@tangle-network/agent-knowledge 17.1.10 → 18.0.0

package/dist/sources/index.d.ts

@@ -30,7 +30,7 @@
  * cache; an undefined value disables caching (useful in tests). `now` is
  * injected for deterministic tests of change-detection windows.
  */
-interface FetchOpts {
+export interface FetchOpts {
   /** Abort signal forwarded to the underlying HTTP fetcher. */
   signal?: AbortSignal;
   /** Absolute path under which the source may cache raw bytes. */
@@ -58,7 +58,7 @@ interface FetchOpts {
  * `KnowledgeFragment` so freshness/change code can pass it around without
  * also dragging the body text.
  */
-interface FragmentProvenance {
+export interface FragmentProvenance {
   /** Canonical URL the fragment was extracted from. */
   url: string;
   /**
@@ -92,7 +92,7 @@ interface FragmentProvenance {
  * One unit of authoritative content. Stable hash on `(id, body)` lets change
  * detection reason about identity across snapshots.
  */
-interface KnowledgeFragment {
+export interface KnowledgeFragment {
   /**
    * Stable identity within (sourceId, selector-space). Two fetches against
    * the same authority section MUST produce the same `id`. The (sourceId,
@@ -130,7 +130,7 @@ interface KnowledgeFragment {
  * application's source list — there is no global registry by design (per
  * the per-tenant isolation contract; see README).
  */
-interface KnowledgeSource {
+export interface KnowledgeSource {
   /** Stable id — used to key freshness state. MUST NOT change once shipped. */
   id: string;
   /** Human-readable name for dashboards. */
@@ -150,7 +150,7 @@ interface KnowledgeSource {
 }
 //#endregion
 //#region src/sources/cornell-lii.d.ts
-interface CornellLiiSelector {
+export interface CornellLiiSelector {
   /** Either 'uscode' or 'wex'. */
   kind: 'uscode' | 'wex';
   /**
@@ -164,7 +164,7 @@ interface CornellLiiSelector {
    */
   dimensionHints?: string[];
 }
-interface CornellLiiSourceOptions {
+export interface CornellLiiSourceOptions {
   /**
    * Selectors to fetch on each `fetch()` call. The caller (a per-tenant
    * workspace config, typically) lists exactly the authorities they need
@@ -188,7 +188,7 @@ interface CornellLiiSourceOptions {
  * })
  * ```
  */
-declare function createCornellLiiSource(options: CornellLiiSourceOptions): KnowledgeSource;
+export declare function createCornellLiiSource(options: CornellLiiSourceOptions): KnowledgeSource;
 //#endregion
 //#region src/sources/html.d.ts
 /**
@@ -211,16 +211,16 @@ declare function createCornellLiiSource(options: CornellLiiSourceOptions): Knowl
  * Preserves paragraph and line breaks (`</p>`, `<br>`, `</li>`, `</div>`,
  * `</h*>`) as `\n` so statute text retains its subsection structure.
  */
-declare function htmlToText(html: string): string;
+export declare function htmlToText(html: string): string;
 /** Extract the first match of a regex's first capture group, or undefined. */
-declare function firstMatch(html: string, pattern: RegExp): string | undefined;
+export declare function firstMatch(html: string, pattern: RegExp): string | undefined;
 /** Extract the inner HTML of the first matching tag with id `id`. */
-declare function innerHtmlById(html: string, id: string): string | undefined;
+export declare function innerHtmlById(html: string, id: string): string | undefined;
 /**
  * Extract every (href, text) pair matching the URL regex.
  * Returns absolute URLs by resolving against `baseUrl`.
  */
-declare function extractLinks(html: string, hrefPattern: RegExp, baseUrl: string): {
+export declare function extractLinks(html: string, hrefPattern: RegExp, baseUrl: string): {
   href: string;
   text: string;
 }[];
@@ -236,12 +236,12 @@ declare function extractLinks(html: string, hrefPattern: RegExp, baseUrl: string
  * use successful status codes.
  */
 /** User-Agent string sent on every outbound request. */
-declare const POLITE_USER_AGENT = "agent-knowledge (+https://github.com/tangle-network/agent-knowledge)";
+export declare const POLITE_USER_AGENT = "agent-knowledge (+https://github.com/tangle-network/agent-knowledge)";
 /** Minimum gap between successive requests to the same origin (ms). */
-declare const MIN_REQUEST_GAP_MS = 1000;
+export declare const MIN_REQUEST_GAP_MS = 1000;
 /** Maximum response body we will buffer in memory (bytes). */
-declare const MAX_RESPONSE_BYTES: number;
-interface PoliteFetchOptions {
+export declare const MAX_RESPONSE_BYTES: number;
+export interface PoliteFetchOptions {
   signal?: AbortSignal;
   cacheDir?: string;
   /**
@@ -256,7 +256,7 @@ interface PoliteFetchOptions {
    */
   headers?: Record<string, string>;
 }
-interface PoliteFetchResult {
+export interface PoliteFetchResult {
   url: string;
   status: number;
   /** Decoded UTF-8 body. Truncated to `MAX_RESPONSE_BYTES`. */
@@ -286,14 +286,14 @@ interface PoliteFetchResult {
  * Throws ONLY on `AbortError` (caller asked to stop) and on cache-write
  * failures that indicate a misconfigured filesystem.
