@autorag/librarian 2.10.2 → 2.10.4

package/dist/index.js

@@ -2109,6 +2109,25 @@ function errorMessage(error) {
 var MANIFEST_FILENAME = "manifest.json";
 var init_bundle = () => {};
 
+// src/filesystem/path-exclusions.ts
+import { posix as posix2, win32 as win322 } from "node:path";
+function createPathExclusionMatcher(excludedPaths, platform = process.platform) {
+  const paths = platform === "win32" ? win322 : posix2;
+  const normalize = (path) => {
+    const absolute = paths.resolve(path);
+    return platform === "win32" ? absolute.toLowerCase() : absolute;
+  };
+  const excluded = excludedPaths.map((path) => {
+    const absolute = normalize(path);
+    return { absolute, prefix: absolute.endsWith(paths.sep) ? absolute : `${absolute}${paths.sep}` };
+  });
+  return (path) => {
+    const absolute = normalize(path);
+    return excluded.some((entry) => absolute === entry.absolute || absolute.startsWith(entry.prefix));
+  };
+}
+var init_path_exclusions = () => {};
+
 // src/everything/client.ts
 import { spawn as spawn2 } from "node:child_process";
 import { createHash as createHash4 } from "node:crypto";
@@ -2226,6 +2245,7 @@ class EverythingClient {
   run;
   launch;
   stateDir;
+  isExcludedPath;
   binaries;
   operation = Promise.resolve();
   constructor(options) {
@@ -2235,6 +2255,7 @@ class EverythingClient {
     this.run = options.run ?? runProcess;
     this.launch = options.launch ?? launchDetached;
     this.stateDir = join6(options.root, ".autorag", "everything");
+    this.isExcludedPath = createPathExclusionMatcher(options.excludeFolders ?? [], "win32");
   }
   isSupported() {
     return this.platform === "win32";
@@ -2254,6 +2275,9 @@ class EverythingClient {
       }
       const es = { ok: true, esPath: binaries.esPath };
       const timeoutMs = this.options.timeoutMs ?? DEFAULT_TIMEOUT_MS2;
+      if ((this.options.excludeFolders?.length ?? 0) > 0) {
+        return this.searchWithExclusions(es.esPath, request, timeoutMs);
+      }
       const args = buildEverythingSearchArgs(this.instanceName, request, timeoutMs);
       const result = await this.run(es.esPath, args, timeoutMs + 5000);
       if (result.code !== 0)
@@ -2265,6 +2289,50 @@ class EverythingClient {
       }
     });
   }
+  async searchWithExclusions(esPath, request, timeoutMs) {
+    const deadline = Date.now() + timeoutMs;
+    const offset = request.offset ?? 0;
+    const maxResults = request.maxResults ?? DEFAULT_MAX_RESULTS;
+    const results = [];
+    let skipped = 0;
+    let nativeOffset = 0;
+    const timedOut = () => ({
+      ok: false,
+      reason: "search-failed",
+      message: `Everything search exceeded ${timeoutMs}ms while filtering excluded paths; narrow the query or increase the Everything timeout.`
+    });
+    for (;; ) {
+      const remainingMs = deadline - Date.now();
+      if (remainingMs <= 0)
+        return timedOut();
+      const args = buildEverythingSearchArgs(this.instanceName, { ...request, offset: nativeOffset, maxResults: MAX_FILTER_CANDIDATES }, remainingMs);
+      const response = await this.run(esPath, args, remainingMs + 5000);
+      if (response.code !== 0)
+        return this.failure("search-failed", esPath, args, response);
+      if (Date.now() >= deadline)
+        return timedOut();
+      let entries;
+      try {
+        entries = parseEverythingJson(response.stdout);
+      } catch (error) {
+        return { ok: false, reason: "search-failed", message: error.message };
+      }
+      for (const entry of entries) {
+        if (this.isExcludedPath(entry.path))
+          continue;
+        if (skipped < offset) {
+          skipped += 1;
+          continue;
+        }
+        results.push(entry);
+        if (results.length >= maxResults)
+          return { ok: true, results };
+      }
+      if (entries.length < MAX_FILTER_CANDIDATES)
+        return { ok: true, results };
+      nativeOffset += entries.length;
+    }
+  }
   index() {
     return this.serialize(async () => {
       const es = await this.ensureRunning(true);
@@ -2338,7 +2406,8 @@ class EverythingClient {
       excludeFolders: [
         ...new Set([
           join6(this.options.root, ".autorag"),
-          ...this.options.folders.map((folder) => join6(folder, ".autorag"))
+          ...this.options.folders.map((folder) => join6(folder, ".autorag")),
+          ...this.options.excludeFolders ?? []
         ])
       ]
     }));
@@ -2432,8 +2501,9 @@ function launchDetached(command, args) {
   child.on("error", () => {});
   child.unref();
 }
-var DEFAULT_TIMEOUT_MS2 = 30000, DEFAULT_STARTUP_TIMEOUT_MS = 20000, DEFAULT_INDEX_TIMEOUT_MS, DEFAULT_POLL_INTERVAL_MS = 250, DEFAULT_MAX_RESULTS = 100, MAX_OUTPUT_BYTES, VERSION_PATTERN;
+var DEFAULT_TIMEOUT_MS2 = 30000, DEFAULT_STARTUP_TIMEOUT_MS = 20000, DEFAULT_INDEX_TIMEOUT_MS, DEFAULT_POLL_INTERVAL_MS = 250, DEFAULT_MAX_RESULTS = 100, MAX_FILTER_CANDIDATES = 1000, MAX_OUTPUT_BYTES, VERSION_PATTERN;
 var init_client = __esm(() => {
+  init_path_exclusions();
   init_bundle();
   DEFAULT_INDEX_TIMEOUT_MS = 10 * 60000;
   MAX_OUTPUT_BYTES = 8 * 1024 * 1024;
@@ -2453,6 +2523,7 @@ function compare(a, b) {
   return a < b ? -1 : a > b ? 1 : 0;
 }
 async function walkFileSearch(folders, request, options = {}) {
+  const isExcluded = createPathExclusionMatcher(options.excludeFolders ?? []);
   const maxVisited = options.maxVisited ?? DEFAULT_MAX_VISITED;
   const maxMatches = options.maxMatches ?? DEFAULT_MAX_MATCHES;
   const deadline = Date.now() + (options.deadlineMs ?? DEFAULT_DEADLINE_MS);
@@ -2475,6 +2546,8 @@ async function walkFileSearch(folders, request, options = {}) {
   outer:
     while (pending.length > 0) {
       const dir = pending.pop();
+      if (isExcluded(dir))
+        continue;
       if (Date.now() >= deadline) {
         truncated = true;
         break;
@@ -2492,6 +2565,8 @@ async function walkFileSearch(folders, request, options = {}) {
         }
         visited += 1;
         const path = join7(dir, dirent.name);
+        if (isExcluded(path))
+          continue;
         if (dirent.isDirectory()) {
           if (dirent.name === ".autorag")
             continue;
@@ -2526,6 +2601,8 @@ async function walkFileSearch(folders, request, options = {}) {
       }
     }
   for (const folder of folders) {
+    if (isExcluded(folder))
+      continue;
     if (!underPath(folder) || !matches(folder))
       continue;
     if (request.kind === "files")
@@ -2552,6 +2629,7 @@ async function walkFileSearch(folders, request, options = {}) {
 }
 var DEFAULT_MAX_VISITED = 500000, DEFAULT_DEADLINE_MS = 20000, DEFAULT_MAX_MATCHES = 1e4, SORT_ORDERS;
 var init_walk = __esm(() => {
+  init_path_exclusions();
   SORT_ORDERS = {
     "name-ascending": (a, b) => compare(a.name, b.name),
     "name-descending": (a, b) => compare(b.name, a.name),
@@ -2661,6 +2739,7 @@ class FSearchClient {
   launch;
   isAlive;
   kill;
+  isExcludedPath;
   stateDir;
   dbPath;
   pidPath;
@@ -2673,6 +2752,7 @@ class FSearchClient {
     this.launch = options.launch ?? launchDetached2;
     this.isAlive = options.isProcessAlive ?? defaultIsProcessAlive;
     this.kill = options.killProcess ?? ((pid) => process.kill(pid, "SIGTERM"));
+    this.isExcludedPath = createPathExclusionMatcher(options.excludeFolders ?? []);
     this.stateDir = join8(options.root, ".autorag", "fsearch");
     this.dbPath = join8(this.stateDir, "fsearch.db");
     this.pidPath = join8(this.stateDir, "watch.pid");
@@ -2690,6 +2770,7 @@ class FSearchClient {
       if (unavailable !== undefined || !binary.ok) {
         try {
           const walked = await walkFileSearch(this.options.folders, request, {
+            excludeFolders: this.options.excludeFolders,
             maxVisited: this.options.walkMaxVisited,
             deadlineMs: this.options.walkDeadlineMs
           });
@@ -2704,7 +2785,9 @@ class FSearchClient {
           return { ok: false, reason: "search-failed", message: error.message };
         }
       }
-      const args = buildFSearchSearchArgs(this.dbPath, this.socketPath(), request);
+      const hasExclusions = (this.options.excludeFolders?.length ?? 0) > 0;
+      const nativeRequest = hasExclusions ? { ...request, offset: 0, maxResults: MAX_RESULTS_CAP } : request;
+      const args = buildFSearchSearchArgs(this.dbPath, this.socketPath(), nativeRequest);
       const result = await this.run(binary.path, args, this.options.timeoutMs ?? DEFAULT_TIMEOUT_MS3);
       if (result.code !== 0)
         return this.failure("search-failed", binary.path, args, result);
@@ -2715,6 +2798,10 @@ class FSearchClient {
         return { ok: false, reason: "search-failed", message: error.message };
       }
       let entries = parsed.entries;
+      if (hasExclusions) {
+        const isExcluded = this.isExcludedPath;
+        entries = entries.filter((entry) => !isExcluded(entry.path));
+      }
       if (request.path !== undefined) {
         const prefix = request.path.endsWith(sep3) ? request.path : `${request.path}${sep3}`;
         entries = entries.filter((entry) => entry.path === request.path || entry.path.startsWith(prefix));
@@ -2724,7 +2811,16 @@ class FSearchClient {
       const maxResults = Math.min(request.maxResults ?? DEFAULT_MAX_RESULTS2, MAX_RESULTS_CAP);
       if (entries.length > maxResults)
         entries = entries.slice(0, maxResults);
-      return { ok: true, backend: "fsearch-cli", results: entries, total: parsed.total };
+      return {
+        ok: true,
+        backend: "fsearch-cli",
+        results: entries,
+        total: parsed.total,
+        ...hasExclusions && parsed.total !== undefined && parsed.total > parsed.entries.length && entries.length < maxResults ? {
+          truncated: true,
+          note: `FSearch checked ${parsed.entries.length} of ${parsed.total} native matches before exclusions (candidate limit ${MAX_RESULTS_CAP}); results may be incomplete. Narrow the query.`
+        } : {}
+      };
     });
   }
   index() {
@@ -2918,6 +3014,7 @@ function defaultIsProcessAlive(pid) {
 }
 var DEFAULT_TIMEOUT_MS3 = 30000, DEFAULT_STARTUP_TIMEOUT_MS2 = 1e4, STOP_TIMEOUT_MS = 2000, DEFAULT_INDEX_TIMEOUT_MS2, DEFAULT_POLL_INTERVAL_MS2 = 250, DEFAULT_MAX_RESULTS2 = 100, MAX_RESULTS_CAP = 1000, MAX_OUTPUT_BYTES2, SORT_KEYS;
 var init_client2 = __esm(() => {
+  init_path_exclusions();
   init_walk();
   DEFAULT_INDEX_TIMEOUT_MS2 = 10 * 60000;
   MAX_OUTPUT_BYTES2 = 8 * 1024 * 1024;
@@ -5513,7 +5610,8 @@ class MinSyncClient {
       restoreConfig();
       return this.lockHeldResult(result.stderr);
     }
-    if (!result.ok) {
+    const partial = !result.ok && result.code === MINSYNC_PARTIAL_SYNC_EXIT_CODE ? readPartialSync(result.stdout) : undefined;
+    if (!result.ok && partial === undefined) {
       restoreConfig();
       return {
         ok: false,
@@ -5533,10 +5631,12 @@ class MinSyncClient {
     }
     if (effective.identity)
       this.writeIdentity(effective.identity);
+    const filesFailed = partial?.filesFailed ?? [];
     return {
       ok: true,
       synced: readSyncedCount(result.stdout),
       workspacePath: this.workspacePath,
+      ...filesFailed.length > 0 ? { filesFailed } : {},
       ...identityChanged ? {
         diagnostic: {
           code: "embedding-identity-mismatch",
@@ -5568,6 +5668,19 @@ function readConfigSnapshot(configPath) {
     throw error;
   }
 }
+function readPartialSync(stdout) {
+  const parsed = parseJson(stdout);
+  if (!isRecord4(parsed) || parsed.partial !== true)
+    return;
+  const failures = Array.isArray(parsed.files_failed) ? parsed.files_failed : [];
+  return {
+    filesFailed: failures.filter(isRecord4).map((failure) => ({
+      path: typeof failure.path === "string" ? failure.path : "(unknown)",
+      stage: typeof failure.stage === "string" ? failure.stage : "unknown",
+      error: typeof failure.error === "string" ? failure.error : "unknown error"
+    }))
+  };
+}
 function readSyncedCount(stdout) {
   const parsed = parseJson(stdout);
   if (!isRecord4(parsed))
@@ -5623,7 +5736,7 @@ function isIdentity(value) {
 function isRecord4(value) {
   return typeof value === "object" && value !== null;
 }
-var MINSYNC_OLLAMA_MIGRATION_MESSAGE = "This workspace uses the legacy 768-dimensional Ollama/TEI embedding path. Reindex explicitly, or pin an explicit Ollama/TEI embedder config before using the new default runtime.", API_KEY_ENV_PATTERN, MINSYNC_LOCK_HELD_EXIT_CODE = 3, MinSyncQueryError;
+var MINSYNC_OLLAMA_MIGRATION_MESSAGE = "This workspace uses the legacy 768-dimensional Ollama/TEI embedding path. Reindex explicitly, or pin an explicit Ollama/TEI embedder config before using the new default runtime.", API_KEY_ENV_PATTERN, MINSYNC_LOCK_HELD_EXIT_CODE = 3, MINSYNC_PARTIAL_SYNC_EXIT_CODE = 6, MinSyncQueryError;
 var init_client4 = __esm(() => {
   init_embedder_config();
   init_errors();
@@ -5867,7 +5980,7 @@ 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 = "0.4.7", 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 = "0.4.8", MINSYNC_CARGO_INSTALL_TIMEOUT_MS, CARGO_PROBE_TIMEOUT_MS = 5000, MinSyncReleaseError;
 var init_installer2 = __esm(() => {
   init_process();
   SHA256_HEX_PATTERN = /^[a-f0-9]{64}$/;
@@ -9170,12 +9283,43 @@ function digitsEndAt(text, from) {
     end += 1;
   return end;
 }
-function continuesAsProse(answer, from) {
-  if (answer[from] !== " " && answer[from] !== " ")
+function continuesAsProse(answer, openIndex, from) {
+  let cursor = from;
+  let leadingEmphasis = 0;
+  while (openIndex - 1 - leadingEmphasis >= 0 && (answer[openIndex - 1 - leadingEmphasis] === "*" || answer[openIndex - 1 - leadingEmphasis] === "_")) {
+    leadingEmphasis += 1;
+  }
+  if (leadingEmphasis > 0) {
+    let trailingEmphasis = 0;
+    while (cursor + trailingEmphasis < answer.length && trailingEmphasis < leadingEmphasis && answer[cursor + trailingEmphasis] === answer[openIndex - 1 - trailingEmphasis]) {
+      trailingEmphasis += 1;
+    }
+    if (trailingEmphasis === leadingEmphasis)
+      cursor += trailingEmphasis;
+  }
+  const tokenStart = openIndex - leadingEmphasis;
+  const linePrefix = answer.slice(answer.lastIndexOf(`
+`, tokenStart - 1) + 1, tokenStart);
+  const atHeadingStart = /^[ \t>*-]*$/u.test(linePrefix);
+  if (atHeadingStart && answer[cursor] === ":")
+    cursor += 1;
+  if (answer[cursor] !== " " && answer[cursor] !== " ")
     return false;
-  const next = answer.codePointAt(from + 1);
+  cursor += 1;
+  while (cursor < answer.length && (answer[cursor] === "*" || answer[cursor] === "_" || answer[cursor] === '"' || answer[cursor] === "'" || answer[cursor] === "“" || answer[cursor] === "‘")) {
+    cursor += 1;
+  }
+  const next = answer.codePointAt(cursor);
   return next !== undefined && /[\p{L}\p{N}]/u.test(String.fromCodePoint(next));
 }
+function isLegalPinpoint(answer, openIndex, afterCloseIndex) {
+  if (PINPOINT_RANGE_AHEAD.test(answer.slice(afterCloseIndex, afterCloseIndex + 16)))
+    return true;
+  if (PINPOINT_RANGE_BEHIND.test(answer.slice(Math.max(0, openIndex - 16), openIndex)))
+    return true;
+  const prefixWindow = answer.slice(Math.max(0, openIndex - 120), openIndex);
+  return PARA_PINPOINT_PREFIX.test(prefixWindow) || CLAUSE_START_AT_PREFIX.test(prefixWindow) || CASE_CITATION_AT_PREFIX.test(prefixWindow);
+}
 function evidenceMarkerAt(answer, index) {
   const ids = [];
   let cursor = index + 1;
@@ -9241,7 +9385,7 @@ function answerMarkers(answer, isFile = () => false) {
     }
     if (char === "[") {
       const digitsEnd = digitsEndAt(answer, index + 1);
-      if (digitsEnd > index + 1 && answer[digitsEnd] === "]" && answer[digitsEnd + 1] !== "(" && !continuesAsProse(answer, digitsEnd + 1)) {
+      if (digitsEnd > index + 1 && answer[digitsEnd] === "]" && answer[digitsEnd + 1] !== "(" && !continuesAsProse(answer, index, digitsEnd + 1) && !isLegalPinpoint(answer, index, digitsEnd + 1)) {
         markers.push({
           kind: "number",
           start: leadingStart(index),
@@ -9313,7 +9457,14 @@ function assertResultsMappingOneToOne(label, results, mapping) {
   ].filter((note) => note !== undefined);
   throw new Error(`${label}: result numbers and mapping numbers must be one-to-one, but results contain ${formatCitationList(resultNumbers)} ` + `and mapping contains ${formatCitationList(mappingNumbers)}` + (duplicateNotes.length > 0 ? ` (${duplicateNotes.join("; ")})` : "") + ". Give every result a unique number with exactly one mapping entry carrying the same number and no mapping entry without a result.");
 }
-var MAX_FILE_MARKER_CLOSERS = 8, FILE_MARKER_PREFIX = "[file:";
+var MAX_FILE_MARKER_CLOSERS = 8, FILE_MARKER_PREFIX = "[file:", PARA_PINPOINT_PREFIX, CLAUSE_START_AT_PREFIX, CASE_CITATION_AT_PREFIX, PINPOINT_RANGE_AHEAD, PINPOINT_RANGE_BEHIND;
+var init_citations = __esm(() => {
+  PARA_PINPOINT_PREFIX = /(?:^|[^\p{L}\p{N}])(?:paras?\.?|paragraphs?)(?:[ \t\u00a0]+\[\d+\][ \t\u00a0]*(?:[-–—]|,(?:[ \t\u00a0]*(?:and|or)\b)?|and\b|or\b))*[ \t\u00a0]*$/iu;
+  CLAUSE_START_AT_PREFIX = /(?:^|[\n.!?؛;:()]|[>*-])[ \t\u00a0]*At(?:[ \t\u00a0]+\[\d+\][ \t\u00a0]*(?:[-–—]|,(?:[ \t\u00a0]*(?:and|or)\b)?|and\b|or\b))*[ \t\u00a0]+$/u;
+  CASE_CITATION_AT_PREFIX = /(?:\[\d{4}\][^\n.;]*?|\b(?:v|vs\.?)\s+[^\n.;]{1,60}?|\bs(?:ection)?[ \t\u00a0]+\d+[A-Za-z]?(?:\(\d+\))?[^\n.;]{0,30}?)[ \t\u00a0]+at(?:[ \t\u00a0]+\[\d+\][ \t\u00a0]*(?:[-–—]|,(?:[ \t\u00a0]*(?:and|or)\b)?|and\b|or\b))*[ \t\u00a0]+$/iu;
+  PINPOINT_RANGE_AHEAD = /^[ \t\u00a0]*[-–—][ \t\u00a0]*\[\d+\]/u;
+  PINPOINT_RANGE_BEHIND = /\[\d+\][ \t\u00a0]*[-–—][ \t\u00a0]*$/u;
+});
 
