@autorag/librarian 2.10.4 → 2.10.5

package/dist/index.js

@@ -5142,7 +5142,20 @@ function configuredVectorDimension(workspacePath) {
   const dimension = options.dimension;
   return typeof dimension === "number" ? dimension : undefined;
 }
-function rewriteEmbedderConfig(workspacePath, embedder, chunkerConfig = {}) {
+function configuredLexicalLanguage(workspacePath) {
+  let raw;
+  try {
+    raw = readFileSync8(minSyncConfigPath(workspacePath), "utf8");
+  } catch {
+    return;
+  }
+  const lexical = parse(raw).lexical;
+  if (typeof lexical !== "object" || lexical === null || Array.isArray(lexical))
+    return "simple";
+  const language = lexical.language;
+  return typeof language === "string" ? language : "simple";
+}
+function rewriteEmbedderConfig(workspacePath, embedder, managed = {}) {
   const configPath = minSyncConfigPath(workspacePath);
   let raw;
   try {
@@ -5185,8 +5198,13 @@ function rewriteEmbedderConfig(workspacePath, embedder, chunkerConfig = {}) {
   }
   if (embedder.dimension !== undefined)
     options.dimension = embedder.dimension;
-  if (chunkerConfig.maxChunkSize !== undefined)
-    chunkerOptions.max_chunk_size = chunkerConfig.maxChunkSize;
+  if (managed.maxChunkSize !== undefined)
+    chunkerOptions.max_chunk_size = managed.maxChunkSize;
+  if (managed.lexicalLanguage !== undefined) {
+    const lexical = typeof parsed.lexical === "object" && parsed.lexical !== null && !Array.isArray(parsed.lexical) ? parsed.lexical : {};
+    parsed.lexical = lexical;
+    lexical.language = managed.lexicalLanguage;
+  }
   writeFileSync6(configPath, stringify(parsed));
   return true;
 }
@@ -5198,6 +5216,66 @@ var init_embedder_config = __esm(() => {
   };
 });
 
+// src/minsync/kiwi.ts
+import { existsSync as existsSync10 } from "node:fs";
+import { homedir as homedir2 } from "node:os";
+import { posix as posix3, win32 as win323 } from "node:path";
+function kiwiAvailable(environment = defaultEnvironment()) {
+  const { env, exists } = environment;
+  const library = env.KIWI_LIBRARY_PATH;
+  const model = env.KIWI_MODEL_PATH;
+  const libraryFound = library !== undefined && library !== "" ? exists(library) : libraryCandidates(environment).some(exists);
+  const modelFound = model !== undefined && model !== "" ? exists(model) : modelCandidates(environment).some(exists);
+  return libraryFound && modelFound;
+}
+function defaultEnvironment() {
+  let home;
+  try {
+    home = homedir2();
+  } catch {
+    home = undefined;
+  }
+  return { env: process.env, platform: process.platform, home, exists: existsSync10 };
+}
+function userRoots({ env, platform, home }) {
+  if (platform === "win32") {
+    const roots = [];
+    if (env.LOCALAPPDATA)
+      roots.push(win323.join(env.LOCALAPPDATA, "kiwi"));
+    const profile = env.USERPROFILE ?? home;
+    if (profile)
+      roots.push(win323.join(profile, "AppData", "Local", "kiwi"));
+    roots.push("C:\\kiwi", "C:\\Program Files\\Kiwi");
+    return roots;
+  }
+  return home ? [posix3.join(home, ".local", "kiwi")] : [];
+}
+function libraryCandidates(environment) {
+  const { platform } = environment;
+  if (platform === "win32")
+    return userRoots(environment).map((root) => win323.join(root, "lib", "kiwi.dll"));
+  const file = platform === "darwin" ? "libkiwi.dylib" : "libkiwi.so";
+  const system = platform === "darwin" ? ["/usr/local/lib", "/opt/homebrew/lib"] : ["/usr/local/lib", "/usr/lib"];
+  return [
+    ...userRoots(environment).map((root) => posix3.join(root, "lib", file)),
+    ...system.map((dir) => posix3.join(dir, file))
+  ];
+}
+function modelCandidates(environment) {
+  const { platform } = environment;
+  if (platform === "win32")
+    return userRoots(environment).map((root) => win323.join(root, "models", "cong", "base"));
+  const system = platform === "darwin" ? ["/usr/local/models", "/opt/homebrew/models", "/usr/local/share/kiwi/models"] : ["/usr/local/models", "/usr/local/share/kiwi/models", "/usr/share/kiwi/models"];
+  return [
+    ...userRoots(environment).map((root) => posix3.join(root, "models", "cong", "base")),
+    ...system.map((dir) => posix3.join(dir, "cong", "base"))
+  ];
+}
+var init_kiwi = () => {};
+
+// src/minsync/lexical-language.ts
+var MINSYNC_LEXICAL_LANGUAGE = "multilingual";
+
 // src/minsync/local-embedder.ts
 import { spawn as spawn7 } from "node:child_process";
 import { request } from "node:http";
@@ -5386,7 +5464,7 @@ var init_process = __esm(() => {
 });
 
 // src/minsync/client.ts
-import { existsSync as existsSync10, readFileSync as readFileSync9, writeFileSync as writeFileSync7 } from "node:fs";
+import { existsSync as existsSync11, readFileSync as readFileSync9, writeFileSync as writeFileSync7 } from "node:fs";
 import { join as join14 } from "node:path";
 
 class MinSyncClient {
@@ -5481,7 +5559,7 @@ class MinSyncClient {
     return previous === undefined || JSON.stringify(previous) !== JSON.stringify(identity);
   }
   async sync(force = false, signal) {
-    if (!existsSync10(this.binaryPath))
+    if (!existsSync11(this.binaryPath))
       throw minSyncBinaryMissingError();
     let effective = { config: {} };
     try {
@@ -5519,14 +5597,16 @@ class MinSyncClient {
       }
     }
     const spawnOptions = signal === undefined ? {} : { signal };
-    const initialized = existsSync10(minSyncConfigPath(this.workspacePath));
+    const lexical = this.resolveLexicalLanguage();
+    const initialized = existsSync11(minSyncConfigPath(this.workspacePath));
     if (!initialized) {
       const initArgs = ["init", "--format", "json"];
       if (this.embedder?.id) {
         initArgs.push("--embedder", this.embedder.id);
       }
+      initArgs.push("--language", lexical.language);
       const init = await this.spawn(initArgs, spawnOptions);
-      if (!init.ok || !existsSync10(minSyncConfigPath(this.workspacePath))) {
+      if (!init.ok || !existsSync11(minSyncConfigPath(this.workspacePath))) {
         return {
           ok: false,
           synced: 0,
@@ -5538,20 +5618,29 @@ class MinSyncClient {
     const chunkSizeChanged = this.maxChunkSize !== undefined && configuredMaxChunkSize(this.workspacePath) !== this.maxChunkSize;
     const dimensionChanged = embedder.dimension !== undefined && configuredVectorDimension(this.workspacePath) !== embedder.dimension;
     const configPath = minSyncConfigPath(this.workspacePath);
-    const shouldRewriteConfig = this.embedder !== undefined || this.maxChunkSize !== undefined || effective.identity !== undefined;
-    const originalConfig = shouldRewriteConfig ? readConfigSnapshot(configPath) : undefined;
-    const configRewritten = shouldRewriteConfig && rewriteEmbedderConfig(this.workspacePath, embedder, { maxChunkSize: this.maxChunkSize });
+    const originalConfig = readConfigSnapshot(configPath);
+    const configRewritten = rewriteEmbedderConfig(this.workspacePath, embedder, {
+      maxChunkSize: this.maxChunkSize,
+      lexicalLanguage: lexical.language
+    });
     const restoreConfig = () => {
       if (configRewritten && originalConfig !== undefined)
         writeFileSync7(configPath, originalConfig);
     };
     try {
-      return await this.checkAndSync(force || chunkSizeChanged || dimensionChanged, signal, effective, restoreConfig);
+      const result = await this.checkAndSync(force || chunkSizeChanged || dimensionChanged, signal, effective, restoreConfig);
+      return result.ok && lexical.fallback !== undefined ? { ...result, lexicalFallback: lexical.fallback } : result;
     } catch (error) {
       restoreConfig();
       throw error;
     }
   }
+  resolveLexicalLanguage() {
+    const requested = MINSYNC_LEXICAL_LANGUAGE;
+    if (kiwiAvailable() || configuredLexicalLanguage(this.workspacePath) === requested)
+      return { language: requested };
+    return { language: "simple", fallback: { requested, used: "simple" } };
+  }
   lockHeldResult(stderr) {
     return {
       ok: false,
@@ -5604,7 +5693,7 @@ class MinSyncClient {
     }
     const identityChanged = effective.identity !== undefined && this.identityMismatch(effective.identity);
     const fullReindex = forceFull || identityChanged;
-    const syncArgs = existsSync10(cursorPath) && !fullReindex ? ["sync", "--format", "json"] : ["sync", "--full", "--format", "json"];
+    const syncArgs = existsSync11(cursorPath) && !fullReindex ? ["sync", "--format", "json"] : ["sync", "--full", "--format", "json"];
     const result = await this.spawn(syncArgs, spawnOptions);
     if (!result.ok && result.code === MINSYNC_LOCK_HELD_EXIT_CODE) {
       restoreConfig();
@@ -5625,7 +5714,7 @@ class MinSyncClient {
         }
       };
     }
-    if (!existsSync10(cursorPath)) {
+    if (!existsSync11(cursorPath)) {
       restoreConfig();
       return { ok: false, synced: 0, workspacePath: this.workspacePath, reason: "not-ready: missing cursor" };
     }
@@ -5646,7 +5735,7 @@ class MinSyncClient {
     };
   }
   async query(text, topK, mode = "vector", signal) {
-    if (!existsSync10(this.binaryPath))
+    if (!existsSync11(this.binaryPath))
       throw minSyncBinaryMissingError();
     const result = await this.spawn(["query", "--format", "json", "--mode", mode, "-k", String(topK), text], {
       ...signal === undefined ? {} : { signal }
@@ -5740,6 +5829,7 @@ var MINSYNC_OLLAMA_MIGRATION_MESSAGE = "This workspace uses the legacy 768-dimen
 var init_client4 = __esm(() => {
   init_embedder_config();
   init_errors();
+  init_kiwi();
   init_local_embedder();
   init_process();
   API_KEY_ENV_PATTERN = /^[A-Za-z_][A-Za-z0-9_]*$/;
@@ -5762,14 +5852,14 @@ var init_client4 = __esm(() => {
 
