });
          child.unref();
          closeSync(logFd); // parent's copy; child inherited the fd

          writeFileSync(pidPath, String(child.pid));
          console.log(`Started on http://localhost:${port}/mcp (PID ${child.pid})`);
          console.log(`Logs: ${logPath}`);
          process.exit(0);
        }

        // Foreground HTTP mode — remove top-level cursor handlers so the
        // async cleanup handlers in startMcpHttpServer actually run.
        process.removeAllListeners("SIGTERM");
        process.removeAllListeners("SIGINT");
        const { startMcpHttpServer } = await import("../mcp/server.js");
        try {
          await startMcpHttpServer(port);
        } catch (e: any) {
          if (e?.code === "EADDRINUSE") {
            console.error(`Port ${port} already in use. Try a different port with --port.`);
            process.exit(1);
          }
          throw e;
        }
      } else {
        // Default: stdio transport
        const { startMcpServer } = await import("../mcp/server.js");
        await startMcpServer();
      }
      break;
    }

    case "skill": {
      const subcommand = cli.args[0];
      switch (subcommand) {
        case "show": {
          showSkill();
          break;
        }

        case "install": {
          try {
            await installSkill(Boolean(cli.values.global), Boolean(cli.values.force), Boolean(cli.values.yes));
          } catch (error) {
            console.error(error instanceof Error ? error.message : String(error));
            process.exit(1);
          }
          break;
        }

        case "help":
        case undefined: {
          console.log("Usage: qmd skill <show|install> [options]");
          console.log("");
          console.log("Commands:");
          console.log("  show                 Print the packaged QMD skill");
          console.log("  install              Install into ./.agents/skills/qmd");
          console.log("");
          console.log("Options:");
          console.log("  --global             Install into ~/.agents/skills/qmd");
          console.log("  --yes                Also create the .claude/skills/qmd symlink");
          console.log("  -f, --force          Replace existing install or symlink");
          process.exit(0);
        }

        default:
          console.error(`Unknown subcommand: ${subcommand}`);
          console.error("Run 'qmd skill help' for usage");
          process.exit(1);
      }
      break;
    }

    case "cleanup": {
      const db = getDb();

      // 1. Clear llm_cache
      const cacheCount = deleteLLMCache(db);
      console.log(`${c.green}✓${c.reset} Cleared ${cacheCount} cached API responses`);

      // 2. Remove orphaned vectors
      const orphanedVecs = cleanupOrphanedVectors(db);
      if (orphanedVecs > 0) {
        console.log(`${c.green}✓${c.reset} Removed ${orphanedVecs} orphaned embedding chunks`);
      } else {
        console.log(`${c.dim}No orphaned embeddings to remove${c.reset}`);
      }

      // 3. Remove inactive documents
      const inactiveDocs = deleteInactiveDocuments(db);
      if (inactiveDocs > 0) {
        console.log(`${c.green}✓${c.reset} Removed ${inactiveDocs} inactive document records`);
      }

      // 4. Vacuum to reclaim space
      vacuumDatabase(db);
      console.log(`${c.green}✓${c.reset} Database vacuumed`);

      closeDb();
      break;
    }

    default:
      console.error(`Unknown command: ${cli.command}`);
      console.error("Run 'qmd --help' for usage.");
      process.exit(1);
  }

  if (cli.command !== "mcp") {
    await disposeDefaultLlamaCpp();
    process.exit(0);
  }

} // end if (main module)