 // src/agent/evidence-ledger.ts
 import { createHash as createHash10 } from "node:crypto";
@@ -9597,6 +9748,7 @@ function deriveResultsFromAnswer(text, ledger, options) {
 }
 var MAX_EXCERPT_CHARS = 400, MAX_CLAIM_CHARS = 280, CLAIM_LOOKBACK_CHARS = 600, SENTENCE_END, LEADING_LIST_MARK, EVIDENCE_MARKER_TEXT, LEADING_BLANK;
 var init_answer_citations = __esm(() => {
+  init_citations();
   init_evidence_ledger();
   SENTENCE_END = /[.!?。！？]/u;
   LEADING_LIST_MARK = /^[\s\-*•>]+/u;
@@ -9605,7 +9757,7 @@ var init_answer_citations = __esm(() => {
 });
 
 // src/agent/answer-guidelines.ts
-var ANSWER_IMAGE_EMBED_RULE = "**Images**: when a retrieved result is itself an image file (png, jpg, jpeg, gif, webp, svg, bmp, avif, heic, tiff) and showing it answers the question, embed it in the answer as a markdown image on its own line: `![short description](<absolute source path>)`, followed by its citation, e.g. `![Q3 revenue chart](</data/reports/q3 chart.png>) [e2]`. Use only the real source path of retrieved evidence. Never invent a path, use relative paths, or embed remote URLs. Apart from these image embeds, the no-file-paths rule stands.", ANSWER_IMAGE_DELTA_RULE = "An image already embedded in the first answer must not be embedded again.", ANSWER_CITATION_RULE = "**Citations**: cite supporting evidence inline with the evidence id shown next to each retrieved result, in square brackets (e.g. [e3], or [e3][e7] for two). Copy ids exactly; never invent one. For a local file you opened yourself with `bash`, cite `[file:<absolute path>]`. Cite the evidence that backs a sentence at the end of that sentence. Do not cite numbers, paths, or anything else.";
+var ANSWER_IMAGE_EMBED_RULE = "**Images**: when a retrieved result is itself an image file (png, jpg, jpeg, gif, webp, svg, bmp, avif, heic, tiff) and showing it answers the question, embed it in the answer as a markdown image on its own line: `![short description](<absolute source path>)`, followed by its citation, e.g. `![Q3 revenue chart](</data/reports/q3 chart.png>) [e2]`. Use only the real source path of retrieved evidence. Never invent a path, use relative paths, or embed remote URLs. Apart from these image embeds, the no-file-paths rule stands.", ANSWER_IMAGE_DELTA_RULE = "An image already embedded in the first answer must not be embedded again.", ANSWER_CITATION_RULE = "**Citations**: cite supporting evidence inline with the evidence id shown next to each retrieved result, in square brackets (e.g. [e3], or [e3][e7] for two). Copy ids exactly; never invent one. For a local file you opened yourself with `bash`, cite `[file:<absolute path>]`. Cite the evidence that backs a sentence at the end of that sentence. Do not cite numbers, paths, or anything else.", ANSWER_DELIVERABLE_RULE = '**Deliverables and artifacts**: if the user asks you to draft, write, generate, or prepare a deliverable or artifact (a reply email, letter, legal opinion, summary document, standard-format case citation block, checklist, report, or workflow document — including requests phrased as "how should I reply..." to a specific message), produce the finished, ready-to-use artifact as a file (using `write`, `edit`, or `bash` in local sessions, without overwriting existing source files) in a file format, extension, and structure aligned with the context and domain (e.g. `.eml` or `.md`/`.txt` with headers for an email reply; `.md`, `.txt`, `.docx`, `.hwp`, or `.pdf` matching workspace conventions for a legal opinion, insertable case-citation block, or summary document; `.csv`, `.xlsx`, or `.json` for structured tabular data), containing the complete unabridged artifact (including required headers, precedent passages, pinpoint paragraphs, and standard-format citations) inside the file instead of merely describing how to produce it. Do NOT dump the full artifact text into your reply: unless the user explicitly requests otherwise, keep your reply concise within the at-most-5-bullet direct-answer format, clearly stating the created file\'s exact path/location and briefly summarizing what was generated with inline citations to the underlying source evidence (`[eN]` or `[file:<source path>]`); never cite the newly created deliverable file itself as `[file:...]`.', ANSWER_DELIVERABLE_DELTA_RULE = "If a deliverable or artifact was requested and the first answer could not write the file yet (the fast phase runs without tool calls) or only gave advice about it, create the complete context-aligned deliverable file now during verification and report the created file's location and a brief summary concisely.";
 
 // src/agent/datasource-skill.ts
 import { Type as Type3 } from "typebox";
@@ -13941,6 +14093,7 @@ function recordStructuredResultsSession(sessionId, query, details, memory, compo
 }
 var init_search_documents = __esm(() => {
   init_memory();
+  init_citations();
 });
 
 // src/agent/search-all-tool.ts
@@ -14257,6 +14410,10 @@ function buildSystemPrompt(config) {
   const modelId = config.modelId ?? "the configured model";
   const toolLines = [
     toolLine(config, "bash", "read and inspect configured document collections with ls, find, grep, cat, and similar tools, and — in local sessions — organize local files when the user asks (rename, move, copy, create folders, move to trash)"),
+    ...config.remoteSession ? [] : [
+      toolLine(config, "write", "create a new file in local sessions — use this to produce requested deliverables and artifacts in a file format aligned with the request and context without overwriting existing source files"),
+      toolLine(config, "edit", "surgically update an existing deliverable or config file when the user asks (local sessions only)")
+    ],
     toolLine(config, "jikji_find", "local discovery through Jikji answer packs"),
     toolLine(config, "everything_search", "instant Windows file/folder name search (voidtools Everything) across the configured search folders"),
     toolLine(config, "fsearch_search", "instant macOS/Linux file/folder name search (fsearch-cli/FSearch) across the configured search folders, with a slow filesystem-walk fallback"),
@@ -14273,6 +14430,12 @@ function buildSystemPrompt(config) {
     ...config.toolNames.filter((name) => name.startsWith("search_datasource_")).map((name) => `- **${name}**: search only the ${name.slice("search_datasource_".length).replace(/_/g, "-")} datasource connection, spawning no other datasource CLIs`),
     ...config.toolNames.filter((name) => ![
       "bash",
+      "read",
+      "write",
+      "edit",
+      "grep",
+      "find",
+      "ls",
       "jikji_find",
       "everything_search",
       "fsearch_search",
@@ -14405,8 +14568,9 @@ Your answer is your final reply, written as plain text, never as a tool call. Ci
 ## Answer Guidelines
 
 - **Complete vs. delta answer**: When no first answer reached the caller, give the complete core answer. When a first answer was already delivered, return only the delta against it — corrections and newly verified findings — and never repeat its unchanged content; if nothing changed, confirm the first answer in one short line.
-- **Bullet-point core answer**: Provide the core answer to the user's question in at most 5 bullet points. If additional explanation or context is necessary, append it after the bullet points.
-- **Direct answer only**: The caller only needs the answer to their question. Never include specific file paths, datasource descriptions, or retrieval mechanics/principles in the answer (source paths are attached to each citation from the evidence ids you cite). Do not quote raw chunk text.
+- **Bullet-point core 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 or context is necessary, append it after the bullet points.
+- **Direct answer only**: Unless the user explicitly requests otherwise, the caller only needs the direct answer to their question. Never include specific file paths, datasource descriptions, or retrieval mechanics/principles in the answer (source paths are attached to each citation from the evidence ids you cite). Do not quote raw chunk text.
+- ${ANSWER_DELIVERABLE_RULE} ${ANSWER_DELIVERABLE_DELTA_RULE}
 - ${ANSWER_CITATION_RULE}
 - ${ANSWER_IMAGE_EMBED_RULE} ${ANSWER_IMAGE_DELTA_RULE}
 - **No per-source negative reports**: Never report individual negative findings per source (e.g. "no information found in Slack" or "checked Drive but found nothing"). Simply omit unproductive sources from the answer and focus on what was found or provide a concise overall conclusion.
@@ -20363,7 +20527,8 @@ class AutoRAGAgent {
       const everythingClient = new EverythingClient({
         ...options.everything ?? {},
         root: this.workspaceProjectRoot,
-        folders: this.searchPaths
+        folders: this.searchPaths,
+        excludeFolders: this.excludePaths
       });
       if (everythingClient.isSupported())
         this.everythingClient = everythingClient;
@@ -20469,8 +20634,9 @@ class AutoRAGAgent {
       return true;
     });
     this.tools = tools;
+    const builtinToolNames = this.remoteSession ? PI_BUILTIN_TOOL_NAMES.filter((name) => name !== "write" && name !== "edit") : PI_BUILTIN_TOOL_NAMES;
     const toolNames = [
-      ...PI_BUILTIN_TOOL_NAMES,
+      ...builtinToolNames,
       ...tools.map((tool) => tool.name),
       ...this.jevExtension !== undefined ? [JEV_TOOL_NAME] : []
     ];
@@ -20486,7 +20652,7 @@ class AutoRAGAgent {
     this.innerAgent = {
       state: {
         systemPrompt,
-        tools: [...PI_BUILTIN_TOOL_NAMES.map((name) => ({ name })), ...tools],
+        tools: [...builtinToolNames.map((name) => ({ name })), ...tools],
         messages: []
       },
       transformContext: (messages) => this.withMemoryContext(messages)
@@ -20803,6 +20969,7 @@ ${context.text}
         ...this.jevExtension !== undefined ? [this.jevExtension] : []
       ],
       ...this.jevExtension !== undefined ? { extensionToolNames: [JEV_TOOL_NAME] } : {},
+      remoteSession: this.remoteSession,
       ...this.updateNotice === undefined ? {} : { updateNotice: this.updateNotice }
     });
     this.boundPiRuntime = runtime.runtime;
@@ -21729,8 +21896,9 @@ ${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.
 
 ` + `Formatting and content rules for the answer:
-` + `- 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.
-` + `- Answer the question directly. Do not include specific file paths, datasource descriptions, or retrieval mechanics in the answer text.
+` + `- 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").
@@ -21769,12 +21937,14 @@ ${baseline}
 ` + `- 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):
-` + `- 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.
-` + `- Answer the question directly. Do not include specific file paths, datasource descriptions, or retrieval mechanics in the answer text.
+` + `- 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").
@@ -21802,8 +21972,9 @@ ${initialRetrievalContext ?? "Pending; continue only with a brief progress state
 ` + `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:
-` + `- 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.
-` + `- Answer the question directly. Do not include specific file paths, datasource descriptions, or retrieval mechanics in the answer text.
+` + `- 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").
@@ -21958,6 +22129,7 @@ ${initialRetrievalContext ?? "Pending; continue only with a brief progress state
           ...summary.diagnostics,
           ...minsyncDiagnostics,
           ...stagingExcludedDiagnostics(minsync?.stagingExcluded),
+          ...fileSkippedDiagnostics(minsync?.filesFailed),
           ...everythingDiagnostics,
           ...fsearchDiagnostics
         ],
@@ -22594,6 +22766,14 @@ function stagingExcludedDiagnostics(excluded) {
     source
   }));
 }
+function fileSkippedDiagnostics(failed) {
+  return (failed ?? []).map((failure) => ({
+    code: "minsync-file-skipped",
+    severity: "warning",
+    message: `MinSync skipped this file at the ${failure.stage} stage and will retry it on the next sync: ${sanitizeDiagnosticMessage(failure.error)}`,
+    source: failure.path
+  }));
+}
 function toSearchDiagnostic(diagnostic) {
   return {
     code: diagnostic.code,
@@ -22812,7 +22992,7 @@ var init_results = __esm(() => {
     confidence: Type18.Optional(Type18.Number({ description: "Evidence confidence, 0..1", minimum: 0, maximum: 1 }))
   });
   answerSchema = Type18.String({
-    description: `Answer for the caller. At most 5 bullet points (plus optional explanation); reference results by bracketed number (e.g. [1], [2]) without file paths or raw chunk text, except the image-embed exception below. ${ANSWER_CITATION_RULE} ${ANSWER_IMAGE_EMBED_RULE}`
+    description: `Answer for the caller. Unless the user explicitly requests otherwise, provide at most 5 bullet points (plus optional explanation) and answer directly; reference results by bracketed number (e.g. [1], [2]) without file paths or raw chunk text, except the image-embed exception below. ${ANSWER_DELIVERABLE_RULE} ${ANSWER_CITATION_RULE} ${ANSWER_IMAGE_EMBED_RULE}`
   });
   warningsSchema = Type18.Optional(Type18.Array(Type18.String(), { description: "Optional warnings about this result set" }));
   curatedResultFields = {
@@ -23110,7 +23290,7 @@ function chatCrawler(id) {
     hierarchy: (metadata) => ["chats", str(metadata.chatName) ?? str(metadata.chatId) ?? "unknown"]
   };
 }
-var str = (value) => typeof value === "string" && value.length > 0 ? value : undefined, slack, CRAWLER_DESCRIPTORS;
+var str = (value) => typeof value === "string" && value.length > 0 ? value : undefined, slack, kakao, CRAWLER_DESCRIPTORS;
 var init_descriptors = __esm(() => {
   slack = {
     id: "slack",
@@ -23141,10 +23321,32 @@ var init_descriptors = __esm(() => {
       str(metadata.channelName) ?? str(metadata.channelId) ?? "unknown"
     ]
   };
+  kakao = {
+    id: "kakao",
+    keyExpression: "m.chunk_id",
+    chunkId: (key) => key,
+    fromClause: "chunks m",
+    liveWhere: "length(trim(coalesce(m.text, ''))) > 0 and not (json_valid(m.text) and (json_extract(m.text, '$.feedType') is not null or json_extract(m.text, '$.csIP') is not null))",
+    textExpression: "coalesce(m.text, '')",
+    titleExpression: "m.chat_name",
+    epochSecondsExpression: "unixepoch(m.started_at)",
+    fts: { table: "chunks_fts", joinOn: "chunks_fts.rowid = m.rowid" },
+    metadata: [
+      { sql: "m.chat_id", key: "chatId" },
+      { sql: "m.chat_name", key: "chatName" },
+      { sql: "m.sender_nickname", key: "senderNickname" },
+      { sql: "m.started_at", key: "startedAt" },
+      { sql: "m.ended_at", key: "endedAt" }
+    ],
+    channelIdExpression: "m.chat_id",
+    channelNameExpression: "m.chat_name",
+    hierarchy: (metadata) => ["chats", str(metadata.chatName) ?? str(metadata.chatId) ?? "unknown"]
+  };
   CRAWLER_DESCRIPTORS = {
     slack,
     telegram: chatCrawler("telegram"),
-    whatsapp: chatCrawler("whatsapp")
+    whatsapp: chatCrawler("whatsapp"),
+    kakao
   };
 });
 
@@ -27292,1436 +27494,1528 @@ var init_methods3 = __esm(() => {
   DEFAULT_KAKAO_TAGS = ["kakaotalk", "personal", "pii"];
 });
 