  */
-declare function politeFetch(url: string, options?: PoliteFetchOptions): Promise<PoliteFetchResult>;
+export declare function politeFetch(url: string, options?: PoliteFetchOptions): Promise<PoliteFetchResult>;
 /** Reset the in-process throttle map. Test-only. */
-declare function __resetHttpThrottle(): void;
+export declare function __resetHttpThrottle(): void;
 /** Cheap heuristic that catches CAPTCHA, WAF block pages, and "Just a moment" interstitials. */
-declare function looksLikeBlockPage(body: string): boolean;
+export declare function looksLikeBlockPage(body: string): boolean;
 //#endregion
 //#region src/sources/irs-publications.d.ts
-interface IrsPublicationsSourceOptions {
+export interface IrsPublicationsSourceOptions {
   /**
    * Specific publication slugs to fetch (e.g. `['p15', 'p17', 'p463']`).
    * When `includeIndex` is true (default), the publications index page is
@@ -310,8 +310,8 @@ interface IrsPublicationsSourceOptions {
   id?: string;
 }
 /** Default eval dimensions for IRS-sourced fragments. */
-declare const IRS_DIMENSION_HINTS: string[];
-declare function createIrsPublicationsSource(options?: IrsPublicationsSourceOptions): KnowledgeSource;
+export declare const IRS_DIMENSION_HINTS: string[];
+export declare function createIrsPublicationsSource(options?: IrsPublicationsSourceOptions): KnowledgeSource;
 //#endregion
 //#region src/sources/state-sos.d.ts
 /**
@@ -331,7 +331,7 @@ declare function createIrsPublicationsSource(options?: IrsPublicationsSourceOpti
  *
  * @experimental Interface will likely grow as we add more state coverage.
  */
-interface StateSosEntity {
+export interface StateSosEntity {
   /** Stable id for this fragment within the state (e.g. 'llc-formation', 'corp-formation'). */
   id: string;
   /** Path under the configured `baseUrl` for this entity. */
@@ -359,7 +359,7 @@ interface StateSosEntity {
   /** Eval dimensions this entity feeds. */
   dimensionHints?: string[];
 }
-interface StateSosSourceConfig {
+export interface StateSosSourceConfig {
   /** US state postal code, e.g. 'CA', 'DE', 'TX'. */
   state: string;
   /** Base URL for the state SOS — e.g. 'https://www.sos.ca.gov'. */
@@ -371,7 +371,6 @@ interface StateSosSourceConfig {
   /** Display name; default `<state> Secretary of State`. */
   name?: string;
 }
-declare function createStateSosSource(config: StateSosSourceConfig): KnowledgeSource;
+export declare function createStateSosSource(config: StateSosSourceConfig): KnowledgeSource;
 //#endregion
-export { CornellLiiSelector, CornellLiiSourceOptions, FetchOpts, FragmentProvenance, IRS_DIMENSION_HINTS, IrsPublicationsSourceOptions, KnowledgeFragment, KnowledgeSource, MAX_RESPONSE_BYTES, MIN_REQUEST_GAP_MS, POLITE_USER_AGENT, PoliteFetchOptions, PoliteFetchResult, StateSosEntity, StateSosSourceConfig, __resetHttpThrottle, createCornellLiiSource, createIrsPublicationsSource, createStateSosSource, extractLinks, firstMatch, htmlToText, innerHtmlById, looksLikeBlockPage, politeFetch };
 //# sourceMappingURL=index.d.ts.map