 // src/minsync/installer.ts
 import { createHash as createHash9 } from "node:crypto";
-import { copyFileSync, createWriteStream as createWriteStream2, existsSync as existsSync11, mkdirSync as mkdirSync8, renameSync as renameSync4, rmSync as rmSync5 } from "node:fs";
+import { copyFileSync, createWriteStream as createWriteStream2, existsSync as existsSync12, mkdirSync as mkdirSync8, renameSync as renameSync4, rmSync as rmSync5 } from "node:fs";
 import { chmod as chmod2, mkdtemp } from "node:fs/promises";
 import { get } from "node:https";
 import { basename as basename7, delimiter as delimiter2, dirname as dirname7, join as join15 } from "node:path";
 import { pipeline } from "node:stream/promises";
 async function ensureMinSyncBinary(options) {
   const binaryPath = join15(options.root, ".autorag", "bin", executableName(options.platform ?? process.platform));
-  if (existsSync11(binaryPath))
+  if (existsSync12(binaryPath))
     return { binaryPath, version: "cached" };
   mkdirSync8(dirname7(binaryPath), { recursive: true });
   const cargoInstaller = options.cargoInstaller ?? ((destination) => installMinSyncFromCargoIfAvailable(options, destination));
@@ -5817,7 +5907,7 @@ async function installMinSyncFromCargo(options, destination, cargo) {
     throw new MinSyncReleaseError(result.stderr || `Could not install MinSync ${MINSYNC_VERSION} from crates.io`);
   }
   const installedBinary = join15(cargoRoot, "bin", executableName(options.platform ?? process.platform));
-  if (!existsSync11(installedBinary)) {
+  if (!existsSync12(installedBinary)) {
     throw new MinSyncReleaseError(`Cargo did not produce the MinSync ${MINSYNC_VERSION} binary`);
   }
   copyFileSync(installedBinary, destination);
@@ -5949,7 +6039,7 @@ function lookupCargo(env, platform) {
     if (dir.length === 0)
       continue;
     const candidate = join15(dir, execName);
-    if (existsSync11(candidate))
+    if (existsSync12(candidate))
       return candidate;
   }
   return;
@@ -5980,7 +6070,7 @@ function isGitHubAsset(value) {
 function isRecord5(value) {
   return typeof value === "object" && value !== null;
 }
-var LATEST_RELEASE_URL = "https://api.github.com/repos/NomaDamas/MinSync/releases/latest", SHA256_HEX_PATTERN, MINSYNC_VERSION = "0.4.8", MINSYNC_CARGO_INSTALL_TIMEOUT_MS, CARGO_PROBE_TIMEOUT_MS = 5000, MinSyncReleaseError;
+var LATEST_RELEASE_URL = "https://api.github.com/repos/NomaDamas/MinSync/releases/latest", SHA256_HEX_PATTERN, MINSYNC_VERSION = "0.4.10", MINSYNC_CARGO_INSTALL_TIMEOUT_MS, CARGO_PROBE_TIMEOUT_MS = 5000, MinSyncReleaseError;
 var init_installer2 = __esm(() => {
   init_process();
   SHA256_HEX_PATTERN = /^[a-f0-9]{64}$/;
@@ -6008,7 +6098,7 @@ var init_paths2 = __esm(() => {
 import { randomUUID as randomUUID4 } from "node:crypto";
 import {
   copyFileSync as copyFileSync2,
-  existsSync as existsSync12,
+  existsSync as existsSync13,
   lstatSync as lstatSync2,
   mkdirSync as mkdirSync9,
   readdirSync as readdirSync2,
@@ -6023,8 +6113,8 @@ function syncMinSyncWorkspace(root, options = {}) {
   const filesRoot = join17(workspacePath, "files");
   const index = loadMirrorIndex(root);
   const candidates = Object.values(index.entries).sort((a, b) => a.virtualPath.localeCompare(b.virtualPath));
-  const excluded = candidates.filter((entry) => normalizeVirtualPath(entry.virtualPath) !== entry.virtualPath || !existsSync12(entry.outputPath)).map((entry) => entry.virtualPath);
-  const entries = candidates.filter((entry) => normalizeVirtualPath(entry.virtualPath) === entry.virtualPath && existsSync12(entry.outputPath)).map((entry) => {
+  const excluded = candidates.filter((entry) => normalizeVirtualPath(entry.virtualPath) !== entry.virtualPath || !existsSync13(entry.outputPath)).map((entry) => entry.virtualPath);
+  const entries = candidates.filter((entry) => normalizeVirtualPath(entry.virtualPath) === entry.virtualPath && existsSync13(entry.outputPath)).map((entry) => {
     const minSyncPath = minSyncDocumentPath(workspacePath, entry.virtualPath);
     return {
       virtualPath: entry.virtualPath,
@@ -6161,7 +6251,7 @@ var init_workspace = __esm(() => {
 });
 
 // src/minsync/method.ts
-import { accessSync, constants, existsSync as existsSync13, realpathSync as realpathSync4 } from "node:fs";
+import { accessSync, constants, existsSync as existsSync14, realpathSync as realpathSync4 } from "node:fs";
 import { basename as basename9, delimiter as delimiter3, join as join18, normalize as normalize2 } from "node:path";
 function lookupInPath(env) {
   const pathEnv = env.PATH;
@@ -6257,19 +6347,19 @@ class MinSyncVectorMethod {
   }
   isBinaryMissing() {
     if (this.binaryPath !== undefined)
-      return !existsSync13(this.binaryPath);
+      return !existsSync14(this.binaryPath);
     if (lookupInPath(process.env) !== undefined)
       return false;
     const cachedBinary = join18(this.root, ".autorag", "bin", executableName(process.platform));
-    if (existsSync13(cachedBinary))
+    if (existsSync14(cachedBinary))
       return false;
     return true;
   }
   isExplicitBinaryMissing() {
-    return this.binaryPath !== undefined && !existsSync13(this.binaryPath);
+    return this.binaryPath !== undefined && !existsSync14(this.binaryPath);
   }
   isReady() {
-    return existsSync13(join18(this.workspacePath, ".minsync", "cursor.json"));
+    return existsSync14(join18(this.workspacePath, ".minsync", "cursor.json"));
   }
   async retrieve(query, options) {
     if ((workspaceSyncInFlight.get(this.workspacePath) ?? 0) > 0) {
@@ -6322,7 +6412,7 @@ class MinSyncVectorMethod {
     return results;
   }
   async resolveBinary(options = {}) {
-    if (this.binaryPath && existsSync13(this.binaryPath))
+    if (this.binaryPath && existsSync14(this.binaryPath))
       return this.binaryPath;
     if (this.binaryPath !== undefined)
       throw minSyncBinaryMissingError();
@@ -6330,7 +6420,7 @@ class MinSyncVectorMethod {
     if (pathBinary)
       return pathBinary;
     const cachedBinary = join18(this.root, ".autorag", "bin", executableName(process.platform));
-    if (existsSync13(cachedBinary))
+    if (existsSync14(cachedBinary))
       return cachedBinary;
     if (this.autoInstall && options.install !== false) {
       try {
@@ -7033,7 +7123,7 @@ var init_defaults = __esm(() => {
 });
 