-// src/datasource/skills/lazykatok/skill.ts
-class LazykatokSkill {
-  client;
-  instanceId;
-  instances;
-  pollingIntervalMs;
-  tags;
-  lastIndexedAt;
-  constructor(options) {
-    this.client = options.client;
-    this.instanceId = options.instanceId ?? DEFAULT_INSTANCE_ID4;
-    this.instances = options.instances !== undefined && options.instances.length > 0 ? options.instances : [this.instanceId];
-    this.pollingIntervalMs = options.pollingIntervalMs ?? DEFAULT_POLLING_INTERVAL_MS3;
-    this.tags = options.tags ?? DEFAULT_KAKAO_TAGS2;
-    this.lastIndexedAt = options.lastIndexedAt;
-  }
-  describe() {
-    return {
-      name: KAKAO_DATASOURCE_ID2,
-      id: KAKAO_DATASOURCE_ID2,
-      type: KAKAO_SKILL_TYPE,
-      description: "KakaoTalk datasource via the external lazykatok CLI",
-      capabilities: ["chat", "external-cli", "polling", "bm25", "semantic"],
-      tags: this.tags,
-      status: "active",
-      requiresExternalCli: true,
-      datasourceId: KAKAO_DATASOURCE_ID2,
-      instanceId: this.instanceId,
-      instances: this.instances
-    };
+// src/datasource/sidecar/fusion.ts
+function encodeVector(vector) {
+  const floats = Float32Array.from(vector);
+  return new Uint8Array(floats.buffer, floats.byteOffset, floats.byteLength);
+}
+function decodeVector(blob) {
+  if (blob.byteLength % Float32Array.BYTES_PER_ELEMENT !== 0)
+    return;
+  return new Float32Array(new Uint8Array(blob).buffer);
+}
+function cosine3(a, b) {
+  if (a.length === 0 || a.length !== b.length)
+    return 0;
+  let dot = 0;
+  let normA = 0;
+  let normB = 0;
+  for (let index = 0;index < a.length; index += 1) {
+    const x = a[index] ?? 0;
+    const y = b[index] ?? 0;
+    dot += x * y;
+    normA += x * x;
+    normB += y * y;
   }
-  polling() {
-    return {
-      mode: "poll",
-      intervalMs: this.pollingIntervalMs,
-      lastIndexedAt: this.lastIndexedAt
-    };
+  if (normA === 0 || normB === 0)
+    return 0;
+  return dot / Math.sqrt(normA * normB);
+}
+function topKByCosine(query, rows, k) {
+  if (k <= 0)
+    return [];
+  const scored = [];
+  for (const row of rows) {
+    const score = cosine3(query, row.vector);
+    if (score > 0)
+      scored.push({ key: row.key, score });
   }
-  async index(options) {
-    try {
-      const doctor = await this.client.doctor(options?.signal);
-      if (!doctor.ok)
-        return this.fail("datasource-unavailable", doctor);
-      const sync = await this.client.sync(options?.signal);
-      if (!sync.ok)
-        return this.fail("datasource-index-failed", sync);
-      const indexResult = await this.client.index(options?.signal);
-      if (!indexResult.ok)
-        return this.fail("datasource-index-failed", indexResult);
-      this.lastIndexedAt = Date.now();
-      const chunkCount = "data" in indexResult && typeof indexResult.data === "object" && indexResult.data !== null && "chunkCount" in indexResult.data && typeof indexResult.data.chunkCount === "number" ? indexResult.data.chunkCount : 0;
-      return {
-        ok: true,
-        instanceId: this.instanceId,
-        skill: KAKAO_DATASOURCE_ID2,
-        chunkCount,
-        indexedAt: this.lastIndexedAt,
-        diagnostics: []
-      };
-    } catch (error) {
-      return this.fail("datasource-unavailable", {
-        ok: false,
-        reason: "spawn-error",
-        stdout: "",
-        stderr: describeRetrievalError(error),
-        code: null
-      });
+  scored.sort((a, b) => b.score - a.score || a.key.localeCompare(b.key));
+  return scored.slice(0, k);
+}
+function reciprocalRankFusion(lists) {
+  const scores = new Map;
+  for (const list of lists) {
+    const seen = new Set;
+    for (const [index, key] of list.entries()) {
+      if (seen.has(key))
+        continue;
+      seen.add(key);
+      scores.set(key, (scores.get(key) ?? 0) + 1 / (RRF_K2 + index + 1));
     }
   }
-  retrievalMethods() {
-    return [
-      new LazykatokBm25Method({ client: this.client, instanceId: this.instanceId, tags: this.tags }),
-      new LazykatokSemanticMethod({ client: this.client, instanceId: this.instanceId, tags: this.tags })
+  return [...scores.entries()].map(([key, score]) => ({ key, score })).sort((a, b) => b.score - a.score);
+}
+var RRF_K2 = 60;
+
+// src/datasource/sidecar/index.ts
+import { createHash as createHash14 } from "node:crypto";
+import { mkdirSync as mkdirSync22, statSync as statSync3 } from "node:fs";
+import { dirname as dirname24 } from "node:path";
+import { DatabaseSync as DatabaseSync2 } from "node:sqlite";
+import { pathToFileURL as pathToFileURL2 } from "node:url";
+
+class CrawlerSidecar {
+  descriptor;
+  options;
+  select;
+  where;
+  opened;
+  attached;
+  lastSemantic;
+  constructor(options) {
+    this.options = options;
+    const { descriptor } = options;
+    this.descriptor = descriptor;
+    const metadataColumns = descriptor.metadata.map((column, index) => `${column.sql} as m${index}`);
+    this.select = [
+      `${descriptor.keyExpression} as k`,
+      `${descriptor.textExpression} as content`,
+      `${descriptor.titleExpression ?? "null"} as title`,
+      `${descriptor.epochSecondsExpression ?? "null"} as epoch`,
+      ...metadataColumns
+    ].join(", ");
+    const scope = descriptor.scopeExpression !== undefined && options.workspaceId !== undefined;
+    const allowlist = [
+      ...(options.channelIds?.length ?? 0) > 0 ? [`${descriptor.channelIdExpression} in (${options.channelIds?.map(() => "?").join(", ")})`] : [],
+      ...(options.channelNames?.length ?? 0) > 0 ? [`${descriptor.channelNameExpression} in (${options.channelNames?.map(() => "?").join(", ")})`] : []
     ];
+    this.where = [
+      descriptor.liveWhere,
+      ...scope ? [`${descriptor.scopeExpression} = ?`] : [],
+      ...allowlist.length > 0 ? [`(${allowlist.join(" or ")})`] : []
+    ].join(" and ");
   }
-  describeSources() {
-    return this.instances.map((instanceId) => ({
-      source: datasourceSourcePath(KAKAO_DATASOURCE_ID2, instanceId),
-      datasourceId: KAKAO_DATASOURCE_ID2,
-      skill: KAKAO_DATASOURCE_ID2,
-      instanceId,
-      contentType: "chat",
-      metadata: {
-        datasourceId: KAKAO_DATASOURCE_ID2,
-        instanceId,
-        tags: this.tags
-      }
-    }));
+  get db() {
+    if (this.opened !== undefined)
+      return this.opened;
+    mkdirSync22(dirname24(this.options.sidecarDbPath), { recursive: true });
+    const db = new DatabaseSync2(this.options.sidecarDbPath);
+    db.exec(`pragma busy_timeout = ${BUSY_TIMEOUT_MS}`);
+    db.exec("pragma journal_mode = wal");
+    db.exec(`
+			create table if not exists sidecar_meta (k text primary key, v text not null);
+			create table if not exists sidecar_vec (message_key text primary key, hash text not null, vec blob not null);
+		`);
+    this.opened = db;
+    return db;
   }
-  skillManifest() {
-    const instanceSources = this.instances.map((instanceId) => `- \`${datasourceSourcePath(KAKAO_DATASOURCE_ID2, instanceId)}\``).join(`
-`);
-    const cadence = this.pollingIntervalMs > 0 ? `roughly every ${Math.round(this.pollingIntervalMs / 60000)} minute(s) when auto-refresh runs` : "on manual refresh only";
-    return {
-      name: `datasource-${KAKAO_DATASOURCE_ID2}`,
-      description: "Search indexed KakaoTalk chat history (messages, senders, room context). Use for questions about KakaoTalk conversations, decisions made in chats, or who said what.",
-      content: [
-        `# KakaoTalk datasource (${KAKAO_SKILL_TYPE})`,
-        "",
-        "This skill searches KakaoTalk chats that are indexed through the external `lazykatok` CLI. AutoRAG never reads KakaoTalk databases directly.",
-        "",
-        "## When to use",
-        "Use this skill when the question is about KakaoTalk conversations, chat participants, or content shared inside chats.",
-        "",
-        "## Indexing",
-        `Indexing is server-managed and refreshed ${cadence}. You do not trigger indexing; just search.`,
-        "",
-        "## How to search",
-        `Call the dedicated \`${datasourceSearchToolName(KAKAO_DATASOURCE_ID2)}\` tool with a natural-language \`query\` and \`topK\`. This datasource does not support per-source scope narrowing. Configured datasource instances:`,
-        instanceSources.length > 0 ? instanceSources : "- (no configured instances)",
-        "",
-        "Chat/channel filtering is owned by lazykatok.",
-        "",
-        "## Native CLI",
-        "The external `lazykatok` CLI owns the archive, index, and credentials. When the dedicated tool cannot express what you need, call lazykatok directly through `bash`:",
-        '- `lazykatok search bm25 "<query>" --json --limit 20` — lexical search (`semantic` mode also available)',
-        "- `lazykatok chunk get <chunkId> --json` — fetch one chunk by id (chunk ids appear in result metadata)",
-        "- `lazykatok chunk context <chunkId> --json` — surrounding messages of a chunk",
-        "Never pass datasource virtual paths (`/kakao/...`) to bash; they are not OS paths.",
-        "",
-        "## Output rules",
-        "Datasource source identifiers such as `/kakao/<instance>/chunks/<id>` are internal and opaque. Never put them, real file paths, account IDs, or phone numbers in the visible answer."
-      ].join(`
-`)
-    };
-  }
-  fail(code, result) {
-    const base = result.stderr.length > 0 ? `${result.reason}: ${result.stderr.trim().slice(0, 4000)}` : result.reason;
-    const message = code === "datasource-index-failed" && result.reason === "timeout" ? `${base} — ${connectorSyncTimeoutHint()}` : base;
-    const diagnostic = {
-      code,
-      severity: code === "datasource-unavailable" ? "warning" : "error",
-      message,
-      instanceId: this.instanceId,
-      source: KAKAO_DATASOURCE_ID2
-    };
-    return {
-      ok: false,
-      instanceId: this.instanceId,
-      skill: KAKAO_DATASOURCE_ID2,
-      indexedAt: Date.now(),
-      diagnostics: [diagnostic],
-      error: result.reason,
-      code,
-      message
-    };
-  }
-}
-var KAKAO_DATASOURCE_ID2 = "kakao", KAKAO_SKILL_TYPE = "kakaotalk", DEFAULT_INSTANCE_ID4 = "default", DEFAULT_POLLING_INTERVAL_MS3, DEFAULT_KAKAO_TAGS2;
-var init_skill7 = __esm(() => {
-  init_skip();
-  init_scope2();
-  init_methods3();
-  DEFAULT_POLLING_INTERVAL_MS3 = 15 * 60 * 1000;
-  DEFAULT_KAKAO_TAGS2 = ["kakaotalk", "personal", "pii"];
-});
-
-// src/datasource/skills/lazykatok/index.ts
-var init_lazykatok = __esm(() => {
-  init_client8();
-  init_methods3();
-  init_skill7();
-  init_types5();
-});
-
-// src/datasource/skills/mail-export/connector.ts
-import { readdir as readdir4, readFile as readFile5, stat as stat6 } from "node:fs/promises";
-import { basename as basename12, dirname as dirname24, extname as extname6, join as join37 } from "node:path";
-import { simpleParser as simpleParser2 } from "mailparser";
-
-class MailExportConnector {
-  options;
-  constructor(options = {}) {
-    this.options = options;
-  }
-  async fetch() {
-    const paths = this.options.paths ?? [];
-    if (paths.length === 0)
-      return { ok: false, reason: "not-configured", message: "no mail export paths configured" };
-    const maxDocuments = this.options.maxDocuments ?? DEFAULT_MAX_DOCUMENTS3;
-    const maxBytes = this.options.maxBytesPerFile ?? DEFAULT_MAX_BYTES_PER_FILE2;
-    const files = [];
-    let missingPaths = 0;
-    for (const path of paths) {
-      try {
-        const info = await stat6(path);
-        if (info.isDirectory()) {
-          const entries = await readdir4(path, { recursive: true, withFileTypes: true });
-          for (const entry of entries) {
-            if (!entry.isFile())
-              continue;
-            const extension = extname6(entry.name).toLowerCase();
-            if (extension === ".eml" || extension === ".mbox") {
-              files.push(join37(entry.parentPath, entry.name));
-            }
-          }
-        } else if (info.isFile()) {
-          files.push(path);
-        }
-      } catch {
-        missingPaths += 1;
-      }
+  attachCrawler() {
+    let file;
+    try {
+      const { dev, ino } = statSync3(this.options.crawlerDbPath);
+      file = { dev, ino };
+    } catch {
+      this.detachCrawler();
+      return false;
     }
-    if (missingPaths === paths.length) {
-      return { ok: false, reason: "unavailable", message: "no configured mail export paths were readable" };
+    if (this.attached !== undefined) {
+      if (this.attached.dev === file.dev && this.attached.ino === file.ino)
+        return true;
+      this.detachCrawler();
     }
-    const documents = [];
-    const warnings = [];
-    if (missingPaths > 0)
-      warnings.push(`${missingPaths} configured path(s) were unreadable`);
-    let parseFailures = 0;
-    let skippedLarge = 0;
-    for (const file of files.sort()) {
-      if (documents.length >= maxDocuments)
-        break;
-      let bytes;
-      try {
-        const info = await stat6(file);
-        if (info.size > maxBytes) {
-          skippedLarge += 1;
-          continue;
-        }
-        bytes = await readFile5(file);
-      } catch {
-        parseFailures += 1;
-        continue;
-      }
-      const stem = basename12(file, extname6(file));
-      const mailboxDir = basename12(dirname24(file));
-      const hierarchy = ["mailboxes", mailboxDir, stem];
-      const extension = extname6(file).toLowerCase();
-      const rawMessages = extension === ".mbox" ? splitMbox(bytes.toString("utf8")) : [bytes.toString("utf8")];
-      for (const [index, raw] of rawMessages.entries()) {
-        if (documents.length >= maxDocuments)
-          break;
-        try {
-          const parsed = await simpleParser2(raw);
-          const subject = parsed.subject ?? "(no subject)";
-          const from = Array.isArray(parsed.from) ? parsed.from[0]?.text : parsed.from?.text;
-          const to = Array.isArray(parsed.to) ? parsed.to[0]?.text : parsed.to?.text;
-          const body = (parsed.text ?? "").trim().slice(0, MAX_BODY_CHARS);
-          const headerBlock = [
-            `Subject: ${subject}`,
-            ...from !== undefined ? [`From: ${from}`] : [],
-            ...to !== undefined ? [`To: ${to}`] : [],
-            ...parsed.date !== undefined ? [`Date: ${parsed.date.toISOString()}`] : []
-          ].join(`
-`);
-          const content = `${headerBlock}
-
-${body}`.trim();
-          if (content.length === 0)
-            continue;
-          documents.push({
-            docId: extension === ".mbox" ? `${stem}-${index + 1}` : stem,
-            hierarchy,
-            title: subject,
-            content,
-            ...parsed.date !== undefined ? { publishedAt: parsed.date.getTime() } : {},
-            metadata: {
-              path: file,
-              subject,
-              ...from !== undefined ? { from } : {},
-              messageIndex: index
-            }
-          });
-        } catch {
-          parseFailures += 1;
-        }
-      }
+    const crawlerUri = `${pathToFileURL2(this.options.crawlerDbPath).href}?mode=ro`.replaceAll("'", "''");
+    try {
+      this.db.exec(`attach database '${crawlerUri}' as crawler`);
+    } catch {
+      return false;
     }
-    if (skippedLarge > 0)
-      warnings.push(`${skippedLarge} file(s) exceeded the size limit and were skipped`);
-    if (parseFailures > 0)
-      warnings.push(`${parseFailures} message(s) failed to parse`);
-    return { ok: true, documents, ...warnings.length > 0 ? { warnings } : {} };
-  }
-}
-function splitMbox(text) {
-  const lines = text.split(/\r?\n/u);
-  const messages = [];
-  let current = [];
-  for (const line of lines) {
-    if (MBOX_SEPARATOR.test(line) && line.startsWith("From ")) {
-      if (current.length > 0)
-        messages.push(current.join(`
-`));
-      current = [];
-      continue;
+    const { descriptor } = this.options;
+    try {
+      this.db.prepare(`select ${this.select} from ${descriptor.fromClause} where ${this.where} limit 0`).all(...this.whereParams());
+      this.db.prepare(`select 1 from ${descriptor.fts.table} limit 0`).all();
+    } catch {
+      this.db.exec("detach database crawler");
+      return false;
     }
-    current.push(line);
+    this.attached = file;
+    this.lastSemantic = undefined;
+    return true;
   }
-  if (current.length > 0)
-    messages.push(current.join(`
-`));
-  return messages.filter((message) => message.trim().length > 0);
-}
-var DEFAULT_MAX_DOCUMENTS3 = 500, DEFAULT_MAX_BYTES_PER_FILE2, MAX_BODY_CHARS = 50000, MBOX_SEPARATOR;
-var init_connector4 = __esm(() => {
-  DEFAULT_MAX_BYTES_PER_FILE2 = 50 * 1024 * 1024;
-  MBOX_SEPARATOR = /^From .*\d{4}$/mu;
-});
-
-// src/datasource/skills/mail-export/skill.ts
-var MAIL_EXPORT_SKILL_DEFINITION, MailExportSkill;
-var init_skill8 = __esm(() => {
-  init_connector_skill();
-  init_connector4();
-  MAIL_EXPORT_SKILL_DEFINITION = {
-    skillName: "mail-export",
-    skillType: "mail-export",
-    description: "Local mail export datasource",
-    capabilities: ["email", "filesystem"],
-    defaultTags: ["mail-export", "email", "pii"],
-    contentType: "email",
-    manifestDescription: "Search indexed local mail exports (mbox/eml archives). Use for questions about archived email conversations.",
-    manifestNotes: [
-      "Only the server-configured export paths are indexed; archives are read locally and never uploaded."
-    ]
-  };
-  MailExportSkill = class MailExportSkill extends ConnectorDatasourceSkill {
-    constructor(options = {}) {
-      const { connector, connectorOptions, ...rest } = options;
-      super(MAIL_EXPORT_SKILL_DEFINITION, {
-        ...rest,
-        connector: connector ?? new MailExportConnector(connectorOptions ?? {})
-      });
-    }
-  };
-});
-
-// src/datasource/skills/mail-export/index.ts
-var init_mail_export = __esm(() => {
-  init_connector4();
-  init_skill8();
-});
-
-// src/datasource/skills/mailcrawl/config.ts
-import { chmodSync, mkdirSync as mkdirSync22 } from "node:fs";
-function validateMailcrawlInstanceId(instanceId) {
-  if (!/^[A-Za-z0-9][A-Za-z0-9._-]{0,127}$/u.test(instanceId)) {
-    throw new Error("mailcrawl instanceId must be a safe single path segment");
+  detachCrawler() {
+    if (this.attached === undefined)
+      return;
+    this.attached = undefined;
+    this.lastSemantic = undefined;
+    try {
+      this.opened?.exec("detach database crawler");
+    } catch {}
   }
-}
-function ensurePrivateMailcrawlDataDir(path) {
-  mkdirSync22(path, { recursive: true, mode: 448 });
-  if (process.platform !== "win32")
-    chmodSync(path, 448);
-}
-var init_config3 = () => {};
-
-// src/datasource/skills/mailcrawl/client.ts
-import { spawn as spawn15 } from "node:child_process";
-
-class MailcrawlClient {
-  options;
-  constructor(options = {}) {
-    this.options = options;
-    if (options.instanceId !== undefined)
-      validateMailcrawlInstanceId(options.instanceId);
+  get semanticEnabled() {
+    return this.options.embedder !== undefined;
   }
-  async sync(signal) {
-    const args = [
-      "sync",
-      "--json",
-      ...this.options.account === undefined ? [] : ["--account", this.options.account],
-      ...this.options.mailbox === undefined ? [] : ["--mailbox", this.options.mailbox],
-      ...this.options.backend === undefined ? [] : ["--backend", this.options.backend],
-      ...this.options.source === undefined ? [] : ["--source", this.options.source],
-      ...this.options.fixture === undefined ? [] : ["--fixture", this.options.fixture],
-      ...this.options.himalayaConfig === undefined ? [] : ["--himalaya-config", this.options.himalayaConfig]
-    ];
-    const result = await this.run(args, signal, this.indexTimeoutMs());
-    if (!result.ok)
-      return failure3(result);
-    const parsed = parseObject2(result.stdout);
-    if (parsed === undefined || !hasAnyNumber(parsed, ["added", "updated", "deleted", "unchanged", "chunksAdded"])) {
-      return failure3({ ...result, ok: false, reason: "invalid-output" });
-    }
-    const messages = number(parsed, "added") ?? number(parsed, "updated") ?? number(parsed, "unchanged") ?? 0;
-    return ok4({
-      messages,
-      added: number(parsed, "added"),
-      updated: number(parsed, "updated"),
-      deleted: number(parsed, "deleted"),
-      unchanged: number(parsed, "unchanged"),
-      chunksAdded: number(parsed, "chunksAdded"),
-      archiveRevision: string(parsed, "archiveRevision")
-    }, result);
+  close() {
+    this.opened?.close();
+    this.opened = undefined;
+    this.attached = undefined;
+    this.lastSemantic = undefined;
   }
-  async search(mode, query, options) {
-    if (query.trim().length === 0)
-      return { ok: true, hits: [], stdout: "", stderr: "", code: 0 };
-    const args = [
-      "search",
-      "--mode",
-      mode === "keyword" ? "bm25" : mode,
-      ...this.options.account === undefined ? [] : ["--account", this.options.account],
-      ...this.options.mailbox === undefined ? [] : ["--mailbox", this.options.mailbox],
-      "--limit",
-      String(options?.topK ?? 20),
-      "--json",
-      query.trim()
+  whereParams() {
+    const { descriptor, workspaceId, channelIds = [], channelNames = [] } = this.options;
+    return [
+      ...descriptor.scopeExpression !== undefined && workspaceId !== undefined ? [workspaceId] : [],
+      ...channelIds,
+      ...channelNames
     ];
-    const result = await this.run(args, options?.signal, this.options.timeoutMs ?? DEFAULT_SEARCH_TIMEOUT_MS2);
-    if (!result.ok)
-      return failure3(result);
-    const parsed = parseJson3(result.stdout);
-    if (parsed === undefined)
-      return failure3({ ...result, ok: false, reason: "invalid-output" });
-    const rows = Array.isArray(parsed) ? parsed : objectArray(parsed, ["hits", "results", "items"]);
-    if (rows === undefined)
-      return failure3({ ...result, ok: false, reason: "invalid-output" });
-    const hits = rows.map((row, index) => normalizeHit2(row, mode, index));
-    if (hits.some((hit) => hit === undefined)) {
-      return failure3({ ...result, ok: false, reason: "invalid-output" });
-    }
-    return {
-      ok: true,
-      hits,
-      stdout: result.stdout,
-      stderr: result.stderr,
-      code: result.code ?? 0
-    };
   }
-  indexTimeoutMs() {
-    return this.options.indexTimeoutMs ?? this.options.timeoutMs ?? DEFAULT_INDEX_TIMEOUT_MS3;
+  searchLexical(query, topK) {
+    const match = toMatchQuery(query);
+    if (match === undefined || topK <= 0 || !this.attachCrawler())
+      return [];
+    const { descriptor } = this.options;
+    const rows = this.db.prepare(`select ${this.select}, bm25(${descriptor.fts.table}) as rank
+				from ${descriptor.fromClause} join ${descriptor.fts.table} on ${descriptor.fts.joinOn}
+				where ${descriptor.fts.table} match ? and ${this.where}
+				order by rank limit ?`).all(match, ...this.whereParams(), topK);
+    return rows.map((row) => this.toHit(this.toRow(row), -Number(row.rank)));
   }
-  async run(args, signal, timeoutMs) {
-    const violation = findRemoteEmbeddingViolation({ ...process.env, ...this.options.env ?? {} });
-    if (violation !== undefined) {
-      return { ok: false, reason: "remote-embedding-rejected", stdout: "", stderr: "", code: null };
+  async searchSemantic(query, topK) {
+    const trimmed = query.trim();
+    if (this.options.embedder === undefined || trimmed.length === 0 || topK <= 0 || !this.attachCrawler())
+      return [];
+    const ranked = await this.rankSemantic(trimmed, topK * POOL_FACTOR);
+    if (ranked.length === 0)
+      return [];
+    const live = this.hydrate(ranked.map((entry) => entry.key));
+    const hits = [];
+    for (const entry of ranked) {
+      const row = live.get(entry.key);
+      if (row !== undefined)
+        hits.push(this.toHit(row, entry.score));
+      if (hits.length >= topK)
+        break;
     }
-    const dataDir = this.options.dataDir;
-    if (dataDir !== undefined)
-      ensurePrivateMailcrawlDataDir(dataDir);
-    const env = {};
-    for (const [key, value] of Object.entries(process.env)) {
-      if (value !== undefined && (SAFE_ENV_KEYS2.has(key) || SAFE_MAILCRAWL_ENV_KEYS.has(key)))
-        env[key] = value;
+    return hits;
+  }
+  async rankSemantic(query, pool) {
+    const now = Date.now();
+    let cached = this.lastSemantic;
+    if (cached === undefined || cached.query !== query || now - cached.at > SEMANTIC_REUSE_MS) {
+      const embedder = this.options.embedder;
+      if (embedder === undefined)
+        return [];
+      cached = {
+        query,
+        at: now,
+        vector: embedder.embed([query]).then((vectors) => vectors[0], () => {
+          return;
+        }),
+        ranked: undefined
+      };
+      this.lastSemantic = cached;
     }
-    for (const [key, value] of Object.entries(this.options.env ?? {})) {
-      if (value === undefined)
-        delete env[key];
-      else if (SAFE_ENV_KEYS2.has(key) || SAFE_MAILCRAWL_ENV_KEYS.has(key))
-        env[key] = value;
+    const vector = await cached.vector;
+    if (vector === undefined) {
+      if (this.lastSemantic === cached)
+        this.lastSemantic = undefined;
+      return [];
     }
-    if (dataDir !== undefined)
-      env.MAILCRAWL_DATA_DIR = dataDir;
-    return spawnMailcrawl(this.options.binaryPath ?? DEFAULT_BINARY4, args, env, this.options, signal, timeoutMs);
+    if (cached.ranked === undefined || cached.ranked.pool < pool) {
+      const wide = pool * POOL_FACTOR;
+      cached.ranked = { pool: wide, keys: topKByCosine(vector, this.storedVectors(), wide) };
+    }
+    return cached.ranked.keys.slice(0, pool);
   }
-  async index(signal) {
-    const result = await this.run(["index", "--json"], signal, this.indexTimeoutMs());
-    if (!result.ok)
-      return failure3(result);
-    const parsed = parseObject2(result.stdout);
-    if (parsed === undefined || number(parsed, "embedded") === undefined && number(parsed, "reused") === undefined && string(parsed, "generation") === undefined) {
-      return failure3({ ...result, ok: false, reason: "invalid-output" });
+  async searchHybrid(query, topK) {
+    if (topK <= 0)
+      return [];
+    const pool = topK * POOL_FACTOR;
+    const lexical = this.searchLexical(query, pool);
+    const semantic = await this.searchSemantic(query, pool);
+    if (semantic.length === 0)
+      return lexical.slice(0, topK);
+    const byKey = new Map;
+    for (const hit of [...semantic, ...lexical])
+      byKey.set(hit.key, hit);
+    const fused = [];
+    for (const entry of reciprocalRankFusion([lexical.map((hit) => hit.key), semantic.map((hit) => hit.key)])) {
+      const hit = byKey.get(entry.key);
+      if (hit !== undefined)
+        fused.push({ ...hit, score: entry.score });
+      if (fused.length >= topK)
+        break;
     }
-    return ok4({
-      embedded: number(parsed, "embedded"),
-      reused: number(parsed, "reused"),
-      generation: string(parsed, "generation"),
-      archiveRevision: string(parsed, "archiveRevision"),
-      rebuilt: typeof parsed.rebuilt === "boolean" ? parsed.rebuilt : undefined,
-      embedder: string(parsed, "embedder")
-    }, result);
+    return fused;
   }
-}
-function spawnMailcrawl(binary, args, env, options, signal, timeoutMs) {
-  return new Promise((resolveResult) => {
-    const portable = portableSpawnCommand(binary, args);
-    const child = spawn15(portable.command, portable.args, {
-      env,
-      detached: process.platform !== "win32",
-      stdio: ["ignore", "pipe", "pipe"]
-    });
-    let stdout = { text: "", bytes: 0, exceeded: false }, stderr = { text: "", bytes: 0, exceeded: false }, settled = false, reason;
-    const max = options.maxBufferBytes ?? DEFAULT_MAX_BUFFER_BYTES4;
-    const timer = setTimeout(() => {
-      reason = "timeout";
-      terminate4(child);
-    }, timeoutMs);
-    const abort = () => {
-      reason = "aborted";
-      terminate4(child);
-    };
-    if (signal?.aborted)
-      abort();
-    signal?.addEventListener("abort", abort, { once: true });
-    child.stdout.setEncoding("utf8");
-    child.stderr.setEncoding("utf8");
-    child.stdout.on("data", (chunk) => {
-      stdout = appendOutput(stdout, chunk, max);
-      if (stdout.exceeded && reason === undefined) {
-        reason = "stdout-too-large";
-        terminate4(child);
+  async sync() {
+    const { embedder } = this.options;
+    if (embedder === undefined)
+      return { ok: false, message: "no embedder configured" };
+    if (!this.attachCrawler())
+      return { ok: false, message: "crawler archive not available yet" };
+    let identity;
+    try {
+      identity = await embedder.identity();
+    } catch (error) {
+      return { ok: false, message: errorMessage2(error) };
+    }
+    this.adoptIdentity(identity);
+    const pruned = this.prune();
+    const pending = this.pendingKeys();
+    const batchSize = Math.max(1, this.options.batchSize ?? DEFAULT_BATCH_SIZE);
+    const budget = Math.max(1, this.options.maxEmbedPerSync ?? DEFAULT_MAX_EMBED_PER_SYNC);
+    const target = Math.min(pending.length, budget);
+    const embed = async (rows) => {
+      const vectors = await embedder.embed(rows.map((row) => row.input));
+      if (vectors.length !== rows.length) {
+        throw new Error(`embedder returned ${vectors.length} vectors for ${rows.length} messages`);
       }
-    });
-    child.stderr.on("data", (chunk) => {
-      stderr = appendOutput(stderr, chunk, max);
-      if (stderr.exceeded && reason === undefined) {
-        reason = "stderr-too-large";
-        terminate4(child);
+      const wrong = vectors.find((vector) => vector.length !== identity.dimension);
+      if (wrong !== undefined) {
+        throw new Error(`embedder returned a ${wrong.length}-dimension vector, expected ${identity.dimension}`);
+      }
+      return vectors;
+    };
+    let embedded = 0;
+    let skipped = 0;
+    for (let offset = 0;offset < target; offset += batchSize) {
+      const rows = [...this.hydrate(pending.slice(offset, Math.min(offset + batchSize, target))).values()];
+      if (rows.length === 0)
+        continue;
+      try {
+        this.store(rows, await embed(rows));
+        embedded += rows.length;
+      } catch (batchError) {
+        const rejected = [];
+        let lastError = batchError;
+        for (const row of rows) {
+          try {
+            this.store([row], await embed([row]));
+            embedded += 1;
+          } catch (error) {
+            rejected.push(row);
+            lastError = error;
+          }
+        }
+        if (rejected.length === 0)
+          continue;
+        if (embedded === 0)
+          return { ok: false, message: errorMessage2(lastError) };
+        this.store(rejected, undefined);
+        skipped += rejected.length;
       }
-    });
-    child.on("error", (error) => {
-      if (settled)
-        return;
-      settled = true;
-      clearTimeout(timer);
-      signal?.removeEventListener("abort", abort);
-      resolveResult({
-        ok: false,
-        reason: error.code === "ENOENT" ? "binary-missing" : "spawn-error",
-        stdout: stdout.text,
-        stderr: "",
-        code: null
-      });
-    });
-    child.on("close", (code) => {
-      if (settled)
-        return;
-      settled = true;
-      clearTimeout(timer);
-      signal?.removeEventListener("abort", abort);
-      resolveResult({
-        ok: code === 0 && reason === undefined,
-        ...reason === undefined ? code === 0 ? {} : { reason: "nonzero-exit" } : { reason },
-        stdout: stdout.text,
-        stderr: stderr.text,
-        code
-      });
-    });
-  });
-}
-function terminate4(child) {
-  if (child.killed)
-    return;
-  if (process.platform !== "win32" && child.pid !== undefined) {
-    try {
-      process.kill(-child.pid, "SIGTERM");
-      return;
-    } catch {}
-  }
-  child.kill("SIGTERM");
-}
-function parseJson3(text) {
-  try {
-    return JSON.parse(text.trim());
-  } catch {
-    return;
-  }
-}
-function parseObject2(text) {
-  const value = parseJson3(text);
-  return value !== null && typeof value === "object" && !Array.isArray(value) ? value : undefined;
-}
-function objectArray(value, keys) {
-  if (value === null || typeof value !== "object" || Array.isArray(value))
-    return;
-  for (const key of keys)
-    if (Array.isArray(value[key]))
-      return value[key];
-  return;
-}
-function hasAnyNumber(value, keys) {
-  return keys.some((key) => number(value, key) !== undefined);
-}
-function normalizeHit2(value, mode, index) {
-  if (value === null || typeof value !== "object" || Array.isArray(value))
-    return;
-  const row = value;
-  const text = (key) => typeof row[key] === "string" ? row[key] : "";
-  const chunkId = text("chunkId") || text("chunk_id");
-  const messageId = text("messageId") || text("message_id");
-  const accountId = text("accountId") || text("account_id");
-  const mailbox = text("mailbox");
-  const snippet = text("snippet") || text("content") || text("text");
-  if (!isSafeOpaqueId(chunkId) || !isSafeOpaqueId(messageId) || !snippet)
-    return;
-  const to = Array.isArray(row.to) ? row.to.filter((item) => typeof item === "string") : [];
-  return {
-    chunkId,
-    messageId,
-    threadId: text("threadId") || text("thread_id"),
-    accountId,
-    mailbox,
-    subject: text("subject"),
-    from: text("from"),
-    to,
-    date: text("date"),
-    snippet,
-    score: typeof row.score === "number" ? row.score : 1 / (index + 1),
-    mode
-  };
-}
-function number(value, key) {
-  return typeof value[key] === "number" && Number.isFinite(value[key]) ? value[key] : undefined;
-}
-function string(value, key) {
-  return typeof value[key] === "string" ? value[key] : undefined;
-}
-function ok4(data, result) {
-  return { ok: true, data, stdout: result.stdout, stderr: result.stderr, code: result.code ?? 0 };
-}
-function failure3(result) {
-  return {
-    ok: false,
-    reason: result.reason ?? "nonzero-exit",
-    stdout: result.stdout,
-    stderr: boundDiagnosticText6(result.stderr),
-    code: result.code
-  };
-}
-function boundDiagnosticText6(value) {
-  if (value.length === 0)
-    return "";
-  return value.trim().slice(0, 4000);
-}
-function appendOutput(current, chunk, maxBytes) {
-  if (current.exceeded)
-    return current;
-  const bytes = Buffer.from(chunk, "utf8");
-  const remaining = Math.max(0, maxBytes - current.bytes);
-  const prefix = truncateUtf8(chunk, remaining);
-  const total = current.bytes + bytes.byteLength;
-  return {
-    text: current.text + prefix,
-    bytes: Math.min(total, maxBytes),
-    exceeded: total > maxBytes
-  };
-}
-function isSafeOpaqueId(value) {
-  return value.length > 0 && value !== "." && value !== ".." && !/[\\/#\u0000-\u001f\u007f]/u.test(value);
-}
-function findRemoteEmbeddingViolation(env) {
-  for (const [key, value] of Object.entries(env)) {
-    if (value === undefined)
-      continue;
-    const lower = key.toLowerCase();
-    if (REMOTE_EMBEDDING_ENV_KEYS.has(lower))
-      return key;
-    if (lower === "mailcrawl_embedder" || lower === "mailcrawl_embed_url") {
-      if (isRemoteHttpEndpoint(value))
-        return key;
     }
+    return { ok: true, embedded, skipped, pruned, remaining: pending.length - embedded - skipped };
   }
-  return;
-}
-function isRemoteHttpEndpoint(value) {
-  const trimmed = value.trim();
-  if (!/^https?:\/\//iu.test(trimmed))
-    return false;
-  try {
-    return !LOOPBACK_HOSTS2.has(new URL(trimmed).hostname);
-  } catch {
-    return true;
+  adoptIdentity(identity) {
+    const serialized = JSON.stringify(identity);
+    const current = this.db.prepare("select v from sidecar_meta where k = ?").get(IDENTITY_KEY);
+    if (current?.v === serialized)
+      return;
+    this.transaction(() => {
+      this.db.exec("delete from sidecar_vec");
+      this.db.prepare("insert or replace into sidecar_meta (k, v) values (?, ?)").run(IDENTITY_KEY, serialized);
+    });
+    this.lastSemantic = undefined;
   }
-}
-function truncateUtf8(value, maxBytes) {
-  if (Buffer.byteLength(value) <= maxBytes)
-    return value;
-  const decoder = new TextDecoder("utf-8", { fatal: true });
-  const bytes = Buffer.from(value, "utf8");
-  for (let end = Math.min(maxBytes, bytes.length);end >= 0; end -= 1) {
+  prune() {
+    const { descriptor } = this.options;
+    const result = this.db.prepare(`delete from sidecar_vec where message_key not in (
+					select ${descriptor.keyExpression} from ${descriptor.fromClause} where ${this.where}
+				)`).run(...this.whereParams());
+    if (Number(result.changes) > 0)
+      this.lastSemantic = undefined;
+    return Number(result.changes);
+  }
+  pendingKeys() {
+    const { descriptor } = this.options;
+    const rows = this.db.prepare(`select ${this.select}, v.hash as stored_hash
+				from ${descriptor.fromClause}
+				left join sidecar_vec v on v.message_key = ${descriptor.keyExpression}
+				where ${this.where}`).iterate(...this.whereParams());
+    const pending = [];
+    for (const row of rows) {
+      if (row.stored_hash !== this.toRow(row).hash)
+        pending.push(String(row.k));
+    }
+    return pending;
+  }
+  hydrate(keys) {
+    const { descriptor } = this.options;
+    const out = new Map;
+    if (keys.length === 0)
+      return out;
+    const rows = this.db.prepare(`select ${this.select} from ${descriptor.fromClause}
+				where ${descriptor.keyExpression} in (${keys.map(() => "?").join(", ")}) and ${this.where}`).all(...keys, ...this.whereParams());
+    const byKey = new Map(rows.map((row) => [String(row.k), this.toRow(row)]));
+    for (const key of keys) {
+      const row = byKey.get(key);
+      if (row !== undefined)
+        out.set(key, row);
+    }
+    return out;
+  }
+  store(rows, vectors) {
+    const upsert = this.db.prepare("insert or replace into sidecar_vec (message_key, hash, vec) values (?, ?, ?)");
+    this.transaction(() => {
+      for (const [index, row] of rows.entries()) {
+        upsert.run(row.key, row.hash, encodeVector(vectors?.[index] ?? []));
+      }
+    });
+    this.lastSemantic = undefined;
+  }
+  *storedVectors() {
+    const rows = this.db.prepare("select message_key, vec from sidecar_vec where length(vec) > 0").iterate();
+    for (const row of rows) {
+      const vector = row.vec instanceof Uint8Array ? decodeVector(row.vec) : undefined;
+      if (vector !== undefined)
+        yield { key: String(row.message_key), vector };
+    }
+  }
+  transaction(work) {
+    this.db.exec("begin immediate");
     try {
-      return decoder.decode(bytes.subarray(0, end));
-    } catch {}
+      work();
+      this.db.exec("commit");
+    } catch (error) {
+      this.db.exec("rollback");
+      throw error;
+    }
+  }
+  toRow(row) {
+    const metadata = {};
+    for (const [index, column] of this.options.descriptor.metadata.entries()) {
+      const value = row[`m${index}`];
+      if (typeof value === "string" && value.length > 0 || typeof value === "number")
+        metadata[column.key] = String(value);
+    }
+    const title = typeof row.title === "string" && row.title.length > 0 ? row.title : undefined;
+    const epoch = typeof row.epoch === "number" && Number.isFinite(row.epoch) ? row.epoch : undefined;
+    const content = String(row.content ?? "");
+    const input = `${title ?? ""}
+${content}`.slice(0, MAX_EMBED_INPUT_CHARS);
+    return {
+      key: String(row.k),
+      content,
+      title,
+      epoch,
+      metadata,
+      input,
+      hash: createHash14("sha1").update(input).digest("hex")
+    };
+  }
+  toHit(row, score) {
+    return {
+      key: row.key,
+      content: row.content,
+      score,
+      ...row.title !== undefined ? { title: row.title } : {},
+      ...row.epoch !== undefined ? { publishedAt: Math.round(row.epoch * 1000) } : {},
+      metadata: row.metadata
+    };
   }
-  return "";
 }
-var DEFAULT_BINARY4 = "mailcrawl", DEFAULT_SEARCH_TIMEOUT_MS2 = 60000, DEFAULT_INDEX_TIMEOUT_MS3 = 1800000, DEFAULT_MAX_BUFFER_BYTES4 = 1048576, LOOPBACK_HOSTS2, SAFE_ENV_KEYS2, SAFE_MAILCRAWL_ENV_KEYS, REMOTE_EMBEDDING_ENV_KEYS;
-var init_client9 = __esm(() => {
-  init_portable_spawn();
-  init_config3();
-  LOOPBACK_HOSTS2 = new Set(["127.0.0.1", "localhost", "[::1]", "::1"]);
-  SAFE_ENV_KEYS2 = new Set(["HOME", "LANG", "LC_ALL", "PATH", "TMPDIR", "TMP", "TEMP", "NO_COLOR"]);
-  SAFE_MAILCRAWL_ENV_KEYS = new Set([
-    "MAILCRAWL_DATA_DIR",
-    "MAILCRAWL_EMBEDDER",
-    "MAILCRAWL_EMBEDDER_PROVIDER",
-    "MAILCRAWL_EMBED_URL",
-    "MAILCRAWL_EMBED_MODEL",
-    "MAILCRAWL_EMBED_DIM",
-    "MAILCRAWL_EMBED_TIMEOUT",
-    "MAILCRAWL_EMBED_BATCH_SIZE",
-    "MAILCRAWL_QUERY_PREFIX",
-    "MAILCRAWL_PASSAGE_PREFIX",
-    "MAILCRAWL_NATIVE_MODEL",
-    "MAILCRAWL_NATIVE_DEVICE",
-    "MAILCRAWL_NATIVE_DTYPE",
-    "MAILCRAWL_KIWI_MODEL",
-    "MAILCRAWL_JA_HELPER",
-    "MAILCRAWL_ZH_HELPER"
-  ]);
-  REMOTE_EMBEDDING_ENV_KEYS = new Set([
-    "mailcrawl_embedder_base_url",
-    "mailcrawl_embedding_base_url",
-    "mailcrawl_allow_remote_embeddings"
-  ]);
-});
+function toMatchQuery(text) {
+  const terms = [...new Set(text.match(/[\p{L}\p{N}_]+/gu) ?? [])];
+  return terms.length === 0 ? undefined : terms.map((term) => `"${term}"`).join(" OR ");
+}
+function errorMessage2(error) {
+  return error instanceof Error ? error.message : String(error);
+}
+var DEFAULT_BATCH_SIZE = 256, DEFAULT_MAX_EMBED_PER_SYNC = 2048, MAX_EMBED_INPUT_CHARS = 8192, BUSY_TIMEOUT_MS = 5000, POOL_FACTOR = 3, SEMANTIC_REUSE_MS = 1e4, IDENTITY_KEY = "identity";
+var init_sidecar = () => {};
 