package/dist/sources/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/sources/types.ts","../../src/sources/cornell-lii.ts","../../src/sources/html.ts","../../src/sources/http.ts","../../src/sources/irs-publications.ts","../../src/sources/state-sos.ts"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;UAgCiB;;EAEf,SAAS;;EAET;;EAEA,YAAY;;;;;;EAMZ;;;;;;;;;EASA;;;;;;;UAQe;;EAEf;;;;;;;EAOA;;EAEA;;;;;;;EAOA;;;;;;;;EAQA;;EAEA;;;;;;UAOe;;;;;;EAMf;;EAEA;;EAEA;;EAEA;EACA,YAAY;;;;;;;;;;;;;EAaZ;;EAEA,WAAW;;;;;;;;;;UAWI;;EAEf;;EAEA;;EAEA;;;;;;;;;;EAUA,MAAM,MAAM,YAAY,QAAQ;;;;UCrIjB;;EAEf;;;;;EAKA;;;;;EAKA;;UAGe;;;;;;;EAOf,WAAW;;EAEX;;;;;;;;;;;;;;;iBAgBc,uBAAuB,SAAS,0BAA0B;;;;;;;;;;;;;;;;;;;;;;;iBCrC1D,WAAW;;iBA4BX,WAAW,cAAc,SAAS;;iBAKlC,cAAc,cAAc;;;;;iBAa5B,aACd,cACA,aAAa,QACb;EACG;EAAc;;;;;;;;;;;;;;cCxDN;;cAIA;;cAGA;UAII;EACf,SAAS;EACT;;;;;;EAMA;;;;;EAKA,UAAU;;UAGK;EACf;EACA;;EAEA;;;;;EAKA;EACA;;EAEA;;;;;;EAMA;EACA;;;;;;;;;;;iBAYoB,YACpB,aACA,UAAS,qBACR,QAAQ;;iBAoEK;;iBA6DA,mBAAmB;;;UChLlB;;;;;;;EAOf;;;;;EAKA;EACA;EACA;;;cAIW;iBAEG,4BACd,UAAS,+BACR;;;;;;;;;;;;;;;;;;;;UC5Bc;;EAEf;;EAEA;;;;;;;;EAQA;IACM;IAAY;;IACZ;IAAe;;IACf;IAAe,OAAO;;IACtB;;EACN;;EAEA;;UAGe;;EAEf;;EAEA;;EAEA,UAAU;;EAEV;;EAEA;;iBAGc,qBAAqB,QAAQ,uBAAuB"}
+{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/sources/types.ts","../../src/sources/cornell-lii.ts","../../src/sources/html.ts","../../src/sources/http.ts","../../src/sources/irs-publications.ts","../../src/sources/state-sos.ts"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBAgCiB;;EAEf,SAAS;;EAET;;EAEA,YAAY;;;;;;EAMZ;;;;;;;;;EASA;;;;;;;iBAQe;;EAEf;;;;;;;EAOA;;EAEA;;;;;;;EAOA;;;;;;;;EAQA;;EAEA;;;;;;iBAOe;;;;;;EAMf;;EAEA;;EAEA;;EAEA;EACA,YAAY;;;;;;;;;;;;;EAaZ;;EAEA,WAAW;;;;;;;;;;iBAWI;;EAEf;;EAEA;;EAEA;;;;;;;;;;EAUA,MAAM,MAAM,YAAY,QAAQ;;;;iBCrIjB;;EAEf;;;;;EAKA;;;;;EAKA;;iBAGe;;;;;;;EAOf,WAAW;;EAEX;;;;;;;;;;;;;;;wBAgBc,uBAAuB,SAAS,0BAA0B;;;;;;;;;;;;;;;;;;;;;;;wBCrC1D,WAAW;;wBA4BX,WAAW,cAAc,SAAS;;wBAKlC,cAAc,cAAc;;;;;wBAa5B,aACd,cACA,aAAa,QACb;EACG;EAAc;;;;;;;;;;;;;;qBCxDN;;qBAIA;;qBAGA;iBAII;EACf,SAAS;EACT;;;;;;EAMA;;;;;EAKA,UAAU;;iBAGK;EACf;EACA;;EAEA;;;;;EAKA;EACA;;EAEA;;;;;;EAMA;EACA;;;;;;;;;;;wBAYoB,YACpB,aACA,UAAS,qBACR,QAAQ;;wBAoEK;;wBA6DA,mBAAmB;;;iBChLlB;;;;;;;EAOf;;;;;EAKA;EACA;EACA;;;qBAIW;wBAEG,4BACd,UAAS,+BACR;;;;;;;;;;;;;;;;;;;;iBC5Bc;;EAEf;;EAEA;;;;;;;;EAQA;IACM;IAAY;;IACZ;IAAe;;IACf;IAAe,OAAO;;IACtB;;EACN;;EAEA;;iBAGe;;EAEf;;EAEA;;EAEA,UAAU;;EAEV;;EAEA;;wBAGc,qBAAqB,QAAQ,uBAAuB"}

package/dist/sources/index.js

@@ -72,7 +72,7 @@ const POLITE_USER_AGENT = "agent-knowledge (+https://github.com/tangle-network/a
 /** Minimum gap between successive requests to the same origin (ms). */
 const MIN_REQUEST_GAP_MS = 1e3;
 /** Maximum response body we will buffer in memory (bytes). */
-const MAX_RESPONSE_BYTES = 8 * 1024 * 1024;
+const MAX_RESPONSE_BYTES = 8388608;
 const hostThrottle = /* @__PURE__ */ new Map();
 /**
 * Fetch one URL with per-host throttling, on-disk cache, and block-page
@@ -84,7 +84,7 @@ const hostThrottle = /* @__PURE__ */ new Map();
 * failures that indicate a misconfigured filesystem.