 // src/parser/isolated.ts
-import { existsSync as existsSync14 } from "node:fs";
+import { existsSync as existsSync15 } from "node:fs";
 import { dirname as dirname10, join as join21 } from "node:path";
 import { fileURLToPath as fileURLToPath2, pathToFileURL } from "node:url";
 import { Worker } from "node:worker_threads";
@@ -7172,7 +7262,7 @@ function defaultWorkerUrl() {
     join21(here, "parser", "parse-worker.js"),
     join21(here, "..", "parser", "parse-worker.js")
   ]) {
-    if (existsSync14(candidate))
+    if (existsSync15(candidate))
       return pathToFileURL(candidate);
   }
   throw new Error(`The AutoRAG parse worker entry was not found near ${here}.`);
@@ -7184,7 +7274,7 @@ var init_isolated = __esm(() => {
 });
 
 // src/mirror/sync.ts
-import { existsSync as existsSync15, rmSync as rmSync7 } from "node:fs";
+import { existsSync as existsSync16, rmSync as rmSync7 } from "node:fs";
 import { opendir, readFile as readFile3, stat as stat5 } from "node:fs/promises";
 import { extname as extname3, resolve as resolve11, sep as sep6 } from "node:path";
 import { setTimeout as sleep3 } from "node:timers/promises";
@@ -7376,7 +7466,7 @@ async function syncWithRegistry(options, registry, isolated) {
     }
     const previousEntry = previous.entries[entry.virtualPath];
     const outputPath = parsedOutputPath(options.root, entry.virtualPath);
-    const unchanged = !options.force && previousEntry?.sourceMtimeNs === entry.mtimeNs && previousEntry.sourceSizeBytes === entry.sizeBytes && previousEntry.parserName === parser.name && existsSync15(outputPath);
+    const unchanged = !options.force && previousEntry?.sourceMtimeNs === entry.mtimeNs && previousEntry.sourceSizeBytes === entry.sizeBytes && previousEntry.parserName === parser.name && existsSync16(outputPath);
     if (!unchanged) {
       let bytes;
       try {
@@ -7757,7 +7847,7 @@ function removePrevious(root, index, virtualPath) {
   return 1;
 }
 function removeFile(path) {
-  if (existsSync15(path))
+  if (existsSync16(path))
     rmSync7(path, { force: true });
 }
 var DEFAULT_MAX_SOURCE_BYTES, MIRROR_CHECKPOINT_EVERY = 25, DEFAULT_MAX_DELETE_RATIO = 0.3, MASS_DELETE_FLOOR = 100, DEFAULT_MAX_SCANNED_ENTRIES = 2000000, DEFAULT_MAX_WALK_DEPTH = 64, PRODUCT_ARTIFACT_FILE_NAMES, SKIP_DIR_NAMES, PARSER_FAILURE_MESSAGE = "The registered parser failed on this file; it was skipped during indexing.", PARSER_CRASH_MESSAGE = "The registered parser crashed unexpectedly on this file; it was skipped during indexing.", DEFAULT_TRANSIENT_RETRIES = 3, DEFAULT_TRANSIENT_BACKOFF_MS = 50, TRANSIENT_ERROR_CODES, WORKSPACE_ERROR_CODES, MirrorWorkspaceError, STANDALONE_OCR_IMAGE_EXTENSIONS, RETRYABLE_SKIP_MESSAGES, SEARCH_ROOT_UNAVAILABLE_MESSAGES;
@@ -11482,6 +11572,20 @@ function interactiveKeyOf(data) {
     return "followUp";
   return;
 }
+function stripTelegramEnvelope(text) {
+  return text.replace(/^\s*\[telegram\]\s?/u, "").replace(/\n*\[time\][^\n]*/gu, "").trim();
+}
+function userInputText(event) {
+  if (event.source === "interactive")
+    return event.text.trim();
+  if (event.source === "extension" && event.text.trimStart().startsWith(TELEGRAM_INPUT_PREFIX)) {
+    return stripTelegramEnvelope(event.text);
+  }
+  if (event.source === "extension" && event.text.trimStart().startsWith(WHATSAPP_INPUT_PREFIX)) {
+    return event.text.trim();
+  }
+  return;
+}
 function createAutoRAGInteractiveExtension(getSystemPrompt, contextTransform, options, updateNotice) {
   return (pi) => {
     pi.on("before_agent_start", () => ({ systemPrompt: getSystemPrompt() }));
@@ -11512,8 +11616,8 @@ function createAutoRAGInteractiveExtension(getSystemPrompt, contextTransform, op
       });
     }
     pi.on("input", (event) => {
-      const text = event.text.trim();
-      if (event.source !== "interactive" || text.length === 0)
+      const text = userInputText(event);
+      if (text === undefined || text.length === 0)
         return;
       if (text.startsWith("/") && isSearchRunning?.() !== true)
         return;
@@ -11610,7 +11714,9 @@ async function createAutoRAGPiInteractiveRuntime(options) {
       agentDir: runtimeAgentDir,
       modelRuntime,
       settingsManager,
+      ...options.extensionFlags === undefined ? {} : { extensionFlagValues: new Map(Object.entries(options.extensionFlags)) },
       resourceLoaderOptions: {
+        ...options.extensionPaths === undefined ? {} : { additionalExtensionPaths: [...options.extensionPaths] },
         extensionFactories: [
           createAutoRAGInteractiveExtension(options.getSystemPrompt, options.contextTransform, { onInput: options.onInput, isSearchRunning: options.isSearchRunning, onKey: options.onKey }, options.updateNotice),
           ...options.extensionFactories ?? []
@@ -11644,7 +11750,7 @@ async function createAutoRAGPiInteractiveRuntime(options) {
   });
   return { runtime, dispose: () => runtime.dispose() };
 }
-var PI_BUILTIN_TOOL_NAMES, EMBEDDED_AUTH_MARKER = "<autorag-embedded-model>";
+var PI_BUILTIN_TOOL_NAMES, EMBEDDED_AUTH_MARKER = "<autorag-embedded-model>", TELEGRAM_INPUT_PREFIX = "[telegram]", WHATSAPP_INPUT_PREFIX = "Message from ";
 var init_pi_session = __esm(() => {
   init_provider();
   init_context_token_guard();
@@ -11900,11 +12006,11 @@ var init_jikji_find_tool = __esm(() => {
 
 // src/agent/local-model.ts
 import { readFileSync as readFileSync12 } from "node:fs";
-import { homedir as homedir2 } from "node:os";
+import { homedir as homedir3 } from "node:os";
 import { join as join24 } from "node:path";
 import { parse as parse2 } from "smol-toml";
 function loadLocalAutoRAGModel(options = {}) {
-  const configPath = options.configPath ?? join24(homedir2(), ".codex", "config.toml");
+  const configPath = options.configPath ?? join24(homedir3(), ".codex", "config.toml");
   const config = parse2(readFileSync12(configPath, "utf8"));
   const provider = config.model_provider;
   if (typeof provider !== "string" || provider.length === 0) {
@@ -11947,7 +12053,7 @@ var init_local_model = () => {};
 
 // src/p2p/approval-store.ts
 import {
-  existsSync as existsSync16,
+  existsSync as existsSync17,
   mkdirSync as mkdirSync11,
   readdirSync as readdirSync3,
   readFileSync as readFileSync13,
@@ -12012,7 +12118,7 @@ function savePendingPeerRequest(workspacePath, request) {
 }
 function loadPendingPeerRequest(workspacePath, id) {
   const path = requestPath(workspacePath, id);
-  if (!existsSync16(path))
+  if (!existsSync17(path))
     return;
   try {
     return parsePending(JSON.parse(readFileSync13(path, "utf8")));
@@ -12022,7 +12128,7 @@ function loadPendingPeerRequest(workspacePath, id) {
 }
 function listPendingPeerRequests(workspacePath) {
   const dir = requestsDir(workspacePath);
-  if (!existsSync16(dir))
+  if (!existsSync17(dir))
     return [];
   const requests = [];
   for (const name of readdirSync3(dir)) {
@@ -12044,7 +12150,7 @@ function writePeerRequestDecision(workspacePath, id, decision) {
 }
 function loadPeerRequestDecision(workspacePath, id) {
   const path = decisionPath(workspacePath, id);
-  if (!existsSync16(path))
+  if (!existsSync17(path))
     return;
   try {
     return parseDecision(JSON.parse(readFileSync13(path, "utf8")));
@@ -12378,7 +12484,7 @@ var init_egress_gate = __esm(() => {
 });
 