-// src/datasource/skills/mailcrawl/methods.ts
-class MailcrawlMethod {
-  client;
-  mode;
-  instanceId;
-  tags;
-  account;
-  mailbox;
-  constructor(client, mode, instanceId, tags, account, mailbox) {
-    this.client = client;
-    this.mode = mode;
-    this.instanceId = instanceId;
-    this.tags = tags;
-    this.account = account;
-    this.mailbox = mailbox;
+// src/datasource/sidecar/methods.ts
+function sidecarMethodName(datasourceId, mode) {
+  return mode === "bm25" ? `${datasourceId}-sidecar-bm25` : `${datasourceId}-${mode}`;
+}
+
+class SidecarRetrievalMethod {
+  options;
+  constructor(options) {
+    this.options = options;
   }
   describe() {
+    const { datasourceId, skillType, mode, tags } = this.options;
+    const { type, text, capabilities } = DESCRIPTIONS[mode];
     return {
-      name: `mailcrawl-${this.mode}`,
-      type: this.mode === "bm25" ? "bm25" : this.mode === "semantic" ? "vector" : "hybrid",
-      description: `${this.mode} retrieval over archived email via the external mailcrawl CLI`,
+      name: sidecarMethodName(datasourceId, mode),
+      type,
+      description: `${text} over ${skillType} via a read-only sidecar index`,
       status: "active",
-      capabilities: ["email", this.mode, "scoped", "external-cli", "path-opaque-sources"],
-      datasourceId: "mailcrawl",
-      tags: this.tags
+      capabilities: [...capabilities, "scoped", "path-opaque-sources"],
+      datasourceId,
+      tags
+    };
+  }
+  async retrieve(query, options) {
+    const trimmed = query.trim();
+    if (trimmed.length === 0)
+      return [];
+    const topK = options.topK ?? DEFAULT_TOP_K5;
+    const { sidecar, mode } = this.options;
+    const hits = mode === "bm25" ? sidecar.searchLexical(trimmed, topK) : mode === "semantic" ? await sidecar.searchSemantic(trimmed, topK) : await sidecar.searchHybrid(trimmed, topK);
+    const results = [];
+    for (const hit of hits) {
+      const result = this.toResult(hit, options);
+      if (result !== undefined)
+        results.push(result);
+    }
+    return results;
+  }
+  toResult(hit, options) {
+    const { descriptor, datasourceId, instanceId, mode } = this.options;
+    const chunkId = descriptor.chunkId(hit.key, hit.metadata);
+    const source = datasourceSourcePath(datasourceId, instanceId, chunkId);
+    if (!matchesDatasourceScope(source, options.scope))
+      return;
+    const method = sidecarMethodName(datasourceId, mode);
+    return {
+      id: `${datasourceId}:${instanceId}:${chunkId}`,
+      content: hit.content,
+      source,
+      score: hit.score,
+      metadata: {
+        ...hit.metadata,
+        method,
+        datasourceId,
+        instanceId,
+        backend: "sidecar",
+        ...hit.title !== undefined ? { title: hit.title } : {},
+        hierarchy: descriptor.hierarchy(hit.metadata).join("/"),
+        ...hit.publishedAt !== undefined ? { publishedAt: hit.publishedAt } : {}
+      }
     };
   }
-  async retrieve(query, options) {
-    if (!query.trim())
-      return [];
-    const result = await this.client.search(this.mode, query, {
-      topK: options.topK,
-      signal: options.signal
-    });
-    if (!result.ok)
-      throw datasourceCliError("mailcrawl", `search --mode ${this.mode}`, result);
-    return result.hits.filter((hit) => {
-      if (this.account !== undefined && hit.accountId !== this.account)
-        return false;
-      if (this.mailbox !== undefined && hit.mailbox !== this.mailbox)
-        return false;
-      const source = datasourceSourcePath("mailcrawl", this.instanceId, hit.chunkId);
-      return matchesDatasourceScope(source, options.scope);
-    }).map((hit) => {
-      const source = datasourceSourcePath("mailcrawl", this.instanceId, hit.chunkId);
-      return {
-        id: `mailcrawl:${this.instanceId}:${hit.chunkId}`,
-        content: hit.snippet,
-        source,
-        score: hit.score,
-        metadata: {
-          method: `mailcrawl-${this.mode}`,
-          datasourceId: "mailcrawl",
-          instanceId: this.instanceId,
-          mode: this.mode,
-          chunkId: hit.chunkId,
-          messageId: hit.messageId,
-          threadId: hit.threadId,
-          accountId: hit.accountId,
-          mailbox: hit.mailbox,
-          subject: hit.subject,
-          from: hit.from,
-          to: hit.to,
-          date: hit.date
-        }
-      };
-    }).slice(0, options.topK ?? 20);
-  }
 }
-var modes, MAILCRAWL_MODES;
+var DEFAULT_TOP_K5 = 20, DESCRIPTIONS;
 var init_methods4 = __esm(() => {
-  init_errors4();
   init_scope2();
-  modes = ["bm25", "semantic", "hybrid"];
-  MAILCRAWL_MODES = modes;
+  DESCRIPTIONS = {
+    bm25: { type: "bm25", text: "BM25-ranked FTS5 retrieval", capabilities: ["lexical", "fts5", "bm25"] },
+    semantic: { type: "vector", text: "Semantic (embedding) retrieval", capabilities: ["semantic"] },
+    hybrid: {
+      type: "hybrid",
+      text: "Hybrid BM25 + semantic retrieval (reciprocal rank fusion)",
+      capabilities: ["lexical", "semantic", "hybrid"]
+    }
+  };
 });
 