 */
 async function politeFetch(url, options = {}) {
-	const cacheTtl = options.cacheTtlMs ?? 3600 * 1e3;
+	const cacheTtl = options.cacheTtlMs ?? 36e5;
 	const cached = options.cacheDir ? await readCache(options.cacheDir, url, cacheTtl) : void 0;
 	if (cached) return cached;
 	const host = safeHost(url);

package/dist/sources/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","names":["BASE_URL","extractTitle"],"sources":["../../src/sources/html.ts","../../src/sources/http.ts","../../src/sources/cornell-lii.ts","../../src/sources/irs-publications.ts","../../src/sources/state-sos.ts"],"sourcesContent":["/**\n * Minimal HTML helpers used by the shipped sources.\n *\n * Deliberately not a full DOM parser: every authority we ship against\n * (Cornell LII, IRS.gov, state SOS portals) has well-behaved server-rendered\n * HTML where regex-based extraction is correct and cheap. Bringing in cheerio\n * would add a 1.5MB dependency to a package whose purpose is shipping\n * primitives, not parsing arbitrary web pages.\n *\n * If a future source needs real DOM traversal, it should depend on its own\n * parser locally rather than promoting one into the package-wide deps.\n *\n * @stable\n */\n\n/**\n * Strip HTML tags, collapse whitespace, decode common entities.\n *\n * Preserves paragraph and line breaks (`</p>`, `<br>`, `</li>`, `</div>`,\n * `</h*>`) as `\\n` so statute text retains its subsection structure.\n */\nexport function htmlToText(html: string): string {\n  return html\n    .replace(/<script[\\s\\S]*?<\\/script>/gi, '')\n    .replace(/<style[\\s\\S]*?<\\/style>/gi, '')\n    .replace(/<noscript[\\s\\S]*?<\\/noscript>/gi, '')\n    .replace(/<!--([\\s\\S]*?)-->/g, '')\n    .replace(/<\\s*br\\s*\\/?>/gi, '\\n')\n    .replace(/<\\/(p|li|div|tr|h[1-6]|blockquote|section|article)>/gi, '\\n')\n    .replace(/<[^>]+>/g, '')\n    .replace(/&nbsp;/gi, ' ')\n    .replace(/&amp;/gi, '&')\n    .replace(/&lt;/gi, '<')\n    .replace(/&gt;/gi, '>')\n    .replace(/&quot;/gi, '\"')\n    .replace(/&#39;/gi, \"'\")\n    .replace(/&sect;/gi, '§')\n    .replace(/&mdash;/gi, '—')\n    .replace(/&ndash;/gi, '–')\n    .replace(/&#(\\d+);/g, (_, code) => String.fromCodePoint(Number(code)))\n    .replace(/&#x([0-9a-f]+);/gi, (_, code) => String.fromCodePoint(Number.parseInt(code, 16)))\n    .split('\\n')\n    .map((line) => line.replace(/[\\t  ]+/g, ' ').trim())\n    .filter((line, idx, all) => !(line === '' && all[idx - 1] === ''))\n    .join('\\n')\n    .trim()\n}\n\n/** Extract the first match of a regex's first capture group, or undefined. */\nexport function firstMatch(html: string, pattern: RegExp): string | undefined {\n  return pattern.exec(html)?.[1]?.trim()\n}\n\n/** Extract the inner HTML of the first matching tag with id `id`. */\nexport function innerHtmlById(html: string, id: string): string | undefined {\n  const escaped = id.replace(/[.*+?^${}()|[\\]\\\\]/g, '\\\\$&')\n  const tagPattern = new RegExp(\n    `<([a-z][a-z0-9]*)\\\\b[^>]*\\\\sid=[\"']${escaped}[\"'][^>]*>([\\\\s\\\\S]*?)<\\\\/\\\\1>`,\n    'i',\n  )\n  return tagPattern.exec(html)?.[2]\n}\n\n/**\n * Extract every (href, text) pair matching the URL regex.\n * Returns absolute URLs by resolving against `baseUrl`.\n */\nexport function extractLinks(\n  html: string,\n  hrefPattern: RegExp,\n  baseUrl: string,\n): { href: string; text: string }[] {\n  const out: { href: string; text: string }[] = []\n  const anchor = /<a\\b[^>]*\\shref=[\"']([^\"']+)[\"'][^>]*>([\\s\\S]*?)<\\/a>/gi\n  for (const match of html.matchAll(anchor)) {\n    const href = match[1]\n    const inner = match[2]\n    if (!href || !inner) continue\n    if (!hrefPattern.test(href)) continue\n    const text = htmlToText(inner)\n    if (!text) continue\n    try {\n      out.push({ href: new URL(href, baseUrl).toString(), text })\n    } catch {\n      /* skip malformed URL */\n    }\n  }\n  return out\n}\n","import { mkdir, readFile, stat, writeFile } from 'node:fs/promises'\nimport { dirname, join } from 'node:path'\nimport { sha256 } from '../ids'\n\n/**\n * Polite HTTP fetcher shared by remote sources.