 // src/p2p/policy.ts
-import { existsSync as existsSync17, mkdirSync as mkdirSync12, readFileSync as readFileSync14, writeFileSync as writeFileSync10 } from "node:fs";
+import { existsSync as existsSync18, mkdirSync as mkdirSync12, readFileSync as readFileSync14, writeFileSync as writeFileSync10 } from "node:fs";
 import { basename as basename10, dirname as dirname13, join as join26, resolve as resolve12 } from "node:path";
 import { parse as parseToml } from "smol-toml";
 function isRecord11(value) {
@@ -12458,7 +12564,7 @@ function normalizePolicyEntries(value, workspacePath, context) {
   return entries;
 }
 function readJsonFile(path) {
-  if (!existsSync17(path))
+  if (!existsSync18(path))
     return;
   let raw;
   try {
@@ -12473,7 +12579,7 @@ function readJsonFile(path) {
   }
 }
 function readTomlFile(path) {
-  if (!existsSync17(path))
+  if (!existsSync18(path))
     return;
   let raw;
   try {
@@ -12614,7 +12720,7 @@ function globToRegExp(pattern) {
   return new RegExp(`${expression}$`, "u");
 }
 function loadSeenSources(path, workspacePath) {
-  if (!existsSync17(path))
+  if (!existsSync18(path))
     return { promoted: new Set, pending: new Set };
   let parsed;
   try {
@@ -12921,7 +13027,7 @@ var init_wire = __esm(() => {
 
 // src/p2p/simplex-query-store.ts
 import { randomUUID as randomUUID5 } from "node:crypto";
-import { existsSync as existsSync18, mkdirSync as mkdirSync13, readdirSync as readdirSync4, readFileSync as readFileSync15, renameSync as renameSync7, unlinkSync as unlinkSync4, writeFileSync as writeFileSync11 } from "node:fs";
+import { existsSync as existsSync19, mkdirSync as mkdirSync13, readdirSync as readdirSync4, readFileSync as readFileSync15, renameSync as renameSync7, unlinkSync as unlinkSync4, writeFileSync as writeFileSync11 } from "node:fs";
 import { dirname as dirname14, join as join27 } from "node:path";
 import { Value as Value2 } from "typebox/value";
 function statePath(workspacePath, directory, id) {
@@ -12957,7 +13063,7 @@ function parseState(value) {
   return value;
 }
 function readState(path) {
-  if (!existsSync18(path))
+  if (!existsSync19(path))
     return;
   try {
     return parseState(JSON.parse(readFileSync15(path, "utf8")));
@@ -12984,14 +13090,14 @@ function moveState(workspacePath, from, to, state) {
 }
 function recoverClaims(workspacePath) {
   const directory = join27(workspacePath, QUERY_DIR, PENDING_DIR);
-  if (!existsSync18(directory))
+  if (!existsSync19(directory))
     return;
   for (const name of readdirSync4(directory)) {
     if (!name.endsWith(".json.claim"))
       continue;
     const claim = join27(directory, name);
     const restored = join27(directory, name.slice(0, -".claim".length));
-    if (!existsSync18(restored))
+    if (!existsSync19(restored))
       renameSync7(claim, restored);
   }
 }
@@ -13015,7 +13121,7 @@ function loadSimplexQueryState(workspacePath, id) {
 }
 function listPendingSimplexQueryStates(workspacePath) {
   const directory = join27(workspacePath, QUERY_DIR, PENDING_DIR);
-  if (!existsSync18(directory))
+  if (!existsSync19(directory))
     return [];
   return readdirSync4(directory).filter((name) => name.endsWith(".json")).map((name) => readState(join27(directory, name))).filter((state) => state?.status === "pending").filter((state) => loadSimplexQueryState(workspacePath, state.id)?.status === "pending").sort((left, right) => left.createdAt.localeCompare(right.createdAt));
 }
@@ -13158,7 +13264,7 @@ var init_simplex_query_store = __esm(() => {
 });
 
 // src/p2p/simplex-server.ts
-import { existsSync as existsSync19, mkdirSync as mkdirSync14, readFileSync as readFileSync16, writeFileSync as writeFileSync12 } from "node:fs";
+import { existsSync as existsSync20, mkdirSync as mkdirSync14, readFileSync as readFileSync16, writeFileSync as writeFileSync12 } from "node:fs";
 import { join as join28 } from "node:path";
 import { Type as Type10 } from "typebox";
 import { Value as Value3 } from "typebox/value";
@@ -13187,7 +13293,7 @@ function rankSimplexPeerTargets(query, registry) {
 }
 function loadSimplexPeerRegistry(workspacePath) {
   const path = join28(workspacePath, PEERS_DIR, PEERS_FILENAME);
-  if (!existsSync19(path))
+  if (!existsSync20(path))
     return {};
   try {
     const parsed = JSON.parse(readFileSync16(path, "utf8"));
@@ -14312,14 +14418,14 @@ var init_search_single_datasource_tool = __esm(() => {
 });
 
 // src/agent/self-config.ts
-import { existsSync as existsSync20, readFileSync as readFileSync18, rmSync as rmSync9, writeFileSync as writeFileSync14 } from "node:fs";
+import { existsSync as existsSync21, readFileSync as readFileSync18, rmSync as rmSync9, writeFileSync as writeFileSync14 } from "node:fs";
 import { dirname as dirname16, join as join29 } from "node:path";
 import { fileURLToPath as fileURLToPath3 } from "node:url";
 function findPackagedSetupSkill() {
   let dir = dirname16(fileURLToPath3(import.meta.url));
   for (let depth = 0;depth < 6; depth += 1) {
     const candidate = join29(dir, SETUP_SKILL_RELATIVE_PATH);
-    if (existsSync20(candidate))
+    if (existsSync21(candidate))
       return candidate;
     const parent = dirname16(dir);
     if (parent === dir)
@@ -14342,7 +14448,7 @@ function loadSetupSkill(skillPath) {
   return text.replace(/^---\r?\n[\s\S]*?\r?\n---(?:\r?\n)+/u, "");
 }
 function snapshotConfigFile(configPath) {
-  return existsSync20(configPath) ? readFileSync18(configPath, "utf8") : undefined;
+  return existsSync21(configPath) ? readFileSync18(configPath, "utf8") : undefined;
 }
 async function rollbackIfBroken(options, before) {
   if (options.validate === undefined)
@@ -16865,11 +16971,6 @@ var init_types3 = __esm(() => {
       label: "xAI",
       description: "xAI Responses web_search via the agent's xAI model credentials (XAI_API_KEY)"
     },
-    {
-      value: "perplexity",
-      label: "Perplexity",
-      description: "Anonymous Perplexity ask endpoint; no key or signup required"
-    },
     {
       value: "parallel",
       label: "Parallel",
@@ -16880,16 +16981,21 @@ var init_types3 = __esm(() => {
       label: "SearXNG",
       description: "Advanced: requires a self-hosted SEARXNG_ENDPOINT; never part of the default experience"
     },
-    {
-      value: "startpage",
-      label: "Startpage",
-      description: "Credential-free scrape of Startpage (Google-backed) results; may be bot-challenged"
-    },
     {
       value: "duckduckgo",
       label: "DuckDuckGo",
       description: "Credential-free best-effort fallback; may be bot-challenged on datacenter/shared-egress IPs"
     },
+    {
+      value: "perplexity",
+      label: "Perplexity",
+      description: "Anonymous Perplexity ask endpoint; no key or signup required"
+    },
+    {
+      value: "startpage",
+      label: "Startpage",
+      description: "Credential-free scrape of Startpage (Google-backed) results; may be bot-challenged"
+    },
     {
       value: "ecosia",
       label: "Ecosia",
@@ -18237,12 +18343,11 @@ function eventAnswer(event) {
   }
   return "";
 }
-var PERPLEXITY_ASK_URL = "https://www.perplexity.ai/rest/sse/perplexity_ask", API_VERSION2 = "2.18", DEFLECTION_PATTERN, MAX_ERROR_BYTES4, ANONYMOUS_USER_AGENT = "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/131.0.0.0 Safari/537.36", PERPLEXITY_QUERY_SYNTAX, PerplexityProvider;
+var PERPLEXITY_ASK_URL = "https://www.perplexity.ai/rest/sse/perplexity_ask", API_VERSION2 = "2.18", MAX_ERROR_BYTES4, ANONYMOUS_USER_AGENT = "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/131.0.0.0 Safari/537.36", PERPLEXITY_QUERY_SYNTAX, PerplexityProvider;
 var init_perplexity = __esm(() => {
   init_query();
   init_types3();
   init_utils();
-  DEFLECTION_PATTERN = /sign up|log in|create an account|verify you are human|are you a robot|가입|로그인|회원 ?가입/i;
   MAX_ERROR_BYTES4 = 8 * 1024;
   PERPLEXITY_QUERY_SYNTAX = {
     phrases: true,
@@ -18356,8 +18461,8 @@ var init_perplexity = __esm(() => {
           break;
       }
       const sources = [...sourcesByUrl.values()];
-      if (sources.length === 0 && DEFLECTION_PATTERN.test(answer)) {
-        throw new SearchProviderError("perplexity", "Perplexity anonymous ask was deflected with a sign-up wall", 403);
+      if (sources.length === 0) {
+        throw new SearchProviderError("perplexity", `Perplexity anonymous ask returned no web sources (sign-up wall or conversational deflection): ${answer.slice(0, 120)}`, 403);
       }
       const numResults = params.numSearchResults ?? params.limit;
       return {
@@ -18775,7 +18880,7 @@ var init_browser_headers = __esm(() => {
 });
 