-// src/datasource/skills/mailcrawl/skill.ts
-class MailcrawlSkill {
+// src/datasource/skills/lazykatok/skill.ts
+class LazykatokSkill {
   client;
   instanceId;
   instances;
-  interval;
+  pollingIntervalMs;
   tags;
-  account;
-  mailbox;
+  sidecar;
   lastIndexedAt;
-  constructor(options = {}) {
-    const connectorOptions = options.connectorOptions ?? options;
-    this.instanceId = options.instanceId ?? "default";
-    this.instances = options.instances?.length ? options.instances : [this.instanceId];
-    validateMailcrawlInstanceId(this.instanceId);
-    for (const instance of this.instances)
-      validateMailcrawlInstanceId(instance);
-    this.interval = options.pollingIntervalMs ?? 15 * 60 * 1000;
-    this.tags = options.tags ?? TAGS;
-    this.account = connectorOptions.account;
-    this.mailbox = connectorOptions.mailbox;
-    this.client = options.client ?? new MailcrawlClient({ ...connectorOptions, instanceId: this.instanceId });
+  constructor(options) {
+    this.client = options.client;
+    this.instanceId = options.instanceId ?? DEFAULT_INSTANCE_ID4;
+    this.instances = options.instances !== undefined && options.instances.length > 0 ? options.instances : [this.instanceId];
+    this.pollingIntervalMs = options.pollingIntervalMs ?? DEFAULT_POLLING_INTERVAL_MS3;
+    this.tags = options.tags ?? DEFAULT_KAKAO_TAGS2;
+    this.lastIndexedAt = options.lastIndexedAt;
+    this.sidecar = options.sidecar === undefined ? undefined : new CrawlerSidecar({
+      ...options.sidecar,
+      channelIds: options.channelIds ?? options.sidecar.channelIds ?? [],
+      channelNames: options.channelNames ?? options.sidecar.channelNames ?? []
+    });
   }
   describe() {
     return {
-      name: "mailcrawl",
-      id: "mailcrawl",
-      type: "mailcrawl-archive",
-      description: "Local email archive datasource via the external mailcrawl CLI",
-      capabilities: ["email", "external-cli", "polling", "incremental", "bm25", "semantic", "hybrid"],
+      name: KAKAO_DATASOURCE_ID2,
+      id: KAKAO_DATASOURCE_ID2,
+      type: KAKAO_SKILL_TYPE,
+      description: "KakaoTalk datasource via the external lazykatok CLI",
+      capabilities: ["chat", "external-cli", "polling", "bm25", "semantic"],
       tags: this.tags,
       status: "active",
       requiresExternalCli: true,
-      datasourceId: "mailcrawl",
+      datasourceId: KAKAO_DATASOURCE_ID2,
       instanceId: this.instanceId,
       instances: this.instances
     };
   }
   polling() {
     return {
-      mode: this.interval > 0 ? "poll" : "none",
-      ...this.interval > 0 ? { intervalMs: this.interval } : {},
+      mode: "poll",
+      intervalMs: this.pollingIntervalMs,
       lastIndexedAt: this.lastIndexedAt
     };
   }
   async index(options) {
     try {
-      const result = await this.client.sync(options?.signal);
-      if (!result.ok) {
-        return this.fail(result.reason === "binary-missing" || result.reason === "spawn-error" || result.reason === "timeout" ? "datasource-unavailable" : result.reason === "remote-embedding-rejected" ? "datasource-embedding-egress-rejected" : "datasource-index-failed", result.reason);
-      }
-      const indexed = await this.client.index(options?.signal);
-      const diagnostics = [];
-      if (!indexed.ok) {
-        diagnostics.push({
-          code: indexed.reason === "remote-embedding-rejected" ? "datasource-embedding-egress-rejected" : "datasource-index-failed",
-          severity: "warning",
-          message: `mailcrawl semantic index unavailable (${indexed.reason}); BM25 remains available`,
+      const doctor = await this.client.doctor(options?.signal);
+      if (!doctor.ok)
+        return this.fail("datasource-unavailable", doctor);
+      const sync = await this.client.sync(options?.signal);
+      if (!sync.ok)
+        return this.fail("datasource-index-failed", sync);
+      if (this.sidecar !== undefined) {
+        this.lastIndexedAt = Date.now();
+        return {
+          ok: true,
           instanceId: this.instanceId,
-          source: "mailcrawl"
-        });
+          skill: KAKAO_DATASOURCE_ID2,
+          chunkCount: syncChunkCount(sync),
+          indexedAt: this.lastIndexedAt,
+          diagnostics: await this.syncSidecar()
+        };
       }
+      const indexResult = await this.client.index(options?.signal);
+      if (!indexResult.ok)
+        return this.fail("datasource-index-failed", indexResult);
       this.lastIndexedAt = Date.now();
+      const chunkCount = "data" in indexResult && typeof indexResult.data === "object" && indexResult.data !== null && "chunkCount" in indexResult.data && typeof indexResult.data.chunkCount === "number" ? indexResult.data.chunkCount : 0;
       return {
         ok: true,
         instanceId: this.instanceId,
-        skill: "mailcrawl",
-        chunkCount: result.data.chunksAdded ?? result.data.messages,
+        skill: KAKAO_DATASOURCE_ID2,
+        chunkCount,
         indexedAt: this.lastIndexedAt,
-        diagnostics
+        diagnostics: []
       };
-    } catch {
-      return this.fail("datasource-unavailable", "mailcrawl sync failed unexpectedly");
+    } catch (error) {
+      return this.fail("datasource-unavailable", {
+        ok: false,
+        reason: "spawn-error",
+        stdout: "",
+        stderr: describeRetrievalError(error),
+        code: null
+      });
     }
   }
+  close() {
+    this.sidecar?.close();
+  }
   retrievalMethods() {
+    const cliBm25 = new LazykatokBm25Method({ client: this.client, instanceId: this.instanceId, tags: this.tags });
+    const { sidecar } = this;
+    if (sidecar === undefined) {
+      return [
+        cliBm25,
+        new LazykatokSemanticMethod({ client: this.client, instanceId: this.instanceId, tags: this.tags })
+      ];
+    }
+    const shared = {
+      sidecar,
+      descriptor: sidecar.descriptor,
+      datasourceId: KAKAO_DATASOURCE_ID2,
+      skillType: KAKAO_SKILL_TYPE,
+      instanceId: this.instanceId,
+      tags: this.tags
+    };
+    const bm25 = new SidecarRetrievalMethod({ ...shared, mode: "bm25" });
+    if (!sidecar.semanticEnabled)
+      return [cliBm25, bm25];
     return [
-      new MailcrawlMethod(this.client, "bm25", this.instanceId, this.tags, this.account, this.mailbox),
-      new MailcrawlMethod(this.client, "semantic", this.instanceId, this.tags, this.account, this.mailbox),
-      new MailcrawlMethod(this.client, "hybrid", this.instanceId, this.tags, this.account, this.mailbox)
+      cliBm25,
+      bm25,
+      new SidecarRetrievalMethod({ ...shared, mode: "semantic" }),
+      new SidecarRetrievalMethod({ ...shared, mode: "hybrid" })
     ];
   }
+  async syncSidecar() {
+    if (this.sidecar === undefined || !this.sidecar.semanticEnabled)
+      return [];
+    const base = { instanceId: this.instanceId, source: KAKAO_DATASOURCE_ID2 };
+    let result;
+    try {
+      result = await this.sidecar.sync();
+    } catch (error) {
+      result = { ok: false, message: error instanceof Error ? error.message : String(error) };
+    }
+    if (!result.ok) {
+      return [
+        {
+          code: "datasource-index-failed",
+          severity: "warning",
+          message: `semantic-unavailable: ${result.message}`.slice(0, 300),
+          ...base
+        }
+      ];
+    }
+    const diagnostics = [];
+    if (result.skipped > 0) {
+      diagnostics.push({
+        code: "datasource-index-partial",
+        severity: "warning",
+        message: `semantic index skipped ${result.skipped} message(s) the embedder rejected; they stay searchable by BM25 and are retried when edited`,
+        ...base
+      });
+    }
+    if (result.remaining > 0) {
+      diagnostics.push({
+        code: "datasource-index-partial",
+        severity: "info",
+        message: `semantic index catching up: ${result.embedded} embedded this refresh, ${result.remaining} remaining; hybrid search uses BM25 for the rest until a later refresh`,
+        ...base
+      });
+    }
+    return diagnostics;
+  }
   describeSources() {
     return this.instances.map((instanceId) => ({
-      source: datasourceSourcePath("mailcrawl", instanceId),
-      datasourceId: "mailcrawl",
-      skill: "mailcrawl",
+      source: datasourceSourcePath(KAKAO_DATASOURCE_ID2, instanceId),
+      datasourceId: KAKAO_DATASOURCE_ID2,
+      skill: KAKAO_DATASOURCE_ID2,
       instanceId,
-      contentType: "email",
-      metadata: { datasourceId: "mailcrawl", instanceId, tags: this.tags }
-    }));
-  }
-  skillManifest() {
-    const scopes = this.instances.map((id) => `- \`${datasourceSourcePath("mailcrawl", id)}\``).join(`
-`);
-    return {
-      name: "datasource-mailcrawl",
-      description: "Search local email synchronized and indexed by mailcrawl.",
-      content: `# mailcrawl email datasource
-
-Use this skill for authorized local email retrieval. AutoRAG invokes the external \`mailcrawl\` CLI and never opens its SQLite archive directly.
-
-## Indexing
-Indexing is server-managed and incremental; do not trigger sync yourself.
-
-## Search
-Call the dedicated \`${datasourceSearchToolName("mailcrawl")}\` tool with a natural-language query. Available scopes:
-${scopes}
-
-The selectable retrieval modes are BM25, semantic, and hybrid. Scope arguments can only narrow trusted access. Source identifiers remain opaque and hierarchical.
-
-## Native CLI
-The external \`mailcrawl\` CLI owns the archive and index. When the dedicated tool cannot express what you need, call mailcrawl directly through \`bash\`:
-- \`mailcrawl search --mode hybrid --limit 20 --json "<query>"\` (\`bm25\` and \`semantic\` modes also available; \`--account\` / \`--mailbox\` narrow the archive)
-Never pass datasource virtual paths (\`/mailcrawl/...\`) to bash; they are not OS paths.`
+      contentType: "chat",
+      metadata: {
+        datasourceId: KAKAO_DATASOURCE_ID2,
+        instanceId,
+        tags: this.tags
+      }
+    }));
+  }
+  skillManifest() {
+    const instanceSources = this.instances.map((instanceId) => `- \`${datasourceSourcePath(KAKAO_DATASOURCE_ID2, instanceId)}\``).join(`
+`);
+    const cadence = this.pollingIntervalMs > 0 ? `roughly every ${Math.round(this.pollingIntervalMs / 60000)} minute(s) when auto-refresh runs` : "on manual refresh only";
+    return {
+      name: `datasource-${KAKAO_DATASOURCE_ID2}`,
+      description: "Search indexed KakaoTalk chat history (messages, senders, room context). Use for questions about KakaoTalk conversations, decisions made in chats, or who said what.",
+      content: [
+        `# KakaoTalk datasource (${KAKAO_SKILL_TYPE})`,
+        "",
+        "This skill searches KakaoTalk chats that are indexed through the external `lazykatok` CLI. AutoRAG never reads KakaoTalk databases directly.",
+        "",
+        "## When to use",
+        "Use this skill when the question is about KakaoTalk conversations, chat participants, or content shared inside chats.",
+        "",
+        "## Indexing",
+        `Indexing is server-managed and refreshed ${cadence}. You do not trigger indexing; just search.`,
+        "",
+        "## How to search",
+        `Call the dedicated \`${datasourceSearchToolName(KAKAO_DATASOURCE_ID2)}\` tool with a natural-language \`query\` and \`topK\`. This datasource does not support per-source scope narrowing. Configured datasource instances:`,
+        instanceSources.length > 0 ? instanceSources : "- (no configured instances)",
+        "",
+        "Chat/channel filtering is owned by lazykatok.",
+        "",
+        "## Native CLI",
+        "The external `lazykatok` CLI owns the archive, index, and credentials. When the dedicated tool cannot express what you need, call lazykatok directly through `bash`:",
+        '- `lazykatok search bm25 "<query>" --json --limit 20` — lexical search (`semantic` mode also available)',
+        "- `lazykatok chunk get <chunkId> --json` — fetch one chunk by id (chunk ids appear in result metadata)",
+        "- `lazykatok chunk context <chunkId> --json` — surrounding messages of a chunk",
+        "Never pass datasource virtual paths (`/kakao/...`) to bash; they are not OS paths.",
+        "",
+        "## Output rules",
+        "Datasource source identifiers such as `/kakao/<instance>/chunks/<id>` are internal and opaque. Never put them, real file paths, account IDs, or phone numbers in the visible answer."
+      ].join(`
+`)
     };
   }
-  fail(code, message) {
+  fail(code, result) {
+    const base = result.stderr.length > 0 ? `${result.reason}: ${result.stderr.trim().slice(0, 4000)}` : result.reason;
+    const message = code === "datasource-index-failed" && result.reason === "timeout" ? `${base} — ${connectorSyncTimeoutHint()}` : base;
+    const diagnostic = {
+      code,
+      severity: code === "datasource-unavailable" ? "warning" : "error",
+      message,
+      instanceId: this.instanceId,
+      source: KAKAO_DATASOURCE_ID2
+    };
     return {
       ok: false,
       instanceId: this.instanceId,
-      skill: "mailcrawl",
+      skill: KAKAO_DATASOURCE_ID2,
       indexedAt: Date.now(),
-      diagnostics: [
-        {
-          code,
-          severity: code === "datasource-unavailable" ? "warning" : "error",
-          message: `mailcrawl ${message}`,
-          instanceId: this.instanceId,
-          source: "mailcrawl"
-        }
-      ],
-      error: message,
+      diagnostics: [diagnostic],
+      error: result.reason,
       code,
       message
     };
   }
 }
-var TAGS;
-var init_skill9 = __esm(() => {
+function syncChunkCount(sync) {
+  if (!("data" in sync) || typeof sync.data !== "object" || sync.data === null)
+    return 0;
+  const meta = "metadata" in sync.data && typeof sync.data.metadata === "object" && sync.data.metadata !== null ? sync.data.metadata : undefined;
+  if (typeof meta?.chunks === "number")
+    return meta.chunks;
+  if ("messageCount" in sync.data && typeof sync.data.messageCount === "number")
+    return sync.data.messageCount;
+  return 0;
+}
+var KAKAO_DATASOURCE_ID2 = "kakao", KAKAO_SKILL_TYPE = "kakaotalk", DEFAULT_INSTANCE_ID4 = "default", DEFAULT_POLLING_INTERVAL_MS3, DEFAULT_KAKAO_TAGS2;
+var init_skill7 = __esm(() => {
+  init_skip();
   init_scope2();
-  init_client9();
-  init_config3();
+  init_sidecar();
   init_methods4();
-  TAGS = ["mailcrawl", "email", "pii"];
+  init_methods3();
+  DEFAULT_POLLING_INTERVAL_MS3 = 15 * 60 * 1000;
+  DEFAULT_KAKAO_TAGS2 = ["kakaotalk", "personal", "pii"];
 });
 