\n *\n * Independent sources share a per-origin throttle because rate-limited sites\n * may return block pages instead of 429 responses. Responses are cached by URL\n * because many publishers omit reliable ETag and Last-Modified headers. Bodies\n * are checked even after a 2xx response because captcha and block pages often\n * use successful status codes.\n */\n\n/** User-Agent string sent on every outbound request. */\nexport const POLITE_USER_AGENT =\n  'agent-knowledge (+https://github.com/tangle-network/agent-knowledge)'\n\n/** Minimum gap between successive requests to the same origin (ms). */\nexport const MIN_REQUEST_GAP_MS = 1_000\n\n/** Maximum response body we will buffer in memory (bytes). */\nexport const MAX_RESPONSE_BYTES = 8 * 1024 * 1024\n\nconst hostThrottle = new Map<string, Promise<void>>()\n\nexport interface PoliteFetchOptions {\n  signal?: AbortSignal\n  cacheDir?: string\n  /**\n   * Cache age beyond which we re-fetch. Default 1 hour, long enough to\n   * batch a cron sweep across many selectors, short enough that hourly\n   * authoritative-page changes get picked up next tick.\n   */\n  cacheTtlMs?: number\n  /**\n   * Extra request headers. The fetcher always sets `User-Agent` and\n   * `Accept`; callers can add `Accept-Language` etc.\n   */\n  headers?: Record<string, string>\n}\n\nexport interface PoliteFetchResult {\n  url: string\n  status: number\n  /** Decoded UTF-8 body. Truncated to `MAX_RESPONSE_BYTES`. */\n  body: string\n  /**\n   * Best-effort source-attested timestamp. Reads `Last-Modified`,\n   * falling back to `Date`, falling back to fetch time. Always ISO 8601.\n   */\n  sourceUpdatedAt: string\n  fetchedAt: string\n  /** True iff the response was satisfied from disk cache. */\n  fromCache: boolean\n  /**\n   * False on: non-2xx status, captcha/block page heuristic match, or\n   * decoded body below 200 chars from a host known to serve real content\n   * (Cornell, IRS, state SOS). `unverifiableReason` carries the why.\n   */\n  verifiable: boolean\n  unverifiableReason?: string\n}\n\n/**\n * Fetch one URL with per-host throttling, on-disk cache, and block-page\n * detection. Never throws on network/HTTP failure. It returns a result with\n * `verifiable: false` and `unverifiableReason` set so the caller can decide\n * whether to skip, retry, or surface.\n *\n * Throws ONLY on `AbortError` (caller asked to stop) and on cache-write\n * failures that indicate a misconfigured filesystem.\n */\nexport async function politeFetch(\n  url: string,\n  options: PoliteFetchOptions = {},\n): Promise<PoliteFetchResult> {\n  const cacheTtl = options.cacheTtlMs ?? 60 * 60 * 1000\n  const cached = options.cacheDir ? await readCache(options.cacheDir, url, cacheTtl) : undefined\n  if (cached) return cached\n\n  const host = safeHost(url)\n  await throttleHost(host)\n\n  const fetchedAt = new Date().toISOString()\n  let response: Response\n  try {\n    response = await fetch(url, {\n      signal: options.signal,\n      redirect: 'follow',\n      headers: {\n        'User-Agent': POLITE_USER_AGENT,\n        Accept: 'text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8',\n        'Accept-Language': 'en-US,en;q=0.9',\n        ...(options.headers ?? {}),\n      },\n    })\n  } catch (error) {\n    if ((error as { name?: string }).name === 'AbortError') throw error\n    const result: PoliteFetchResult = {\n      url,\n      status: 0,\n      body: '',\n      sourceUpdatedAt: fetchedAt,\n      fetchedAt,\n      fromCache: false,\n      verifiable: false,\n      unverifiableReason: `network error: ${(error as Error).message}`,\n    }\n    if (options.cacheDir) await writeCache(options.cacheDir, url, result)\n    return result\n  }\n\n  const text = await readBoundedText(response)\n  const lastModified = response.headers.get('last-modified')\n  const dateHeader = response.headers.get('date')\n  const sourceUpdatedAt = parseHttpDate(lastModified) ?? parseHttpDate(dateHeader) ?? fetchedAt\n\n  const result: PoliteFetchResult = {\n    url,\n    status: response.status,\n    body: text,\n    sourceUpdatedAt,\n    fetchedAt,\n    fromCache: false,\n    verifiable: true,\n  }\n\n  if (response.status < 200 || response.status >= 300) {\n    result.verifiable = false\n    result.unverifiableReason = `non-2xx status: ${response.status}`\n  } else if (looksLikeBlockPage(text)) {\n    result.verifiable = false\n    result.unverifiableReason = 'block-page heuristic matched'\n  } else if (text.length < 200 && knownLargeAuthority(host)) {\n    result.verifiable = false\n    result.unverifiableReason = `body shorter than expected (${text.length} chars)`\n  }\n\n  if (options.cacheDir) await writeCache(options.cacheDir, url, result)\n  return result\n}\n\n/** Reset the in-process throttle map. Test-only. */\nexport function __resetHttpThrottle(): void {\n  hostThrottle.clear()\n}\n\nfunction safeHost(url: string): string {\n  try {\n    return new URL(url).host\n  } catch {\n    return 'unknown'\n  }\n}\n\nasync function throttleHost(host: string): Promise<void> {\n  const prev = hostThrottle.get(host) ?? Promise.resolve()\n  let release: () => void = () => {}\n  const next = new Promise<void>((resolve) => {\n    release = resolve\n  })\n  hostThrottle.set(\n    host,\n    prev.then(() => next),\n  )\n  await prev\n  setTimeout(release, MIN_REQUEST_GAP_MS)\n}\n\nasync function readBoundedText(response: Response): Promise<string> {\n  if (!response.body) return ''\n  const reader = response.body.getReader()\n  const chunks: Uint8Array[] = []\n  let total = 0\n  while (true) {\n    const { done, value } = await reader.read()\n    if (done) break\n    if (!value) continue\n    total += value.length\n    if (total > MAX_RESPONSE_BYTES) {\n      // Stop reading; release the underlying connection.\n      await reader.cancel()\n      break\n    }\n    chunks.push(value)\n  }\n  const merged = new Uint8Array(Math.min(total, MAX_RESPONSE_BYTES))\n  let offset = 0\n  for (const chunk of chunks) {\n    const take = Math.min(chunk.length, merged.length - offset)\n    if (take <= 0) break\n    merged.set(chunk.subarray(0, take), offset)\n    offset += take\n  }\n  return new TextDecoder('utf-8', { fatal: false }).decode(merged)\n}\n\nfunction parseHttpDate(value: string | null): string | undefined {\n  if (!value) return undefined\n  const ms = Date.parse(value)\n  return Number.isFinite(ms) ? new Date(ms).toISOString() : undefined\n}\n\n/** Cheap heuristic that catches CAPTCHA, WAF block pages, and \"Just a moment\" interstitials. */\nexport function looksLikeBlockPage(body: string): boolean {\n  if (!body) return false\n  const lower = body.toLowerCase()\n  const markers = [\n    'verify you are human',\n    'please enable javascript and cookies',\n    'just a moment',\n    'access denied',\n    'request unsuccessful',\n    'cf-error-details',\n    'captcha',\n    'incapsula',\n    'pardon our interruption',\n  ]\n  for (const marker of markers) {\n    if (lower.includes(marker)) return true\n  }\n  return false\n}\n\nfunction knownLargeAuthority(host: string): boolean {\n  return (\n    host.endsWith('law.cornell.edu') ||\n    host.endsWith('irs.gov') ||\n    host.endsWith('sos.ca.gov') ||\n    host.endsWith('sos.state.tx.us') ||\n    host.endsWith('sos.state.us')\n  )\n}\n\nfunction cachePath(cacheDir: string, url: string): string {\n  const key = sha256(url)\n  return join(cacheDir, 'http', `${key.slice(0, 2)}`, `${key}.json`)\n}\n\nasync function readCache(\n  cacheDir: string,\n  url: string,\n  ttlMs: number,\n): Promise<PoliteFetchResult | undefined> {\n  const path = cachePath(cacheDir, url)\n  try {\n    const info = await stat(path)\n    if (Date.now() - info.mtimeMs > ttlMs) return undefined\n    const raw = await readFile(path, 'utf8')\n    const parsed = JSON.parse(raw) as PoliteFetchResult\n    return { ...parsed, fromCache: true }\n  } catch {\n    return undefined\n  }\n}\n\nasync function writeCache(cacheDir: string, url: string, value: PoliteFetchResult): Promise<void> {\n  const path = cachePath(cacheDir, url)\n  await mkdir(dirname(path), { recursive: true })\n  await writeFile(path, JSON.stringify(value), 'utf8')\n}\n","import { sha256 } from '../ids'\nimport { htmlToText, innerHtmlById } from './html'\nimport { politeFetch } from './http'\nimport type { FetchOpts, KnowledgeFragment, KnowledgeSource } from './types'\n\n/**\n * Cornell Legal Information Institute (LII) source.