 // src/web/search/browser-loader.ts
-import { existsSync as existsSync21 } from "node:fs";
+import { existsSync as existsSync22 } from "node:fs";
 async function closeBounded(close) {
   let timer;
   try {
@@ -18854,7 +18959,7 @@ function ensureWebSearchBrowserLoader(options = {}) {
       return true;
     const exists = options.exists ?? ((path) => {
       try {
-        return existsSync21(path);
+        return existsSync22(path);
       } catch {
         return false;
       }
@@ -18959,6 +19064,8 @@ function normalizeText(value) {
 function isChallengeResponse(page) {
   if (/\/(?:errors|captcha)\//.test(page.url) || page.url.includes("/sp/captcha"))
     return true;
+  if (ANUBIS_CHALLENGE_PATTERN.test(page.html))
+    return true;
   return page.html.includes("component---src-pages-captcha") || page.html.includes("/sp/captcha");
 }
 function parseSearchFormInputs(html) {
@@ -19063,7 +19170,7 @@ async function callStartpageHtml(params) {
     });
   }
   if (isChallengeResponse(page)) {
-    throw new SearchProviderError("startpage", "Startpage blocked the request with a CAPTCHA challenge. Startpage rate-limits automated searches from datacenter/shared-egress IPs; try another provider such as DuckDuckGo or Mojeek, or retry later.", 429);
+    throw new SearchProviderError("startpage", "Startpage blocked the request with a bot challenge (CAPTCHA or Anubis proof-of-work). Startpage rate-limits automated searches from datacenter/shared-egress IPs; try another provider such as DuckDuckGo or Mojeek, or retry later.", 429);
   }
   if (page.status < 200 || page.status >= 300) {
     const classified = classifyProviderHttpError("startpage", page.status, page.html);
@@ -19089,7 +19196,7 @@ async function searchStartpage(params) {
   }
   return { provider: "startpage", sources };
 }
-var STARTPAGE_HOME_URL = "https://www.startpage.com/", STARTPAGE_SEARCH_URL = "https://www.startpage.com/sp/search", DEFAULT_NUM_RESULTS3 = 10, MAX_NUM_RESULTS3 = 20, RECENCY_TO_STARTPAGE_WITH_DATE, StartpageProvider;
+var STARTPAGE_HOME_URL = "https://www.startpage.com/", STARTPAGE_SEARCH_URL = "https://www.startpage.com/sp/search", DEFAULT_NUM_RESULTS3 = 10, MAX_NUM_RESULTS3 = 20, RECENCY_TO_STARTPAGE_WITH_DATE, ANUBIS_CHALLENGE_PATTERN, StartpageProvider;
 var init_startpage = __esm(() => {
   init_entities();
   init_browser_page();
@@ -19102,6 +19209,7 @@ var init_startpage = __esm(() => {
     month: "m",
     year: "y"
   };
+  ANUBIS_CHALLENGE_PATTERN = /<script\b[^>]*\bid\s*=\s*["']anubis_(?:challenge|version)["']/i;
   StartpageProvider = class StartpageProvider extends SearchProvider {
     id = "startpage";
     label = "Startpage";
@@ -20319,7 +20427,7 @@ var init_web_search_tool = __esm(() => {
 
 // src/agent/agent.ts
 import { randomUUID as randomUUID7 } from "node:crypto";
-import { existsSync as existsSync22, watch as fsWatch, mkdirSync as mkdirSync16, realpathSync as realpathSync5, statSync as statSync2, writeFileSync as writeFileSync15 } from "node:fs";
+import { existsSync as existsSync23, watch as fsWatch, mkdirSync as mkdirSync16, realpathSync as realpathSync5, statSync as statSync2, writeFileSync as writeFileSync15 } from "node:fs";
 import { basename as basename11, dirname as dirname17, join as join31, resolve as resolve13 } from "node:path";
 import { clampThinkingLevel } from "@earendil-works/pi-ai/compat";
 import { getAgentDir as getAgentDir3 } from "@earendil-works/pi-coding-agent";
@@ -20450,6 +20558,9 @@ class AutoRAGAgent {
   piSessionDir;
   persistPiSessions;
   updateNotice;
+  piExtensionPaths;
+  piExtensionFlags;
+  telegram;
   boundPiRuntime;
   memoryEmbedder;
   memoryVectorPath;
@@ -20475,6 +20586,9 @@ class AutoRAGAgent {
     this.piSessionDir = options.piSessionDir;
     this.persistPiSessions = options.persistPiSessions ?? true;
     this.updateNotice = options.updateNotice;
+    this.piExtensionPaths = options.piExtensionPaths;
+    this.piExtensionFlags = options.piExtensionFlags;
+    this.telegram = options.telegram === true;
     this.datasourceSkills = options.datasourceSkills ?? [];
     this.datasourceRetry = options.datasourceRetry ?? {};
     this.datasourceVirtualScopePrefixes = this.datasourceSkills.map((skill) => normalizeVirtualPath2(`/${skill.describe().name}`));
@@ -20946,6 +21060,7 @@ ${context.text}
   }
   async createPiInteractiveRuntime() {
     const resolved = this.configuredModel === undefined ? undefined : await this.resolveSessionModel();
+    const telegramBridge = this.telegram ? (await import("@llblab/pi-telegram")).default : undefined;
     const runtime = await createAutoRAGPiInteractiveRuntime({
       cwd: this.workspaceProjectRoot,
       agentDir: this.piAgentDir,
@@ -20966,11 +21081,14 @@ ${context.text}
       onKey: (key) => this.handleInteractiveKey(key),
       extensionFactories: [
         createEvidencePersistence(this.evidenceLedger),
-        ...this.jevExtension !== undefined ? [this.jevExtension] : []
+        ...this.jevExtension !== undefined ? [this.jevExtension] : [],
+        ...telegramBridge !== undefined ? [telegramBridge] : []
       ],
       ...this.jevExtension !== undefined ? { extensionToolNames: [JEV_TOOL_NAME] } : {},
       remoteSession: this.remoteSession,
-      ...this.updateNotice === undefined ? {} : { updateNotice: this.updateNotice }
+      ...this.updateNotice === undefined ? {} : { updateNotice: this.updateNotice },
+      ...this.piExtensionPaths === undefined ? {} : { extensionPaths: this.piExtensionPaths },
+      ...this.piExtensionFlags === undefined ? {} : { extensionFlags: this.piExtensionFlags }
     });
     this.boundPiRuntime = runtime.runtime;
     const dispose = runtime.dispose;
@@ -21890,22 +22008,22 @@ ${result.content.replace(/\s+/gu, " ")}`).join(`
     const scope = options.scope ? ` Restrict search to virtual path scope ${options.scope}.` : "";
     return `Answer this original query immediately: ${query}${limit}${scope}
 
-` + `Baseline retrieval evidence (already gathered for you):
+Baseline retrieval evidence (already gathered for you):
 ${baseline}
 
-` + `Produce the best complete, self-contained answer you can RIGHT NOW from this evidence. Do NOT call any search, retrieval, or file-reading tools and do NOT wait for more evidence. ` + `If the query is answerable from general knowledge alone, answer directly.
+Produce the best complete, self-contained answer you can RIGHT NOW from this evidence. Do NOT call any search, retrieval, or file-reading tools and do NOT wait for more evidence. If the query is answerable from general knowledge alone, answer directly.
 
-` + `Formatting and content rules for the answer:
-` + `- Unless the user explicitly requests a different format or length, provide the core answer to the user's question in at most 5 bullet points. If additional explanation is necessary, append it after the bullet points.
-` + `- Unless the user explicitly requests otherwise, answer the question directly. Do not include specific file paths, datasource descriptions, or retrieval mechanics in the answer text.
-` + `- ${ANSWER_DELIVERABLE_RULE}
-` + `- ${ANSWER_CITATION_RULE}
-` + `- ${ANSWER_IMAGE_EMBED_RULE}
-` + `- Do not report per-source negative findings (e.g. "no information found in Slack" or "checked Drive but found nothing").
-` + `- When evidence conflicts, treat the freshest (most recent) information as the correct source of truth.
-` + `- If information is incomplete or uncertain, acknowledge it briefly without lengthy explanations, stating that it is difficult to answer fully with the given information and searching continues. If there are partial clues or leads (even if not the exact answer), mention those clues concisely.
+Formatting and content rules for the answer:
+- Unless the user explicitly requests a different format or length, provide the core answer to the user's question in at most 5 bullet points. If additional explanation is necessary, append it after the bullet points.
+- Unless the user explicitly requests otherwise, answer the question directly. Do not include specific file paths, datasource descriptions, or retrieval mechanics in the answer text.
+- ${ANSWER_DELIVERABLE_RULE}
+- ${ANSWER_CITATION_RULE}
+- ${ANSWER_IMAGE_EMBED_RULE}
+- Do not report per-source negative findings (e.g. "no information found in Slack" or "checked Drive but found nothing").
+- When evidence conflicts, treat the freshest (most recent) information as the correct source of truth.
+- If information is incomplete or uncertain, acknowledge it briefly without lengthy explanations, stating that it is difficult to answer fully with the given information and searching continues. If there are partial clues or leads (even if not the exact answer), mention those clues concisely.
 