-// src/datasource/skills/mailcrawl/index.ts
-var init_mailcrawl = __esm(() => {
-  init_client9();
-  init_methods4();
-  init_skill9();
+// src/datasource/skills/lazykatok/index.ts
+var init_lazykatok = __esm(() => {
+  init_client8();
+  init_methods3();
+  init_skill7();
+  init_types5();
+});
+
+// src/datasource/skills/mail-export/connector.ts
+import { readdir as readdir4, readFile as readFile5, stat as stat6 } from "node:fs/promises";
+import { basename as basename12, dirname as dirname25, extname as extname6, join as join37 } from "node:path";
+import { simpleParser as simpleParser2 } from "mailparser";
+
+class MailExportConnector {
+  options;
+  constructor(options = {}) {
+    this.options = options;
+  }
+  async fetch() {
+    const paths = this.options.paths ?? [];
+    if (paths.length === 0)
+      return { ok: false, reason: "not-configured", message: "no mail export paths configured" };
+    const maxDocuments = this.options.maxDocuments ?? DEFAULT_MAX_DOCUMENTS3;
+    const maxBytes = this.options.maxBytesPerFile ?? DEFAULT_MAX_BYTES_PER_FILE2;
+    const files = [];
+    let missingPaths = 0;
+    for (const path of paths) {
+      try {
+        const info = await stat6(path);
+        if (info.isDirectory()) {
+          const entries = await readdir4(path, { recursive: true, withFileTypes: true });
+          for (const entry of entries) {
+            if (!entry.isFile())
+              continue;
+            const extension = extname6(entry.name).toLowerCase();
+            if (extension === ".eml" || extension === ".mbox") {
+              files.push(join37(entry.parentPath, entry.name));
+            }
+          }
+        } else if (info.isFile()) {
+          files.push(path);
+        }
+      } catch {
+        missingPaths += 1;
+      }
+    }
+    if (missingPaths === paths.length) {
+      return { ok: false, reason: "unavailable", message: "no configured mail export paths were readable" };
+    }
+    const documents = [];
+    const warnings = [];
+    if (missingPaths > 0)
+      warnings.push(`${missingPaths} configured path(s) were unreadable`);
+    let parseFailures = 0;
+    let skippedLarge = 0;
+    for (const file of files.sort()) {
+      if (documents.length >= maxDocuments)
+        break;
+      let bytes;
+      try {
+        const info = await stat6(file);
+        if (info.size > maxBytes) {
+          skippedLarge += 1;
+          continue;
+        }
+        bytes = await readFile5(file);
+      } catch {
+        parseFailures += 1;
+        continue;
+      }
+      const stem = basename12(file, extname6(file));
+      const mailboxDir = basename12(dirname25(file));
+      const hierarchy = ["mailboxes", mailboxDir, stem];
+      const extension = extname6(file).toLowerCase();
+      const rawMessages = extension === ".mbox" ? splitMbox(bytes.toString("utf8")) : [bytes.toString("utf8")];
+      for (const [index, raw] of rawMessages.entries()) {
+        if (documents.length >= maxDocuments)
+          break;
+        try {
+          const parsed = await simpleParser2(raw);
+          const subject = parsed.subject ?? "(no subject)";
+          const from = Array.isArray(parsed.from) ? parsed.from[0]?.text : parsed.from?.text;
+          const to = Array.isArray(parsed.to) ? parsed.to[0]?.text : parsed.to?.text;
+          const body = (parsed.text ?? "").trim().slice(0, MAX_BODY_CHARS);
+          const headerBlock = [
+            `Subject: ${subject}`,
+            ...from !== undefined ? [`From: ${from}`] : [],
+            ...to !== undefined ? [`To: ${to}`] : [],
+            ...parsed.date !== undefined ? [`Date: ${parsed.date.toISOString()}`] : []
+          ].join(`
+`);
+          const content = `${headerBlock}
+
+${body}`.trim();
+          if (content.length === 0)
+            continue;
+          documents.push({
+            docId: extension === ".mbox" ? `${stem}-${index + 1}` : stem,
+            hierarchy,
+            title: subject,
+            content,
+            ...parsed.date !== undefined ? { publishedAt: parsed.date.getTime() } : {},
+            metadata: {
+              path: file,
+              subject,
+              ...from !== undefined ? { from } : {},
+              messageIndex: index
+            }
+          });
+        } catch {
+          parseFailures += 1;
+        }
+      }
+    }
+    if (skippedLarge > 0)
+      warnings.push(`${skippedLarge} file(s) exceeded the size limit and were skipped`);
+    if (parseFailures > 0)
+      warnings.push(`${parseFailures} message(s) failed to parse`);
+    return { ok: true, documents, ...warnings.length > 0 ? { warnings } : {} };
+  }
+}
+function splitMbox(text) {
+  const lines = text.split(/\r?\n/u);
+  const messages = [];
+  let current = [];
+  for (const line of lines) {
+    if (MBOX_SEPARATOR.test(line) && line.startsWith("From ")) {
+      if (current.length > 0)
+        messages.push(current.join(`
+`));
+      current = [];
+      continue;
+    }
+    current.push(line);
+  }
+  if (current.length > 0)
+    messages.push(current.join(`
+`));
+  return messages.filter((message) => message.trim().length > 0);
+}
+var DEFAULT_MAX_DOCUMENTS3 = 500, DEFAULT_MAX_BYTES_PER_FILE2, MAX_BODY_CHARS = 50000, MBOX_SEPARATOR;
+var init_connector4 = __esm(() => {
+  DEFAULT_MAX_BYTES_PER_FILE2 = 50 * 1024 * 1024;
+  MBOX_SEPARATOR = /^From .*\d{4}$/mu;
+});
+
+// src/datasource/skills/mail-export/skill.ts
+var MAIL_EXPORT_SKILL_DEFINITION, MailExportSkill;
+var init_skill8 = __esm(() => {
+  init_connector_skill();
+  init_connector4();
+  MAIL_EXPORT_SKILL_DEFINITION = {
+    skillName: "mail-export",
+    skillType: "mail-export",
+    description: "Local mail export datasource",
+    capabilities: ["email", "filesystem"],
+    defaultTags: ["mail-export", "email", "pii"],
+    contentType: "email",
+    manifestDescription: "Search indexed local mail exports (mbox/eml archives). Use for questions about archived email conversations.",
+    manifestNotes: [
+      "Only the server-configured export paths are indexed; archives are read locally and never uploaded."
+    ]
+  };
+  MailExportSkill = class MailExportSkill extends ConnectorDatasourceSkill {
+    constructor(options = {}) {
+      const { connector, connectorOptions, ...rest } = options;
+      super(MAIL_EXPORT_SKILL_DEFINITION, {
+        ...rest,
+        connector: connector ?? new MailExportConnector(connectorOptions ?? {})
+      });
+    }
+  };
+});
+
+// src/datasource/skills/mail-export/index.ts
+var init_mail_export = __esm(() => {
+  init_connector4();
+  init_skill8();
 });
 
-// src/datasource/sidecar/fusion.ts
-function encodeVector(vector) {
-  const floats = Float32Array.from(vector);
-  return new Uint8Array(floats.buffer, floats.byteOffset, floats.byteLength);
-}
-function decodeVector(blob) {
-  if (blob.byteLength % Float32Array.BYTES_PER_ELEMENT !== 0)
-    return;
-  return new Float32Array(new Uint8Array(blob).buffer);
-}
-function cosine3(a, b) {
-  if (a.length === 0 || a.length !== b.length)
-    return 0;
-  let dot = 0;
-  let normA = 0;
-  let normB = 0;
-  for (let index = 0;index < a.length; index += 1) {
-    const x = a[index] ?? 0;
-    const y = b[index] ?? 0;
-    dot += x * y;
-    normA += x * x;
-    normB += y * y;
-  }
-  if (normA === 0 || normB === 0)
-    return 0;
-  return dot / Math.sqrt(normA * normB);
-}
-function topKByCosine(query, rows, k) {
-  if (k <= 0)
-    return [];
-  const scored = [];
-  for (const row of rows) {
-    const score = cosine3(query, row.vector);
-    if (score > 0)
-      scored.push({ key: row.key, score });
+// src/datasource/skills/mailcrawl/config.ts
+import { chmodSync, mkdirSync as mkdirSync23 } from "node:fs";
+function validateMailcrawlInstanceId(instanceId) {
+  if (!/^[A-Za-z0-9][A-Za-z0-9._-]{0,127}$/u.test(instanceId)) {
+    throw new Error("mailcrawl instanceId must be a safe single path segment");
   }
-  scored.sort((a, b) => b.score - a.score || a.key.localeCompare(b.key));
-  return scored.slice(0, k);
 }
-function reciprocalRankFusion(lists) {
-  const scores = new Map;
-  for (const list of lists) {
-    const seen = new Set;
-    for (const [index, key] of list.entries()) {
-      if (seen.has(key))
-        continue;
-      seen.add(key);
-      scores.set(key, (scores.get(key) ?? 0) + 1 / (RRF_K2 + index + 1));
-    }
-  }
-  return [...scores.entries()].map(([key, score]) => ({ key, score })).sort((a, b) => b.score - a.score);
+function ensurePrivateMailcrawlDataDir(path) {
+  mkdirSync23(path, { recursive: true, mode: 448 });
+  if (process.platform !== "win32")
+    chmodSync(path, 448);
 }
-var RRF_K2 = 60;
+var init_config3 = () => {};
 
-// src/datasource/sidecar/index.ts
-import { createHash as createHash14 } from "node:crypto";
-import { mkdirSync as mkdirSync23, statSync as statSync3 } from "node:fs";
-import { dirname as dirname25 } from "node:path";
-import { DatabaseSync as DatabaseSync2 } from "node:sqlite";
-import { pathToFileURL as pathToFileURL2 } from "node:url";
+// src/datasource/skills/mailcrawl/client.ts
+import { spawn as spawn15 } from "node:child_process";
 