\n *\n * Pulls federal US Code sections and Wex encyclopedia entries — the two\n * Cornell LII surfaces an agent typically grounds against. The Wex\n * \"non-compete\" page is the canonical test case for the Ryan-LLC v. FTC\n * vacatur drift the continuous-ingestion story is designed to catch.\n *\n * @stable\n */\n\nconst BASE_URL = 'https://www.law.cornell.edu'\n\nexport interface CornellLiiSelector {\n  /** Either 'uscode' or 'wex'. */\n  kind: 'uscode' | 'wex'\n  /**\n   * For `uscode`: `<title>/<section>` (e.g. `'18/1836'` for DTSA).\n   * For `wex`: the slug (e.g. `'non-compete'`).\n   */\n  path: string\n  /**\n   * Optional pre-declared eval dimensions affected by this section. If\n   * omitted, defaults are chosen from `kind` + path heuristics.\n   */\n  dimensionHints?: string[]\n}\n\nexport interface CornellLiiSourceOptions {\n  /**\n   * Selectors to fetch on each `fetch()` call. The caller (a per-tenant\n   * workspace config, typically) lists exactly the authorities they need\n   * tracked. There is no auto-discovery; that would crawl Cornell at\n   * cron speed, which is what the polite-fetch contract exists to avoid.\n   */\n  selectors: CornellLiiSelector[]\n  /** Source id override; default is `'cornell-lii'`. */\n  id?: string\n}\n\n/**\n * Build a Cornell LII source for the listed selectors.\n *\n * Example: track DTSA + non-compete:\n * ```\n * createCornellLiiSource({\n *   selectors: [\n *     { kind: 'uscode', path: '18/1836' },\n *     { kind: 'wex', path: 'non-compete', dimensionHints: ['jurisdictional_accuracy'] },\n *   ],\n * })\n * ```\n */\nexport function createCornellLiiSource(options: CornellLiiSourceOptions): KnowledgeSource {\n  const id = options.id ?? 'cornell-lii'\n  return {\n    id,\n    name: 'Cornell Legal Information Institute',\n    description:\n      'Federal US Code sections (uscode/text/...) and Wex legal encyclopedia entries from law.cornell.edu.',\n    async fetch(opts: FetchOpts): Promise<KnowledgeFragment[]> {\n      const limit = opts.limit ?? options.selectors.length\n      const selectors = options.selectors.slice(0, limit)\n      const out: KnowledgeFragment[] = []\n      for (const selector of selectors) {\n        out.push(await fetchOne(id, selector, opts))\n      }\n      return out\n    },\n  }\n}\n\nasync function fetchOne(\n  sourceId: string,\n  selector: CornellLiiSelector,\n  opts: FetchOpts,\n): Promise<KnowledgeFragment> {\n  const path = selector.path.replace(/^\\/+/, '')\n  const url =\n    selector.kind === 'uscode' ? `${BASE_URL}/uscode/text/${path}` : `${BASE_URL}/wex/${path}`\n\n  const response = await politeFetch(url, {\n    signal: opts.signal,\n    cacheDir: opts.cacheDir,\n  })\n\n  const fragmentId = `${selector.kind}:${selector.path}`\n  const dimensionHints = selector.dimensionHints ?? defaultDimensionHints(selector)\n\n  if (!response.verifiable) {\n    return {\n      id: fragmentId,\n      title: `Cornell LII ${selector.kind} ${selector.path}`,\n      body: '',\n      bodyHash: sha256(''),\n      provenance: {\n        url,\n        sourceUpdatedAt: response.sourceUpdatedAt,\n        fetchedAt: response.fetchedAt,\n        jurisdiction: 'US-FED',\n        verifiable: false,\n        unverifiableReason: response.unverifiableReason,\n      },\n      dimensionHints,\n      metadata: { sourceId, status: response.status, fromCache: response.fromCache },\n    }\n  }\n\n  const html = response.body\n  const title = extractTitle(html, selector)\n  const body = extractBody(html, selector)\n  const effective = extractEffectiveDate(html) ?? response.sourceUpdatedAt\n\n  const verifiable = body.length > 50\n  return {\n    id: fragmentId,\n    title,\n    body,\n    bodyHash: sha256(body),\n    provenance: {\n      url,\n      sourceUpdatedAt: effective,\n      fetchedAt: response.fetchedAt,\n      jurisdiction: 'US-FED',\n      verifiable,\n      unverifiableReason: verifiable ? undefined : 'extracted body too short',\n    },\n    dimensionHints,\n    metadata: { sourceId, status: response.status, fromCache: response.fromCache },\n  }\n}\n\nfunction extractTitle(html: string, selector: CornellLiiSelector): string {\n  const h1 = /<h1[^>]*\\bid=[\"']page_title[\"'][^>]*>([\\s\\S]*?)