-` + `Write the answer as your reply, then stop.`;
+Write the answer as your reply, then stop.`;
   }
   isConversationTurn() {
     if (this.remoteSession || this.boundPiRuntime === undefined)
@@ -21915,7 +22033,7 @@ ${baseline}
   buildDirectAnswerPrompt(query) {
     return `Answer this query directly from your own general knowledge: ${query}
 
-` + `It needs no search: it is general knowledge, simple reasoning, or conversation. Do NOT call any search, retrieval, web, or file-reading tools. ` + `Reply naturally and concisely; for small talk, just respond conversationally. Do not cite sources or mention retrieval.`;
+It needs no search: it is general knowledge, simple reasoning, or conversation. Do NOT call any search, retrieval, web, or file-reading tools. Reply naturally and concisely; for small talk, just respond conversationally. Do not cite sources or mention retrieval.`;
   }
   discoveryHint(sentence) {
     const names = this.tools.map((tool) => tool.name).filter((name) => name === JIKJI_FIND_TOOL_NAME || name === EVERYTHING_SEARCH_TOOL_NAME);
@@ -21933,55 +22051,55 @@ ${baseline}
     const scope = options.scope ? ` Restrict search to virtual path scope ${options.scope}.` : "";
     const firstAnswer = formatFirstAnswerContext(fastAnswer, fastAnswerDelivered);
     const answerRules = fastAnswerDelivered ? `Formatting and content rules for your final answer (DELTA ONLY):
-` + `- The user already has the first answer above. Your final answer MUST contain only the delta against it: (a) corrections to anything in the first answer that is wrong, unsupported, or outdated, and (b) newly verified findings that were not present in the first answer.
-` + `- NEVER restate or re-list first-answer facts that remain correct, and never repeat its bullet list.
-` + `- Mark each item clearly as a correction or as a new finding.
-` + `- If verification changed nothing and found nothing new, say so in one short line (the first answer is confirmed as-is) instead of restating it.
-` + `- ${ANSWER_DELIVERABLE_RULE} ${ANSWER_DELIVERABLE_DELTA_RULE}
-` + `- ${ANSWER_CITATION_RULE} A correction or new finding that relies on evidence the first answer cited must cite that evidence's id again.
-` + `- ${ANSWER_IMAGE_EMBED_RULE} ${ANSWER_IMAGE_DELTA_RULE}
-` + `- Do not report per-source negative findings (e.g. "no information found in Slack").
-` + `- When evidence conflicts, treat the freshest (most recent) information as the correct source of truth.` : `Formatting and content rules for the final answer (COMPLETE — no first answer reached the caller):
+- The user already has the first answer above. Your final answer MUST contain only the delta against it: (a) corrections to anything in the first answer that is wrong, unsupported, or outdated, and (b) newly verified findings that were not present in the first answer.
+- NEVER restate or re-list first-answer facts that remain correct, and never repeat its bullet list.
+- Mark each item clearly as a correction or as a new finding.
+- If verification changed nothing and found nothing new, say so in one short line (the first answer is confirmed as-is) instead of restating it.
+- ${ANSWER_DELIVERABLE_RULE} ${ANSWER_DELIVERABLE_DELTA_RULE}
+- ${ANSWER_CITATION_RULE} A correction or new finding that relies on evidence the first answer cited must cite that evidence's id again.
+- ${ANSWER_IMAGE_EMBED_RULE} ${ANSWER_IMAGE_DELTA_RULE}
+- Do not report per-source negative findings (e.g. "no information found in Slack").
+- When evidence conflicts, treat the freshest (most recent) information as the correct source of truth.` : `Formatting and content rules for the final answer (COMPLETE — no first answer reached the caller):
 ` + `- Unless the user explicitly requests a different format or length, provide the core answer to the user's question in at most 5 bullet points. If additional explanation is necessary, append it after the bullet points.
-` + `- Unless the user explicitly requests otherwise, answer the question directly. Do not include specific file paths, datasource descriptions, or retrieval mechanics in the answer text.
-` + `- ${ANSWER_DELIVERABLE_RULE}${fastAnswer !== undefined ? ` ${ANSWER_DELIVERABLE_DELTA_RULE}` : ""}
-` + `- ${ANSWER_CITATION_RULE}
-` + `- ${ANSWER_IMAGE_EMBED_RULE}
-` + `- Do not report per-source negative findings (e.g. "no information found in Slack").
-` + `- When evidence conflicts, treat the freshest (most recent) information as the correct source of truth.`;
+- Unless the user explicitly requests otherwise, answer the question directly. Do not include specific file paths, datasource descriptions, or retrieval mechanics in the answer text.
+- ${ANSWER_DELIVERABLE_RULE}${fastAnswer !== undefined ? ` ${ANSWER_DELIVERABLE_DELTA_RULE}` : ""}
+- ${ANSWER_CITATION_RULE}
+- ${ANSWER_IMAGE_EMBED_RULE}
+- Do not report per-source negative findings (e.g. "no information found in Slack").
+- When evidence conflicts, treat the freshest (most recent) information as the correct source of truth.`;
     return `Original query: ${query}${limit}${scope}
 
-` + `${firstAnswer}
+${firstAnswer}
 
 ` + formatSteerContext(steers) + `Now verify it rigorously. ${this.discoveryHint((tools) => `Actively use ${tools} when discovering or exploring local files and folders. `)}Check important claims against source files with bash when needed, correct anything wrong or unsupported, fill gaps with retrieval tools, and resolve conflicts and freshness. ${this.webFallbackHint()}
 
-` + `${answerRules}
+${answerRules}
 
-` + `Do not use broad grep/find or recursive filesystem scans: only inspect a path or narrow neighborhood surfaced by retrieval, and only when evidence clearly points there. ` + `Avoid spinning repeated near-identical queries against the same datasource; once additional attempts stop surfacing new evidence, conclude from the evidence available. ` + `If more search is needed, first write a brief 1–2 line progress update stating the best current hypothesis and what you are checking next, then call retrieval tools. ` + `When finished, reply with the final answer and no tool call: that last reply is what the user receives, so keep progress notes out of it.`;
+Do not use broad grep/find or recursive filesystem scans: only inspect a path or narrow neighborhood surfaced by retrieval, and only when evidence clearly points there. Avoid spinning repeated near-identical queries against the same datasource; once additional attempts stop surfacing new evidence, conclude from the evidence available. ` + `If more search is needed, first write a brief 1–2 line progress update stating the best current hypothesis and what you are checking next, then call retrieval tools. ` + `When finished, reply with the final answer and no tool call: that last reply is what the user receives, so keep progress notes out of it.`;
   }
   buildSearchPrompt(query, options, initialRetrievalContext) {
     const limit = typeof options.topK === "number" ? ` Cite at most ${options.topK} sources.` : "";
     const scope = options.scope ? ` Restrict search to virtual path scope ${options.scope}.` : "";
     return `Find and curate information for this original query: ${query}${limit}${scope}
 
-` + `Start by deciding whether this is answerable from general knowledge or memory. If it is a generic, stable question, answer it immediately without retrieval. ` + `Otherwise, baseline MinSync and Jikji retrieval is already running in parallel; do not give the final answer until its next message arrives.
+Start by deciding whether this is answerable from general knowledge or memory. If it is a generic, stable question, answer it immediately without retrieval. Otherwise, baseline MinSync and Jikji retrieval is already running in parallel; do not give the final answer until its next message arrives.
 
-` + `Baseline retrieval context:
+Baseline retrieval context:
 ${initialRetrievalContext ?? "Pending; continue only with a brief progress statement."}
 
-` + `Treat candidates as unverified evidence, verify important claims against source files when needed, and use additional tools when needed. ` + `Judge relevance, conflicts, freshness, and sufficiency in this agent loop.
+Treat candidates as unverified evidence, verify important claims against source files when needed, and use additional tools when needed. Judge relevance, conflicts, freshness, and sufficiency in this agent loop.
 
-` + `Formatting and content rules for the answer:
-` + `- Unless the user explicitly requests a different format or length, provide the core answer to the user's question in at most 5 bullet points. If additional explanation is necessary, append it after the bullet points.
-` + `- Unless the user explicitly requests otherwise, answer the question directly. Do not include specific file paths, datasource descriptions, or retrieval mechanics in the answer text.
-` + `- ${ANSWER_DELIVERABLE_RULE}
-` + `- ${ANSWER_CITATION_RULE}
-` + `- ${ANSWER_IMAGE_EMBED_RULE}
-` + `- Do not report per-source negative findings (e.g. "no information found in Slack").
-` + `- When evidence conflicts, treat the freshest (most recent) information as the correct source of truth.
-` + `- If information is incomplete or uncertain, acknowledge it briefly without lengthy explanations. If there are partial clues or leads, mention them concisely.
+Formatting and content rules for the answer:
+- Unless the user explicitly requests a different format or length, provide the core answer to the user's question in at most 5 bullet points. If additional explanation is necessary, append it after the bullet points.
+- Unless the user explicitly requests otherwise, answer the question directly. Do not include specific file paths, datasource descriptions, or retrieval mechanics in the answer text.
+- ${ANSWER_DELIVERABLE_RULE}
+- ${ANSWER_CITATION_RULE}
+- ${ANSWER_IMAGE_EMBED_RULE}
+- Do not report per-source negative findings (e.g. "no information found in Slack").
+- When evidence conflicts, treat the freshest (most recent) information as the correct source of truth.
+- If information is incomplete or uncertain, acknowledge it briefly without lengthy explanations. If there are partial clues or leads, mention them concisely.
 