-class CrawlerSidecar {
-  descriptor;
+class MailcrawlClient {
   options;
-  select;
-  where;
-  opened;
-  attached;
-  lastSemantic;
-  constructor(options) {
+  constructor(options = {}) {
     this.options = options;
-    const { descriptor } = options;
-    this.descriptor = descriptor;
-    const metadataColumns = descriptor.metadata.map((column, index) => `${column.sql} as m${index}`);
-    this.select = [
-      `${descriptor.keyExpression} as k`,
-      `${descriptor.textExpression} as content`,
-      `${descriptor.titleExpression ?? "null"} as title`,
-      `${descriptor.epochSecondsExpression ?? "null"} as epoch`,
-      ...metadataColumns
-    ].join(", ");
-    const scope = descriptor.scopeExpression !== undefined && options.workspaceId !== undefined;
-    const allowlist = [
-      ...(options.channelIds?.length ?? 0) > 0 ? [`${descriptor.channelIdExpression} in (${options.channelIds?.map(() => "?").join(", ")})`] : [],
-      ...(options.channelNames?.length ?? 0) > 0 ? [`${descriptor.channelNameExpression} in (${options.channelNames?.map(() => "?").join(", ")})`] : []
-    ];
-    this.where = [
-      descriptor.liveWhere,
-      ...scope ? [`${descriptor.scopeExpression} = ?`] : [],
-      ...allowlist.length > 0 ? [`(${allowlist.join(" or ")})`] : []
-    ].join(" and ");
-  }
-  get db() {
-    if (this.opened !== undefined)
-      return this.opened;
-    mkdirSync23(dirname25(this.options.sidecarDbPath), { recursive: true });
-    const db = new DatabaseSync2(this.options.sidecarDbPath);
-    db.exec(`pragma busy_timeout = ${BUSY_TIMEOUT_MS}`);
-    db.exec("pragma journal_mode = wal");
-    db.exec(`
-			create table if not exists sidecar_meta (k text primary key, v text not null);
-			create table if not exists sidecar_vec (message_key text primary key, hash text not null, vec blob not null);
-		`);
-    this.opened = db;
-    return db;
-  }
-  attachCrawler() {
-    let file;
-    try {
-      const { dev, ino } = statSync3(this.options.crawlerDbPath);
-      file = { dev, ino };
-    } catch {
-      this.detachCrawler();
-      return false;
-    }
-    if (this.attached !== undefined) {
-      if (this.attached.dev === file.dev && this.attached.ino === file.ino)
-        return true;
-      this.detachCrawler();
-    }
-    const crawlerUri = `${pathToFileURL2(this.options.crawlerDbPath).href}?mode=ro`.replaceAll("'", "''");
-    try {
-      this.db.exec(`attach database '${crawlerUri}' as crawler`);
-    } catch {
-      return false;
-    }
-    const { descriptor } = this.options;
-    try {
-      this.db.prepare(`select ${this.select} from ${descriptor.fromClause} where ${this.where} limit 0`).all(...this.whereParams());
-      this.db.prepare(`select 1 from ${descriptor.fts.table} limit 0`).all();
-    } catch {
-      this.db.exec("detach database crawler");
-      return false;
-    }
-    this.attached = file;
-    this.lastSemantic = undefined;
-    return true;
-  }
-  detachCrawler() {
-    if (this.attached === undefined)
-      return;
-    this.attached = undefined;
-    this.lastSemantic = undefined;
-    try {
-      this.opened?.exec("detach database crawler");
-    } catch {}
-  }
-  get semanticEnabled() {
-    return this.options.embedder !== undefined;
-  }
-  close() {
-    this.opened?.close();
-    this.opened = undefined;
-    this.attached = undefined;
-    this.lastSemantic = undefined;
-  }
-  whereParams() {
-    const { descriptor, workspaceId, channelIds = [], channelNames = [] } = this.options;
-    return [
-      ...descriptor.scopeExpression !== undefined && workspaceId !== undefined ? [workspaceId] : [],
-      ...channelIds,
-      ...channelNames
-    ];
-  }
-  searchLexical(query, topK) {
-    const match = toMatchQuery(query);
-    if (match === undefined || topK <= 0 || !this.attachCrawler())
-      return [];
-    const { descriptor } = this.options;
-    const rows = this.db.prepare(`select ${this.select}, bm25(${descriptor.fts.table}) as rank
-				from ${descriptor.fromClause} join ${descriptor.fts.table} on ${descriptor.fts.joinOn}
-				where ${descriptor.fts.table} match ? and ${this.where}
-				order by rank limit ?`).all(match, ...this.whereParams(), topK);
-    return rows.map((row) => this.toHit(this.toRow(row), -Number(row.rank)));
-  }
-  async searchSemantic(query, topK) {
-    const trimmed = query.trim();
-    if (this.options.embedder === undefined || trimmed.length === 0 || topK <= 0 || !this.attachCrawler())
-      return [];
-    const ranked = await this.rankSemantic(trimmed, topK * POOL_FACTOR);
-    if (ranked.length === 0)
-      return [];
-    const live = this.hydrate(ranked.map((entry) => entry.key));
-    const hits = [];
-    for (const entry of ranked) {
-      const row = live.get(entry.key);
-      if (row !== undefined)
-        hits.push(this.toHit(row, entry.score));
-      if (hits.length >= topK)
-        break;
-    }
-    return hits;
-  }
-  async rankSemantic(query, pool) {
-    const now = Date.now();
-    let cached = this.lastSemantic;
-    if (cached === undefined || cached.query !== query || now - cached.at > SEMANTIC_REUSE_MS) {
-      const embedder = this.options.embedder;
-      if (embedder === undefined)
-        return [];
-      cached = {
-        query,
-        at: now,
-        vector: embedder.embed([query]).then((vectors) => vectors[0], () => {
-          return;
-        }),
-        ranked: undefined
-      };
-      this.lastSemantic = cached;
-    }
-    const vector = await cached.vector;
-    if (vector === undefined) {
-      if (this.lastSemantic === cached)
-        this.lastSemantic = undefined;
-      return [];
-    }
-    if (cached.ranked === undefined || cached.ranked.pool < pool) {
-      const wide = pool * POOL_FACTOR;
-      cached.ranked = { pool: wide, keys: topKByCosine(vector, this.storedVectors(), wide) };
-    }
-    return cached.ranked.keys.slice(0, pool);
+    if (options.instanceId !== undefined)
+      validateMailcrawlInstanceId(options.instanceId);
   }
-  async searchHybrid(query, topK) {
-    if (topK <= 0)
-      return [];
-    const pool = topK * POOL_FACTOR;
-    const lexical = this.searchLexical(query, pool);
-    const semantic = await this.searchSemantic(query, pool);
-    if (semantic.length === 0)
-      return lexical.slice(0, topK);
-    const byKey = new Map;
-    for (const hit of [...semantic, ...lexical])
-      byKey.set(hit.key, hit);
-    const fused = [];
-    for (const entry of reciprocalRankFusion([lexical.map((hit) => hit.key), semantic.map((hit) => hit.key)])) {
-      const hit = byKey.get(entry.key);
-      if (hit !== undefined)
-        fused.push({ ...hit, score: entry.score });
-      if (fused.length >= topK)
-        break;
+  async sync(signal) {
+    const args = [
+      "sync",
+      "--json",
+      ...this.options.account === undefined ? [] : ["--account", this.options.account],
+      ...this.options.mailbox === undefined ? [] : ["--mailbox", this.options.mailbox],
+      ...this.options.backend === undefined ? [] : ["--backend", this.options.backend],
+      ...this.options.source === undefined ? [] : ["--source", this.options.source],
+      ...this.options.fixture === undefined ? [] : ["--fixture", this.options.fixture],
+      ...this.options.himalayaConfig === undefined ? [] : ["--himalaya-config", this.options.himalayaConfig]
+    ];
+    const result = await this.run(args, signal, this.indexTimeoutMs());
+    if (!result.ok)
+      return failure3(result);
+    const parsed = parseObject2(result.stdout);
+    if (parsed === undefined || !hasAnyNumber(parsed, ["added", "updated", "deleted", "unchanged", "chunksAdded"])) {
+      return failure3({ ...result, ok: false, reason: "invalid-output" });
     }
-    return fused;
+    const messages = number(parsed, "added") ?? number(parsed, "updated") ?? number(parsed, "unchanged") ?? 0;
+    return ok4({
+      messages,
+      added: number(parsed, "added"),
+      updated: number(parsed, "updated"),
+      deleted: number(parsed, "deleted"),
+      unchanged: number(parsed, "unchanged"),
+      chunksAdded: number(parsed, "chunksAdded"),
+      archiveRevision: string(parsed, "archiveRevision")
+    }, result);
   }
-  async sync() {
-    const { embedder } = this.options;
-    if (embedder === undefined)
-      return { ok: false, message: "no embedder configured" };
-    if (!this.attachCrawler())
-      return { ok: false, message: "crawler archive not available yet" };
-    let identity;
-    try {
-      identity = await embedder.identity();
-    } catch (error) {
-      return { ok: false, message: errorMessage2(error) };
+  async search(mode, query, options) {
+    if (query.trim().length === 0)
+      return { ok: true, hits: [], stdout: "", stderr: "", code: 0 };
+    const args = [
+      "search",
+      "--mode",
+      mode === "keyword" ? "bm25" : mode,
+      ...this.options.account === undefined ? [] : ["--account", this.options.account],
+      ...this.options.mailbox === undefined ? [] : ["--mailbox", this.options.mailbox],
+      "--limit",
+      String(options?.topK ?? 20),
+      "--json",
+      query.trim()
+    ];
+    const result = await this.run(args, options?.signal, this.options.timeoutMs ?? DEFAULT_SEARCH_TIMEOUT_MS2);
+    if (!result.ok)
+      return failure3(result);
+    const parsed = parseJson3(result.stdout);
+    if (parsed === undefined)
+      return failure3({ ...result, ok: false, reason: "invalid-output" });
+    const rows = Array.isArray(parsed) ? parsed : objectArray(parsed, ["hits", "results", "items"]);
+    if (rows === undefined)
+      return failure3({ ...result, ok: false, reason: "invalid-output" });
+    const hits = rows.map((row, index) => normalizeHit2(row, mode, index));
+    if (hits.some((hit) => hit === undefined)) {
+      return failure3({ ...result, ok: false, reason: "invalid-output" });
     }
-    this.adoptIdentity(identity);
-    const pruned = this.prune();
-    const pending = this.pendingKeys();
-    const batchSize = Math.max(1, this.options.batchSize ?? DEFAULT_BATCH_SIZE);
-    const budget = Math.max(1, this.options.maxEmbedPerSync ?? DEFAULT_MAX_EMBED_PER_SYNC);
-    const target = Math.min(pending.length, budget);
-    const embed = async (rows) => {
-      const vectors = await embedder.embed(rows.map((row) => row.input));
-      if (vectors.length !== rows.length) {
-        throw new Error(`embedder returned ${vectors.length} vectors for ${rows.length} messages`);
-      }
-      const wrong = vectors.find((vector) => vector.length !== identity.dimension);
-      if (wrong !== undefined) {
-        throw new Error(`embedder returned a ${wrong.length}-dimension vector, expected ${identity.dimension}`);
-      }
-      return vectors;
+    return {
+      ok: true,
+      hits,
+      stdout: result.stdout,
+      stderr: result.stderr,
+      code: result.code ?? 0
     };
-    let embedded = 0;
-    let skipped = 0;
-    for (let offset = 0;offset < target; offset += batchSize) {
-      const rows = [...this.hydrate(pending.slice(offset, Math.min(offset + batchSize, target))).values()];
-      if (rows.length === 0)
-        continue;
-      try {
-        this.store(rows, await embed(rows));
-        embedded += rows.length;
-      } catch (batchError) {
-        const rejected = [];
-        let lastError = batchError;
-        for (const row of rows) {
-          try {
-            this.store([row], await embed([row]));
-            embedded += 1;
-          } catch (error) {
-            rejected.push(row);
-            lastError = error;
-          }
-        }
-        if (rejected.length === 0)
-          continue;
-        if (embedded === 0)
-          return { ok: false, message: errorMessage2(lastError) };
-        this.store(rejected, undefined);
-        skipped += rejected.length;
-      }
-    }
-    return { ok: true, embedded, skipped, pruned, remaining: pending.length - embedded - skipped };
-  }
-  adoptIdentity(identity) {
-    const serialized = JSON.stringify(identity);
-    const current = this.db.prepare("select v from sidecar_meta where k = ?").get(IDENTITY_KEY);
-    if (current?.v === serialized)
-      return;
-    this.transaction(() => {
-      this.db.exec("delete from sidecar_vec");
-      this.db.prepare("insert or replace into sidecar_meta (k, v) values (?, ?)").run(IDENTITY_KEY, serialized);
-    });
-    this.lastSemantic = undefined;
   }
-  prune() {
-    const { descriptor } = this.options;
-    const result = this.db.prepare(`delete from sidecar_vec where message_key not in (
-					select ${descriptor.keyExpression} from ${descriptor.fromClause} where ${this.where}
-				)`).run(...this.whereParams());
-    if (Number(result.changes) > 0)
-      this.lastSemantic = undefined;
-    return Number(result.changes);
+  indexTimeoutMs() {
+    return this.options.indexTimeoutMs ?? this.options.timeoutMs ?? DEFAULT_INDEX_TIMEOUT_MS3;
   }
-  pendingKeys() {
-    const { descriptor } = this.options;
-    const rows = this.db.prepare(`select ${this.select}, v.hash as stored_hash
-				from ${descriptor.fromClause}
-				left join sidecar_vec v on v.message_key = ${descriptor.keyExpression}
-				where ${this.where}`).iterate(...this.whereParams());
-    const pending = [];
-    for (const row of rows) {
-      if (row.stored_hash !== this.toRow(row).hash)
-        pending.push(String(row.k));
+  async run(args, signal, timeoutMs) {
+    const violation = findRemoteEmbeddingViolation({ ...process.env, ...this.options.env ?? {} });
+    if (violation !== undefined) {
+      return { ok: false, reason: "remote-embedding-rejected", stdout: "", stderr: "", code: null };
     }
-    return pending;
+    const dataDir = this.options.dataDir;
+    if (dataDir !== undefined)
+      ensurePrivateMailcrawlDataDir(dataDir);
+    const env = {};
+    for (const [key, value] of Object.entries(process.env)) {
+      if (value !== undefined && (SAFE_ENV_KEYS2.has(key) || SAFE_MAILCRAWL_ENV_KEYS.has(key)))
+        env[key] = value;
+    }
+    for (const [key, value] of Object.entries(this.options.env ?? {})) {
+      if (value === undefined)
+        delete env[key];
+      else if (SAFE_ENV_KEYS2.has(key) || SAFE_MAILCRAWL_ENV_KEYS.has(key))
+        env[key] = value;
+    }
+    if (dataDir !== undefined)
+      env.MAILCRAWL_DATA_DIR = dataDir;
+    return spawnMailcrawl(this.options.binaryPath ?? DEFAULT_BINARY4, args, env, this.options, signal, timeoutMs);
   }
-  hydrate(keys) {
-    const { descriptor } = this.options;
-    const out = new Map;
-    if (keys.length === 0)
-      return out;
-    const rows = this.db.prepare(`select ${this.select} from ${descriptor.fromClause}
-				where ${descriptor.keyExpression} in (${keys.map(() => "?").join(", ")}) and ${this.where}`).all(...keys, ...this.whereParams());
-    const byKey = new Map(rows.map((row) => [String(row.k), this.toRow(row)]));
-    for (const key of keys) {
-      const row = byKey.get(key);
-      if (row !== undefined)
-        out.set(key, row);
+  async index(signal) {
+    const result = await this.run(["index", "--json"], signal, this.indexTimeoutMs());
+    if (!result.ok)
+      return failure3(result);
+    const parsed = parseObject2(result.stdout);
+    if (parsed === undefined || number(parsed, "embedded") === undefined && number(parsed, "reused") === undefined && string(parsed, "generation") === undefined) {
+      return failure3({ ...result, ok: false, reason: "invalid-output" });
     }
-    return out;
+    return ok4({
+      embedded: number(parsed, "embedded"),
+      reused: number(parsed, "reused"),
+      generation: string(parsed, "generation"),
+      archiveRevision: string(parsed, "archiveRevision"),
+      rebuilt: typeof parsed.rebuilt === "boolean" ? parsed.rebuilt : undefined,
+      embedder: string(parsed, "embedder")
+    }, result);
   }
-  store(rows, vectors) {
-    const upsert = this.db.prepare("insert or replace into sidecar_vec (message_key, hash, vec) values (?, ?, ?)");
-    this.transaction(() => {
-      for (const [index, row] of rows.entries()) {
-        upsert.run(row.key, row.hash, encodeVector(vectors?.[index] ?? []));
+}
+function spawnMailcrawl(binary, args, env, options, signal, timeoutMs) {
+  return new Promise((resolveResult) => {
+    const portable = portableSpawnCommand(binary, args);
+    const child = spawn15(portable.command, portable.args, {
+      env,
+      detached: process.platform !== "win32",
+      stdio: ["ignore", "pipe", "pipe"]
+    });
+    let stdout = { text: "", bytes: 0, exceeded: false }, stderr = { text: "", bytes: 0, exceeded: false }, settled = false, reason;
+    const max = options.maxBufferBytes ?? DEFAULT_MAX_BUFFER_BYTES4;
+    const timer = setTimeout(() => {
+      reason = "timeout";
+      terminate4(child);
+    }, timeoutMs);
+    const abort = () => {
+      reason = "aborted";
+      terminate4(child);
+    };
+    if (signal?.aborted)
+      abort();
+    signal?.addEventListener("abort", abort, { once: true });
+    child.stdout.setEncoding("utf8");
+    child.stderr.setEncoding("utf8");
+    child.stdout.on("data", (chunk) => {
+      stdout = appendOutput(stdout, chunk, max);
+      if (stdout.exceeded && reason === undefined) {
+        reason = "stdout-too-large";
+        terminate4(child);
+      }
+    });
+    child.stderr.on("data", (chunk) => {
+      stderr = appendOutput(stderr, chunk, max);
+      if (stderr.exceeded && reason === undefined) {
+        reason = "stderr-too-large";
+        terminate4(child);
       }
     });
-    this.lastSemantic = undefined;
+    child.on("error", (error) => {
+      if (settled)
+        return;
+      settled = true;
+      clearTimeout(timer);
+      signal?.removeEventListener("abort", abort);
+      resolveResult({
+        ok: false,
+        reason: error.code === "ENOENT" ? "binary-missing" : "spawn-error",
+        stdout: stdout.text,
+        stderr: "",
+        code: null
+      });
+    });
+    child.on("close", (code) => {
+      if (settled)
+        return;
+      settled = true;
+      clearTimeout(timer);
+      signal?.removeEventListener("abort", abort);
+      resolveResult({
+        ok: code === 0 && reason === undefined,
+        ...reason === undefined ? code === 0 ? {} : { reason: "nonzero-exit" } : { reason },
+        stdout: stdout.text,
+        stderr: stderr.text,
+        code
+      });
+    });
+  });
+}
+function terminate4(child) {
+  if (child.killed)
+    return;
+  if (process.platform !== "win32" && child.pid !== undefined) {
+    try {
+      process.kill(-child.pid, "SIGTERM");
+      return;
+    } catch {}
   }
-  *storedVectors() {
-    const rows = this.db.prepare("select message_key, vec from sidecar_vec where length(vec) > 0").iterate();
-    for (const row of rows) {
-      const vector = row.vec instanceof Uint8Array ? decodeVector(row.vec) : undefined;
-      if (vector !== undefined)
-        yield { key: String(row.message_key), vector };
+  child.kill("SIGTERM");
+}
+function parseJson3(text) {
+  try {
+    return JSON.parse(text.trim());
+  } catch {
+    return;
+  }
+}
+function parseObject2(text) {
+  const value = parseJson3(text);
+  return value !== null && typeof value === "object" && !Array.isArray(value) ? value : undefined;
+}
+function objectArray(value, keys) {
+  if (value === null || typeof value !== "object" || Array.isArray(value))
+    return;
+  for (const key of keys)
+    if (Array.isArray(value[key]))
+      return value[key];
+  return;
+}
+function hasAnyNumber(value, keys) {
+  return keys.some((key) => number(value, key) !== undefined);
+}
+function normalizeHit2(value, mode, index) {
+  if (value === null || typeof value !== "object" || Array.isArray(value))
+    return;
+  const row = value;
+  const text = (key) => typeof row[key] === "string" ? row[key] : "";
+  const chunkId = text("chunkId") || text("chunk_id");
+  const messageId = text("messageId") || text("message_id");
+  const accountId = text("accountId") || text("account_id");
+  const mailbox = text("mailbox");
+  const snippet = text("snippet") || text("content") || text("text");
+  if (!isSafeOpaqueId(chunkId) || !isSafeOpaqueId(messageId) || !snippet)
+    return;
+  const to = Array.isArray(row.to) ? row.to.filter((item) => typeof item === "string") : [];
+  return {
+    chunkId,
+    messageId,
+    threadId: text("threadId") || text("thread_id"),
+    accountId,
+    mailbox,
+    subject: text("subject"),
+    from: text("from"),
+    to,
+    date: text("date"),
+    snippet,
+    score: typeof row.score === "number" ? row.score : 1 / (index + 1),
+    mode
+  };
+}
+function number(value, key) {
+  return typeof value[key] === "number" && Number.isFinite(value[key]) ? value[key] : undefined;
+}
+function string(value, key) {
+  return typeof value[key] === "string" ? value[key] : undefined;
+}
+function ok4(data, result) {
+  return { ok: true, data, stdout: result.stdout, stderr: result.stderr, code: result.code ?? 0 };
+}
+function failure3(result) {
+  return {
+    ok: false,
+    reason: result.reason ?? "nonzero-exit",
+    stdout: result.stdout,
+    stderr: boundDiagnosticText6(result.stderr),
+    code: result.code
+  };
+}
+function boundDiagnosticText6(value) {
+  if (value.length === 0)
+    return "";
+  return value.trim().slice(0, 4000);
+}
+function appendOutput(current, chunk, maxBytes) {
+  if (current.exceeded)
+    return current;
+  const bytes = Buffer.from(chunk, "utf8");
+  const remaining = Math.max(0, maxBytes - current.bytes);
+  const prefix = truncateUtf8(chunk, remaining);
+  const total = current.bytes + bytes.byteLength;
+  return {
+    text: current.text + prefix,
+    bytes: Math.min(total, maxBytes),
+    exceeded: total > maxBytes
+  };
+}
+function isSafeOpaqueId(value) {
+  return value.length > 0 && value !== "." && value !== ".." && !/[\\/#\u0000-\u001f\u007f]/u.test(value);
+}
+function findRemoteEmbeddingViolation(env) {
+  for (const [key, value] of Object.entries(env)) {
+    if (value === undefined)
+      continue;
+    const lower = key.toLowerCase();
+    if (REMOTE_EMBEDDING_ENV_KEYS.has(lower))
+      return key;
+    if (lower === "mailcrawl_embedder" || lower === "mailcrawl_embed_url") {
+      if (isRemoteHttpEndpoint(value))
+        return key;
     }
   }
-  transaction(work) {
-    this.db.exec("begin immediate");
+  return;
+}
+function isRemoteHttpEndpoint(value) {
+  const trimmed = value.trim();
+  if (!/^https?:\/\//iu.test(trimmed))
+    return false;
+  try {
+    return !LOOPBACK_HOSTS2.has(new URL(trimmed).hostname);
+  } catch {
+    return true;
+  }
+}
+function truncateUtf8(value, maxBytes) {
+  if (Buffer.byteLength(value) <= maxBytes)
+    return value;
+  const decoder = new TextDecoder("utf-8", { fatal: true });
+  const bytes = Buffer.from(value, "utf8");
+  for (let end = Math.min(maxBytes, bytes.length);end >= 0; end -= 1) {
     try {
-      work();
-      this.db.exec("commit");
-    } catch (error) {
-      this.db.exec("rollback");
-      throw error;
-    }
+      return decoder.decode(bytes.subarray(0, end));
+    } catch {}
   }
-  toRow(row) {
-    const metadata = {};
-    for (const [index, column] of this.options.descriptor.metadata.entries()) {
-      const value = row[`m${index}`];
-      if (typeof value === "string" && value.length > 0 || typeof value === "number")
-        metadata[column.key] = String(value);
-    }
-    const title = typeof row.title === "string" && row.title.length > 0 ? row.title : undefined;
-    const epoch = typeof row.epoch === "number" && Number.isFinite(row.epoch) ? row.epoch : undefined;
-    const content = String(row.content ?? "");
-    const input = `${title ?? ""}
-${content}`.slice(0, MAX_EMBED_INPUT_CHARS);
-    return {
-      key: String(row.k),
-      content,
-      title,
-      epoch,
-      metadata,
-      input,
-      hash: createHash14("sha1").update(input).digest("hex")
-    };
+  return "";
+}
+var DEFAULT_BINARY4 = "mailcrawl", DEFAULT_SEARCH_TIMEOUT_MS2 = 60000, DEFAULT_INDEX_TIMEOUT_MS3 = 1800000, DEFAULT_MAX_BUFFER_BYTES4 = 1048576, LOOPBACK_HOSTS2, SAFE_ENV_KEYS2, SAFE_MAILCRAWL_ENV_KEYS, REMOTE_EMBEDDING_ENV_KEYS;
+var init_client9 = __esm(() => {
+  init_portable_spawn();
+  init_config3();
+  LOOPBACK_HOSTS2 = new Set(["127.0.0.1", "localhost", "[::1]", "::1"]);
+  SAFE_ENV_KEYS2 = new Set(["HOME", "LANG", "LC_ALL", "PATH", "TMPDIR", "TMP", "TEMP", "NO_COLOR"]);
+  SAFE_MAILCRAWL_ENV_KEYS = new Set([
+    "MAILCRAWL_DATA_DIR",
+    "MAILCRAWL_EMBEDDER",
+    "MAILCRAWL_EMBEDDER_PROVIDER",
+    "MAILCRAWL_EMBED_URL",
+    "MAILCRAWL_EMBED_MODEL",
+    "MAILCRAWL_EMBED_DIM",
+    "MAILCRAWL_EMBED_TIMEOUT",
+    "MAILCRAWL_EMBED_BATCH_SIZE",
+    "MAILCRAWL_QUERY_PREFIX",
+    "MAILCRAWL_PASSAGE_PREFIX",
+    "MAILCRAWL_NATIVE_MODEL",
+    "MAILCRAWL_NATIVE_DEVICE",
+    "MAILCRAWL_NATIVE_DTYPE",
+    "MAILCRAWL_KIWI_MODEL",
+    "MAILCRAWL_JA_HELPER",
+    "MAILCRAWL_ZH_HELPER"
+  ]);
+  REMOTE_EMBEDDING_ENV_KEYS = new Set([
+    "mailcrawl_embedder_base_url",
+    "mailcrawl_embedding_base_url",
+    "mailcrawl_allow_remote_embeddings"
+  ]);
+});
+
+// src/datasource/skills/mailcrawl/methods.ts
+class MailcrawlMethod {
+  client;
+  mode;
+  instanceId;
+  tags;
+  account;
+  mailbox;
+  constructor(client, mode, instanceId, tags, account, mailbox) {
+    this.client = client;
+    this.mode = mode;
+    this.instanceId = instanceId;
+    this.tags = tags;
+    this.account = account;
+    this.mailbox = mailbox;
   }
-  toHit(row, score) {
+  describe() {
     return {
-      key: row.key,
-      content: row.content,
-      score,
-      ...row.title !== undefined ? { title: row.title } : {},
-      ...row.epoch !== undefined ? { publishedAt: Math.round(row.epoch * 1000) } : {},
-      metadata: row.metadata
+      name: `mailcrawl-${this.mode}`,
+      type: this.mode === "bm25" ? "bm25" : this.mode === "semantic" ? "vector" : "hybrid",
+      description: `${this.mode} retrieval over archived email via the external mailcrawl CLI`,
+      status: "active",
+      capabilities: ["email", this.mode, "scoped", "external-cli", "path-opaque-sources"],
+      datasourceId: "mailcrawl",
+      tags: this.tags
     };
   }
+  async retrieve(query, options) {
+    if (!query.trim())
+      return [];
+    const result = await this.client.search(this.mode, query, {
+      topK: options.topK,
+      signal: options.signal
+    });
+    if (!result.ok)
+      throw datasourceCliError("mailcrawl", `search --mode ${this.mode}`, result);
+    return result.hits.filter((hit) => {
+      if (this.account !== undefined && hit.accountId !== this.account)
+        return false;
+      if (this.mailbox !== undefined && hit.mailbox !== this.mailbox)
+        return false;
+      const source = datasourceSourcePath("mailcrawl", this.instanceId, hit.chunkId);
+      return matchesDatasourceScope(source, options.scope);
+    }).map((hit) => {
+      const source = datasourceSourcePath("mailcrawl", this.instanceId, hit.chunkId);
+      return {
+        id: `mailcrawl:${this.instanceId}:${hit.chunkId}`,
+        content: hit.snippet,
+        source,
+        score: hit.score,
+        metadata: {
+          method: `mailcrawl-${this.mode}`,
+          datasourceId: "mailcrawl",
+          instanceId: this.instanceId,
+          mode: this.mode,
+          chunkId: hit.chunkId,
+          messageId: hit.messageId,
+          threadId: hit.threadId,
+          accountId: hit.accountId,
+          mailbox: hit.mailbox,
+          subject: hit.subject,
+          from: hit.from,
+          to: hit.to,
+          date: hit.date
+        }
+      };
+    }).slice(0, options.topK ?? 20);
+  }
 }
-function toMatchQuery(text) {
-  const terms = [...new Set(text.match(/[\p{L}\p{N}_]+/gu) ?? [])];
-  return terms.length === 0 ? undefined : terms.map((term) => `"${term}"`).join(" OR ");
-}
-function errorMessage2(error) {
-  return error instanceof Error ? error.message : String(error);
-}
-var DEFAULT_BATCH_SIZE = 256, DEFAULT_MAX_EMBED_PER_SYNC = 2048, MAX_EMBED_INPUT_CHARS = 8192, BUSY_TIMEOUT_MS = 5000, POOL_FACTOR = 3, SEMANTIC_REUSE_MS = 1e4, IDENTITY_KEY = "identity";
-var init_sidecar = () => {};
-
-// src/datasource/sidecar/methods.ts
-function sidecarMethodName(datasourceId, mode) {
-  return mode === "bm25" ? `${datasourceId}-sidecar-bm25` : `${datasourceId}-${mode}`;
-}
+var modes, MAILCRAWL_MODES;
+var init_methods5 = __esm(() => {
+  init_errors4();
+  init_scope2();
+  modes = ["bm25", "semantic", "hybrid"];
+  MAILCRAWL_MODES = modes;
+});
 
-class SidecarRetrievalMethod {
-  options;
-  constructor(options) {
-    this.options = options;
+// src/datasource/skills/mailcrawl/skill.ts
+class MailcrawlSkill {
+  client;
+  instanceId;
+  instances;
+  interval;
+  tags;
+  account;
+  mailbox;
+  lastIndexedAt;
+  constructor(options = {}) {
+    const connectorOptions = options.connectorOptions ?? options;
+    this.instanceId = options.instanceId ?? "default";
+    this.instances = options.instances?.length ? options.instances : [this.instanceId];
+    validateMailcrawlInstanceId(this.instanceId);
+    for (const instance of this.instances)
+      validateMailcrawlInstanceId(instance);
+    this.interval = options.pollingIntervalMs ?? 15 * 60 * 1000;
+    this.tags = options.tags ?? TAGS;
+    this.account = connectorOptions.account;
+    this.mailbox = connectorOptions.mailbox;
+    this.client = options.client ?? new MailcrawlClient({ ...connectorOptions, instanceId: this.instanceId });
   }
   describe() {
-    const { datasourceId, skillType, mode, tags } = this.options;
-    const { type, text, capabilities } = DESCRIPTIONS[mode];
     return {
-      name: sidecarMethodName(datasourceId, mode),
-      type,
-      description: `${text} over ${skillType} via a read-only sidecar index`,
+      name: "mailcrawl",
+      id: "mailcrawl",
+      type: "mailcrawl-archive",
+      description: "Local email archive datasource via the external mailcrawl CLI",
+      capabilities: ["email", "external-cli", "polling", "incremental", "bm25", "semantic", "hybrid"],
+      tags: this.tags,
       status: "active",
-      capabilities: [...capabilities, "scoped", "path-opaque-sources"],
-      datasourceId,
-      tags
+      requiresExternalCli: true,
+      datasourceId: "mailcrawl",
+      instanceId: this.instanceId,
+      instances: this.instances
     };
   }
-  async retrieve(query, options) {
-    const trimmed = query.trim();
-    if (trimmed.length === 0)
-      return [];
-    const topK = options.topK ?? DEFAULT_TOP_K5;
-    const { sidecar, mode } = this.options;
-    const hits = mode === "bm25" ? sidecar.searchLexical(trimmed, topK) : mode === "semantic" ? await sidecar.searchSemantic(trimmed, topK) : await sidecar.searchHybrid(trimmed, topK);
-    const results = [];
-    for (const hit of hits) {
-      const result = this.toResult(hit, options);
-      if (result !== undefined)
-        results.push(result);
+  polling() {
+    return {
+      mode: this.interval > 0 ? "poll" : "none",
+      ...this.interval > 0 ? { intervalMs: this.interval } : {},
+      lastIndexedAt: this.lastIndexedAt
+    };
+  }
+  async index(options) {
+    try {
+      const result = await this.client.sync(options?.signal);
+      if (!result.ok) {
+        return this.fail(result.reason === "binary-missing" || result.reason === "spawn-error" || result.reason === "timeout" ? "datasource-unavailable" : result.reason === "remote-embedding-rejected" ? "datasource-embedding-egress-rejected" : "datasource-index-failed", result.reason);
+      }
+      const indexed = await this.client.index(options?.signal);
+      const diagnostics = [];
+      if (!indexed.ok) {
+        diagnostics.push({
+          code: indexed.reason === "remote-embedding-rejected" ? "datasource-embedding-egress-rejected" : "datasource-index-failed",
+          severity: "warning",
+          message: `mailcrawl semantic index unavailable (${indexed.reason}); BM25 remains available`,
+          instanceId: this.instanceId,
+          source: "mailcrawl"
+        });
+      }
+      this.lastIndexedAt = Date.now();
+      return {
+        ok: true,
+        instanceId: this.instanceId,
+        skill: "mailcrawl",
+        chunkCount: result.data.chunksAdded ?? result.data.messages,
+        indexedAt: this.lastIndexedAt,
+        diagnostics
+      };
+    } catch {
+      return this.fail("datasource-unavailable", "mailcrawl sync failed unexpectedly");
     }
-    return results;
   }
-  toResult(hit, options) {
-    const { descriptor, datasourceId, instanceId, mode } = this.options;
-    const chunkId = descriptor.chunkId(hit.key, hit.metadata);
-    const source = datasourceSourcePath(datasourceId, instanceId, chunkId);
-    if (!matchesDatasourceScope(source, options.scope))
-      return;
-    const method = sidecarMethodName(datasourceId, mode);
+  retrievalMethods() {
+    return [
+      new MailcrawlMethod(this.client, "bm25", this.instanceId, this.tags, this.account, this.mailbox),
+      new MailcrawlMethod(this.client, "semantic", this.instanceId, this.tags, this.account, this.mailbox),
+      new MailcrawlMethod(this.client, "hybrid", this.instanceId, this.tags, this.account, this.mailbox)
+    ];
+  }
+  describeSources() {
+    return this.instances.map((instanceId) => ({
+      source: datasourceSourcePath("mailcrawl", instanceId),
+      datasourceId: "mailcrawl",
+      skill: "mailcrawl",
+      instanceId,
+      contentType: "email",
+      metadata: { datasourceId: "mailcrawl", instanceId, tags: this.tags }
+    }));
+  }
+  skillManifest() {
+    const scopes = this.instances.map((id) => `- \`${datasourceSourcePath("mailcrawl", id)}\``).join(`
+`);
     return {
-      id: `${datasourceId}:${instanceId}:${chunkId}`,
-      content: hit.content,
-      source,
-      score: hit.score,
-      metadata: {
-        ...hit.metadata,
-        method,
-        datasourceId,
-        instanceId,
-        backend: "sidecar",
-        ...hit.title !== undefined ? { title: hit.title } : {},
-        hierarchy: descriptor.hierarchy(hit.metadata).join("/"),
-        ...hit.publishedAt !== undefined ? { publishedAt: hit.publishedAt } : {}
-      }
+      name: "datasource-mailcrawl",
+      description: "Search local email synchronized and indexed by mailcrawl.",
+      content: `# mailcrawl email datasource
+
+Use this skill for authorized local email retrieval. AutoRAG invokes the external \`mailcrawl\` CLI and never opens its SQLite archive directly.
+
+## Indexing
+Indexing is server-managed and incremental; do not trigger sync yourself.
+
+## Search
+Call the dedicated \`${datasourceSearchToolName("mailcrawl")}\` tool with a natural-language query. Available scopes:
+${scopes}
+
+The selectable retrieval modes are BM25, semantic, and hybrid. Scope arguments can only narrow trusted access. Source identifiers remain opaque and hierarchical.
+
+## Native CLI
+The external \`mailcrawl\` CLI owns the archive and index. When the dedicated tool cannot express what you need, call mailcrawl directly through \`bash\`:
+- \`mailcrawl search --mode hybrid --limit 20 --json "<query>"\` (\`bm25\` and \`semantic\` modes also available; \`--account\` / \`--mailbox\` narrow the archive)
+Never pass datasource virtual paths (\`/mailcrawl/...\`) to bash; they are not OS paths.`
+    };
+  }
+  fail(code, message) {
+    return {
+      ok: false,
+      instanceId: this.instanceId,
+      skill: "mailcrawl",
+      indexedAt: Date.now(),
+      diagnostics: [
+        {
+          code,
+          severity: code === "datasource-unavailable" ? "warning" : "error",
+          message: `mailcrawl ${message}`,
+          instanceId: this.instanceId,
+          source: "mailcrawl"
+        }
+      ],
+      error: message,
+      code,
+      message
     };
   }
 }
-var DEFAULT_TOP_K5 = 20, DESCRIPTIONS;
-var init_methods5 = __esm(() => {
+var TAGS;
+var init_skill9 = __esm(() => {
   init_scope2();
-  DESCRIPTIONS = {
-    bm25: { type: "bm25", text: "BM25-ranked FTS5 retrieval", capabilities: ["lexical", "fts5", "bm25"] },
-    semantic: { type: "vector", text: "Semantic (embedding) retrieval", capabilities: ["semantic"] },
-    hybrid: {
-      type: "hybrid",
-      text: "Hybrid BM25 + semantic retrieval (reciprocal rank fusion)",
-      capabilities: ["lexical", "semantic", "hybrid"]
-    }
-  };
+  init_client9();
+  init_config3();
+  init_methods5();
+  TAGS = ["mailcrawl", "email", "pii"];
+});
+
+// src/datasource/skills/mailcrawl/index.ts
+var init_mailcrawl = __esm(() => {
+  init_client9();
+  init_methods5();
+  init_skill9();
 });
 
 // src/datasource/crawler-skill.ts
@@ -29004,7 +29298,7 @@ var init_crawler_skill = __esm(() => {
   init_errors4();
   init_scope2();
   init_sidecar();
-  init_methods5();
+  init_methods4();
   DEFAULT_POLLING_INTERVAL_MS4 = 15 * 60 * 1000;
 });
 
@@ -30793,14 +31087,27 @@ var init_factory = __esm(() => {
       ...config.channels?.ids !== undefined ? { chatIds: config.channels.ids } : {},
       ...larkConnectorOptions(config.connector)
     }),
-    kakao: (config, _workspaceRoot, _registrationName) => new LazykatokSkill({
-      client: new LazykatokClient({
-        ...config.connector
-      }),
-      ...config.instanceId !== undefined ? { instanceId: config.instanceId } : {},
-      ...config.pollingIntervalMs !== undefined ? { pollingIntervalMs: config.pollingIntervalMs } : {},
-      ...config.tags !== undefined ? { tags: config.tags } : {}
-    }),
+    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 sidecar = resolveSidecarOptions({
+        descriptor: CRAWLER_DESCRIPTORS.kakao,
+        connector: config.connector,
+        workspaceRoot,
+        registrationName,
+        instanceId: config.instanceId ?? "default",
+        defaultCrawlerDb: join39(dataDir, "archive.sqlite3")
+      });
+      return new LazykatokSkill({
+        client: new LazykatokClient(connector),
+        ...config.instanceId !== undefined ? { instanceId: config.instanceId } : {},
+        ...config.pollingIntervalMs !== undefined ? { pollingIntervalMs: config.pollingIntervalMs } : {},
+        ...config.tags !== undefined ? { tags: config.tags } : {},
+        channelIds: config.channels?.ids,
+        channelNames: config.channels?.names,
+        ...sidecar === undefined ? {} : { sidecar }
+      });
+    },
     github: (config, workspaceRoot, registrationName) => new GitHubSkill({
       ...common(config, workspaceRoot),
       skillName: registrationName,
@@ -30946,7 +31253,7 @@ function migrateLegacyConfig(configPath, legacyPath) {
   const legacy = readConfigFile(legacyPath, true);
   const legacyBytes = readFileSync25(legacyPath);
   const migrated = normalizeLegacyConfigPaths(legacy ?? {}, dirname26(legacyPath));
-  const migratedPathsUnchanged = legacy?.workspacePath === migrated.workspacePath && legacy?.memoryPath === migrated.memoryPath && JSON.stringify(legacy?.searchPaths) === JSON.stringify(migrated.searchPaths);
+  const migratedPathsUnchanged = legacy?.workspacePath === migrated.workspacePath && legacy?.memoryPath === migrated.memoryPath && JSON.stringify(legacy?.searchPaths) === JSON.stringify(migrated.searchPaths) && JSON.stringify(legacy?.excludePaths) === JSON.stringify(migrated.excludePaths);
   const migratedBytes = migratedPathsUnchanged ? legacyBytes : `${JSON.stringify(migrated, null, 2)}
 `;
   mkdirSync25(dirname26(configPath), { recursive: true });
@@ -30984,15 +31291,25 @@ function resolveConfigFileReadOnly(configPath, explicit, legacyPath) {
 function resolveSearchPaths(searchPaths, origin) {
   return searchPaths.map((searchPath) => resolvePersistedPath(searchPath, origin));
 }
+function resolveExcludePaths(value, origin) {
+  if (value === undefined)
+    return;
+  if (!Array.isArray(value) || value.some((path) => typeof path !== "string" || path.length === 0)) {
+    throw new ConfigError("Config field 'excludePaths' must be a non-empty string array");
+  }
+  return resolveSearchPaths(value, origin);
+}
 function resolvePersistedPath(path, origin) {
   return isAbsolute10(path) ? path : resolve16(origin, path);
 }
 function normalizeLegacyConfigPaths(partial, origin) {
   const workspacePath = resolvePersistedPath(partial.workspacePath ?? ".", origin);
+  const excludePaths = resolveExcludePaths(partial.excludePaths, workspacePath);
   return {
     ...partial,
     workspacePath,
     ...partial.searchPaths === undefined ? {} : { searchPaths: resolveSearchPaths(partial.searchPaths, workspacePath) },
+    ...excludePaths === undefined ? {} : { excludePaths },
     ...partial.memoryPath === undefined ? {} : { memoryPath: resolvePersistedPath(partial.memoryPath, workspacePath) }
   };
 }
@@ -31447,10 +31764,7 @@ function resolveConfig(input) {
   const configOrigin = dirname26(resolve16(configPath));
   const fileWorkspacePath = typeof file.workspacePath === "string" ? resolvePersistedPath(file.workspacePath, configOrigin) : undefined;
   const fileSearchPaths = file.searchPaths ? resolveSearchPaths(file.searchPaths, fileWorkspacePath ?? configOrigin) : undefined;
-  if (file.excludePaths !== undefined && (!Array.isArray(file.excludePaths) || file.excludePaths.some((path) => typeof path !== "string" || path.length === 0))) {
-    throw new ConfigError("Config field 'excludePaths' must be a non-empty string array");
-  }
-  const fileExcludePaths = file.excludePaths ? resolveSearchPaths(file.excludePaths, fileWorkspacePath ?? configOrigin) : undefined;
+  const fileExcludePaths = resolveExcludePaths(file.excludePaths, fileWorkspacePath ?? configOrigin);
   const searchPaths = flagSearchPaths ?? envSearchPaths ?? fileSearchPaths ?? defaultSearchPaths;
   const flagWorkspacePath = flagString(flags, "workspace");
   const envWorkspacePath = envString(env, "AUTORAG_WORKSPACE");
@@ -32186,7 +32500,7 @@ function writeDefaultConfig(path, partial, opts = {}) {
   full.dupey = partial.dupey ?? { enabled: true };
   full.excludeExactDuplicates = partial.excludeExactDuplicates ?? true;
   if (partial.excludePaths !== undefined)
-    full.excludePaths = resolveSearchPaths(partial.excludePaths, workspacePath);
+    full.excludePaths = resolveExcludePaths(partial.excludePaths, workspacePath);
   if (partial.limits !== undefined)
     full.limits = normalizeLimitsConfig(partial.limits);
   if (partial.refresh !== undefined)