-` + `${this.discoveryHint((tools) => `When exploring local files and folders, actively use ${tools} rather than exploratory bash commands. `)}If more search is needed, first write a brief 1–2 line progress update stating the best current hypothesis and what you are checking next, then call retrieval tools. ` + `Never repeat a generic status message. Do not use broad grep/find or recursive filesystem scans: only inspect a path or narrow neighborhood surfaced by retrieval, and only when evidence clearly points there. ` + `Avoid spinning repeated near-identical queries against the same datasource; once additional attempts stop surfacing new evidence, conclude from the evidence available. ` + `When finished, reply with the final answer and no tool call: that last reply is what the user receives, so keep progress notes out of it.`;
+${this.discoveryHint((tools) => `When exploring local files and folders, actively use ${tools} rather than exploratory bash commands. `)}If more search is needed, first write a brief 1–2 line progress update stating the best current hypothesis and what you are checking next, then call retrieval tools. ` + `Never repeat a generic status message. Do not use broad grep/find or recursive filesystem scans: only inspect a path or narrow neighborhood surfaced by retrieval, and only when evidence clearly points there. Avoid spinning repeated near-identical queries against the same datasource; once additional attempts stop surfacing new evidence, conclude from the evidence available. When finished, reply with the final answer and no tool call: that last reply is what the user receives, so keep progress notes out of it.`;
   }
   async refresh(force = false, opts) {
     const signal = opts?.signal;
@@ -22217,7 +22335,7 @@ ${initialRetrievalContext ?? "Pending; continue only with a brief progress state
       });
     }
     const state = this.refreshState.inFlight || persistedRunning && ownerAlive ? "indexing" : interrupted ? "failed" : persistedProgress?.state === "success" || persistedProgress?.state === "failed" ? persistedProgress.state : this.refreshState.lastOutcome === "never" ? "idle" : this.refreshState.lastOutcome;
-    const parsedMirrorReady = this.refreshState.lastOutcome === "success" || persistedProgress?.state === "success" || existsSync22(refreshReadinessPath(this.workspaceProjectRoot));
+    const parsedMirrorReady = this.refreshState.lastOutcome === "success" || persistedProgress?.state === "success" || existsSync23(refreshReadinessPath(this.workspaceProjectRoot));
     const effectiveProgress = this.refreshState.inFlight || persistedRunning ? persistedProgress : undefined;
     return {
       state,
@@ -22677,7 +22795,7 @@ function buildSteerMessage(text) {
 
 ${text}
 
-` + `Treat it as direction for the search in progress, not as a new question: adjust which sources, queries, or claims you pursue, then continue. ` + `When finished, reply with the final answer to the original question, updated for this correction, and no tool call.`;
+Treat it as direction for the search in progress, not as a new question: adjust which sources, queries, or claims you pursue, then continue. When finished, reply with the final answer to the original question, updated for this correction, and no tool call.`;
 }
 function frameSteers(messages, steerTexts) {
   if (steerTexts.size === 0)
@@ -22697,12 +22815,12 @@ function formatSteerContext(steers) {
   return `While the search ran, the user corrected it:
 ${steers.map((steer) => `- ${steer}`).join(`
 `)}
-` + `The first answer predates these corrections; follow them over it.
+The first answer predates these corrections; follow them over it.
 
 `;
 }
 function buildSteerContinuationPrompt(steers) {
-  return `${formatSteerContext(steers)}Apply the latest correction: search or verify further if it needs that, ` + `then reply with the complete final answer to the original question and no tool call.`;
+  return `${formatSteerContext(steers)}Apply the latest correction: search or verify further if it needs that, then reply with the complete final answer to the original question and no tool call.`;
 }
 function formatFirstAnswerContext(fastAnswer, delivered) {
   if (fastAnswer === undefined) {
@@ -22825,11 +22943,20 @@ function fsearchRefreshDiagnostics(fsearch) {
 function minSyncRefreshDiagnostics(minsync) {
   if (!minsync)
     return [];
+  const diagnostics = [];
   if (minsync.diagnostic)
-    return [toMinSyncDiagnostic(minsync.diagnostic, minsync.ok)];
-  if (!minsync.ok && minsync.reason)
-    return [toMinSyncReasonDiagnostic(minsync.reason)];
-  return [];
+    diagnostics.push(toMinSyncDiagnostic(minsync.diagnostic, minsync.ok));
+  else if (!minsync.ok && minsync.reason)
+    diagnostics.push(toMinSyncReasonDiagnostic(minsync.reason));
+  if (minsync.lexicalFallback) {
+    diagnostics.push({
+      code: "lexical-tokenizer-unavailable",
+      severity: "warning",
+      message: `The "${minsync.lexicalFallback.requested}" BM25 tokenizer needs the Kiwi native library and model, which were not found, so MinSync tokenizes on whitespace and a Korean word with a particle (인증서를) does not match its stem (인증서). Install Kiwi (set KIWI_LIBRARY_PATH and KIWI_MODEL_PATH, or install it under ~/.local/kiwi), then refresh again.`,
+      source: "minsync"
+    });
+  }
+  return diagnostics;
 }
 function toMinSyncDiagnostic(diag, ok) {
   const code = diag.code === "embedder-unavailable" ? "embedder-unavailable" : diag.code === "embedding-identity-mismatch" ? "embedding-identity-mismatch" : diag.code === "minsync-lock-held" ? "minsync-lock-held" : "minsync-sync-failed";
@@ -23219,7 +23346,7 @@ var init_connector = () => {};
 
 // src/datasource/sidecar/config.ts
 import { readFileSync as readFileSync19 } from "node:fs";
-import { homedir as homedir3 } from "node:os";
+import { homedir as homedir4 } from "node:os";
 import { dirname as dirname18, isAbsolute as isAbsolute8, join as join32, resolve as resolve14, sep as sep7 } from "node:path";
 import { parse as parseToml2 } from "smol-toml";
 function resolveSidecarOptions(input) {
@@ -23259,9 +23386,9 @@ function archivePathFromCrawlerConfig(input) {
 }
 function expandHome(path) {
   if (path === "~")
-    return homedir3();
+    return homedir4();
   const prefixed = path.startsWith("~/") || sep7 === "\\" && path.startsWith("~\\");
-  return prefixed ? join32(homedir3(), path.slice(2)) : path;
+  return prefixed ? join32(homedir4(), path.slice(2)) : path;
 }
 var init_config2 = __esm(() => {
   init_connector();
@@ -23352,7 +23479,7 @@ var init_descriptors = __esm(() => {
 
 // src/datasource/skills/clawgallery/client.ts
 import { spawn as spawn10 } from "node:child_process";
-import { existsSync as existsSync23 } from "node:fs";
+import { existsSync as existsSync24 } from "node:fs";
 
 class ClawGalleryClient {
   options;
@@ -23360,7 +23487,7 @@ class ClawGalleryClient {
     this.options = options;
   }
   async bootstrap(signal) {
-    if (this.options.path !== undefined && !existsSync23(this.options.path)) {
+    if (this.options.path !== undefined && !existsSync24(this.options.path)) {
       return {
         ok: false,
         reason: "nonzero-exit",
@@ -24278,7 +24405,7 @@ var init_rclone_connector = __esm(() => {
 });
 
 // src/datasource/chunk-store.ts
-import { existsSync as existsSync24, mkdirSync as mkdirSync18, readFileSync as readFileSync21, writeFileSync as writeFileSync17 } from "node:fs";
+import { existsSync as existsSync25, mkdirSync as mkdirSync18, readFileSync as readFileSync21, writeFileSync as writeFileSync17 } from "node:fs";
 import { dirname as dirname20, join as join34 } from "node:path";
 function datasourceChunkStorePath(workspaceRoot, skillName, instanceId) {
   return join34(workspaceRoot, ".autorag", "datasources", sanitizeIdSegment(skillName), sanitizeIdSegment(instanceId), "chunks.json");
@@ -24318,7 +24445,7 @@ class DatasourceChunkStore {
     if (this.storePath === undefined)
       return false;
     try {
-      if (!existsSync24(this.storePath))
+      if (!existsSync25(this.storePath))
         return false;
       const parsed = JSON.parse(readFileSync21(this.storePath, "utf8"));
       if (parsed.version !== STORE_VERSION || !Array.isArray(parsed.chunks))
@@ -25845,7 +25972,7 @@ var init_github = __esm(() => {
 
 // src/datasource/skills/github-gist/connector.ts
 import { execFile as execFile2 } from "node:child_process";
-import { existsSync as existsSync25, mkdirSync as mkdirSync20, readFileSync as readFileSync23, writeFileSync as writeFileSync19 } from "node:fs";
+import { existsSync as existsSync26, mkdirSync as mkdirSync20, readFileSync as readFileSync23, writeFileSync as writeFileSync19 } from "node:fs";
 import { dirname as dirname22 } from "node:path";
 async function defaultGhTokenRunner() {
   return new Promise((resolve) => {
@@ -25965,7 +26092,7 @@ class GitHubGistConnector {
   }
   loadState() {
     const statePath = this.options.statePath;
-    if (statePath === undefined || !existsSync25(statePath))
+    if (statePath === undefined || !existsSync26(statePath))
       return { version: STATE_VERSION, gists: {} };
     try {
       const parsed = JSON.parse(readFileSync23(statePath, "utf8"));
@@ -26028,7 +26155,7 @@ var init_connector3 = () => {};
 
 // src/datasource/skills/github-gist/semantic.ts
 import { createHash as createHash13 } from "node:crypto";
-import { existsSync as existsSync26, mkdirSync as mkdirSync21, readFileSync as readFileSync24, writeFileSync as writeFileSync20 } from "node:fs";
+import { existsSync as existsSync27, mkdirSync as mkdirSync21, readFileSync as readFileSync24, writeFileSync as writeFileSync20 } from "node:fs";
 import { dirname as dirname23 } from "node:path";
 function contentHash(chunk) {
   return createHash13("sha256").update(`${chunk.title ?? ""}
@@ -26119,7 +26246,7 @@ ${chunk.content}`));
     if (this.loaded)
       return;
     this.loaded = true;
-    if (this.statePath === undefined || !existsSync26(this.statePath))
+    if (this.statePath === undefined || !existsSync27(this.statePath))
       return;
     try {
       const parsed = JSON.parse(readFileSync24(this.statePath, "utf8"));
@@ -26982,7 +27109,7 @@ var init_types5 = __esm(() => {
 
 // src/datasource/skills/lazykatok/client.ts
 import { spawn as spawn14 } from "node:child_process";
-import { existsSync as existsSync27 } from "node:fs";
+import { existsSync as existsSync28 } from "node:fs";
 import { delimiter as delimiter4, join as join36 } from "node:path";
 
 class LazykatokClient {
@@ -27172,7 +27299,7 @@ function lookupInPath2(executable, env) {
     if (dir.length === 0)
       continue;
     const candidate = join36(dir, executable);
-    if (existsSync27(candidate))
+    if (existsSync28(candidate))
       return candidate;
   }
   return;
@@ -30925,7 +31052,7 @@ var init_wacrawl = __esm(() => {
 });
 
 // src/datasource/skills/factory.ts
-import { homedir as homedir4 } from "node:os";
+import { homedir as homedir5 } from "node:os";
 import { join as join39 } from "node:path";
 function chatCrawlerArgs(descriptor, defaultCrawlerDb, config, workspaceRoot, registrationName) {
   const sidecar = resolveSidecarOptions({
@@ -31029,9 +31156,9 @@ var init_factory = __esm(() => {
   init_telecrawl();
   init_wacrawl();
   BUILDERS = {
-    telegram: (config, workspaceRoot, registrationName) => new TelecrawlSkill(chatCrawlerArgs(CRAWLER_DESCRIPTORS.telegram, join39(homedir4(), ".telecrawl", "telecrawl.db"), config, workspaceRoot, registrationName)),
-    whatsapp: (config, workspaceRoot, registrationName) => new WacrawlSkill(chatCrawlerArgs(CRAWLER_DESCRIPTORS.whatsapp, join39(homedir4(), ".wacrawl", "wacrawl.db"), config, workspaceRoot, registrationName)),
-    slack: (config, workspaceRoot, registrationName) => new SlackSkill(chatCrawlerArgs(CRAWLER_DESCRIPTORS.slack, join39(homedir4(), ".slacrawl", "slacrawl.db"), config, workspaceRoot, registrationName)),
+    telegram: (config, workspaceRoot, registrationName) => new TelecrawlSkill(chatCrawlerArgs(CRAWLER_DESCRIPTORS.telegram, join39(homedir5(), ".telecrawl", "telecrawl.db"), config, workspaceRoot, registrationName)),
+    whatsapp: (config, workspaceRoot, registrationName) => new WacrawlSkill(chatCrawlerArgs(CRAWLER_DESCRIPTORS.whatsapp, join39(homedir5(), ".wacrawl", "wacrawl.db"), config, workspaceRoot, registrationName)),
+    slack: (config, workspaceRoot, registrationName) => new SlackSkill(chatCrawlerArgs(CRAWLER_DESCRIPTORS.slack, join39(homedir5(), ".slacrawl", "slacrawl.db"), config, workspaceRoot, registrationName)),
     discord: (config, workspaceRoot, registrationName, embeddingRuntime) => {
       const connector = config.connector ?? {};
       const clientOptions = {
@@ -31089,7 +31216,7 @@ var init_factory = __esm(() => {
     }),
     kakao: (config, workspaceRoot, registrationName) => {
       const connector = config.connector ?? {};
-      const dataDir = typeof connector.workspacePath === "string" && connector.workspacePath.length > 0 ? connector.workspacePath : join39(homedir4(), "Library", "Application Support", "katok");
+      const dataDir = typeof connector.workspacePath === "string" && connector.workspacePath.length > 0 ? connector.workspacePath : join39(homedir5(), "Library", "Application Support", "katok");
       const sidecar = resolveSidecarOptions({
         descriptor: CRAWLER_DESCRIPTORS.kakao,
         connector: config.connector,
@@ -31113,11 +31240,16 @@ var init_factory = __esm(() => {
       skillName: registrationName,
       connectorOptions: config.connector
     }),
-    "github-gist": (config, workspaceRoot, registrationName) => new GitHubGistSkill({
-      ...common(config, workspaceRoot),
-      skillName: registrationName,
-      connectorOptions: config.connector
-    }),
+    "github-gist": (config, workspaceRoot, registrationName) => {
+      const { semantic, ...connectorOptions } = config.connector ?? {};
+      const semanticOff = semantic === false || typeof semantic === "object" && semantic !== null && semantic.enabled === false;
+      return new GitHubGistSkill({
+        ...common(config, workspaceRoot),
+        skillName: registrationName,
+        connectorOptions,
+        ...semanticOff ? { semantic: { enabled: false } } : {}
+      });
+    },
     "cloud-drive": (config, workspaceRoot, registrationName) => new CloudDriveSkill({
       ...common(config, workspaceRoot),
       skillName: registrationName,
@@ -31163,7 +31295,7 @@ var init_factory = __esm(() => {
 
 // src/cli/config.ts
 import { randomUUID as randomUUID8 } from "node:crypto";
-import { existsSync as existsSync28, mkdirSync as mkdirSync25, readFileSync as readFileSync25, renameSync as renameSync10, unlinkSync as unlinkSync5, writeFileSync as writeFileSync22 } from "node:fs";
+import { existsSync as existsSync29, mkdirSync as mkdirSync25, readFileSync as readFileSync25, renameSync as renameSync10, unlinkSync as unlinkSync5, writeFileSync as writeFileSync22 } from "node:fs";
 import { basename as basename14, dirname as dirname26, isAbsolute as isAbsolute10, join as join40, resolve as resolve16 } from "node:path";
 import { findEnvKeys, getEnvApiKey } from "@earendil-works/pi-ai/compat";
 import { getAgentDir as getAgentDir4, ModelRuntime as ModelRuntime2, SettingsManager as SettingsManager2 } from "@earendil-works/pi-coding-agent";
@@ -31188,7 +31320,7 @@ function resolveConfigPath(input) {
 function readConfigFile(configPath, explicit) {
   let exists;
   try {
-    exists = existsSync28(configPath);
+    exists = existsSync29(configPath);
   } catch {
     exists = false;
   }
@@ -31248,7 +31380,7 @@ function replaceFileAtomically(path, contents, assertCommitAllowed) {
   }
 }
 function migrateLegacyConfig(configPath, legacyPath) {
-  if (existsSync28(configPath) || !existsSync28(legacyPath))
+  if (existsSync29(configPath) || !existsSync29(legacyPath))
     return;
   const legacy = readConfigFile(legacyPath, true);
   const legacyBytes = readFileSync25(legacyPath);
@@ -32519,7 +32651,7 @@ function writeDefaultConfig(path, partial, opts = {}) {
 `;
   const lock = acquireConfigWriteLock(path);
   try {
-    const exists = existsSync28(path);
+    const exists = existsSync29(path);
     if (!opts.force && exists) {
       throw new ConfigError(`Config file already exists: ${path}`);
     }