<\\/h1>/i.exec(html)?.[1]\n  if (h1) return htmlToText(h1)\n  const t = /<title>([\\s\\S]*?)<\\/title>/i.exec(html)?.[1]\n  if (t) return htmlToText(t).split(' | ')[0] ?? `Cornell LII ${selector.path}`\n  return `Cornell LII ${selector.kind} ${selector.path}`\n}\n\nfunction extractBody(html: string, selector: CornellLiiSelector): string {\n  if (selector.kind === 'uscode') {\n    // The statute text lives inside a <text><div class=\"text\">…</div></text>\n    // block on US Code section pages. Prefer it; fall back to #tab_default_1\n    // which always contains the section body.\n    const text = /<text>([\\s\\S]*?)<\\/text>/i.exec(html)?.[1]\n    if (text) return htmlToText(text)\n    const tab = innerHtmlById(html, 'tab_default_1')\n    if (tab) return htmlToText(tab)\n  }\n  // Wex pages wrap the encyclopedia entry under <div id=\"main-content\"> (newer\n  // Drupal template) or directly inside <div id=\"extracted-content\"> (older\n  // template). Try both — lazy regex matching against a nested-div container\n  // returns the wrong (shorter) slice, so we anchor on the leaf containers.\n  const mainContent = innerHtmlById(html, 'main-content')\n  if (mainContent) {\n    return htmlToText(mainContent.replace(/<h1[\\s\\S]*?<\\/h1>/i, ''))\n  }\n  const extracted = innerHtmlById(html, 'extracted-content')\n  if (extracted) {\n    return htmlToText(extracted.replace(/<h1[\\s\\S]*?<\\/h1>/i, ''))\n  }\n  return htmlToText(html)\n}\n\nfunction extractEffectiveDate(html: string): string | undefined {\n  // Cornell LII includes \"Editorial Notes\" / \"Amendments\" blocks with\n  // dates; the most reliable machine-readable signal is the last\n  // amendment year embedded near the section text.\n  const amend = /Amendments[\\s\\S]{0,200}?(\\d{4})/i.exec(html)?.[1]\n  if (amend) {\n    const y = Number.parseInt(amend, 10)\n    if (Number.isFinite(y) && y > 1900 && y <= new Date().getUTCFullYear() + 1) {\n      return new Date(Date.UTC(y, 11, 31)).toISOString()\n    }\n  }\n  return undefined\n}\n\nfunction defaultDimensionHints(selector: CornellLiiSelector): string[] {\n  if (selector.kind === 'uscode') return ['jurisdictional_accuracy', 'citation_hygiene']\n  return ['citation_hygiene']\n}\n","import { sha256 } from '../ids'\nimport { htmlToText } from './html'\nimport { politeFetch } from './http'\nimport type { FetchOpts, KnowledgeFragment, KnowledgeSource } from './types'\n\n/**\n * IRS publications source.\n *\n * Two surfaces:\n *\n *   1. The publications index at https://www.irs.gov/publications enumerates\n *      every active publication with its revision year — a single fragment\n *      with the full table lets change detection notice when a publication\n *      year flips (e.g. Pub 15 (2025) → Pub 15 (2026)).\n *\n *   2. Individual publication landing pages at /publications/p<N>[<suffix>]\n *      return one fragment per publication with summary text. Callers list\n *      the publications they need tracked via `selectors`.\n *\n * Revenue procedures are fetched under their numbered URLs; the IRS does\n * not maintain a stable HTML index of rev-procs, so the caller passes the\n * specific rev-proc paths they care about.\n *\n * @stable\n */\n\nconst BASE_URL = 'https://www.irs.gov'\nconst INDEX_URL = `${BASE_URL}/publications`\n\nexport interface IrsPublicationsSourceOptions {\n  /**\n   * Specific publication slugs to fetch (e.g. `['p15', 'p17', 'p463']`).\n   * When `includeIndex` is true (default), the publications index page is\n   * also fetched as a single fragment so change detection can notice\n   * year/revision shifts across the whole catalogue.\n   */\n  publications?: string[]\n  /**\n   * Revenue procedure paths to fetch (e.g. `['/irb/2024-31_IRB']`). The\n   * caller passes the exact path; this source does not auto-discover.\n   */\n  revenueProcedures?: string[]\n  includeIndex?: boolean\n  id?: string\n}\n\n/** Default eval dimensions for IRS-sourced fragments. */\nexport const IRS_DIMENSION_HINTS = ['tax_compliance', 'regulatory_currency', 'citation_hygiene']\n\nexport function createIrsPublicationsSource(\n  options: IrsPublicationsSourceOptions = {},\n): KnowledgeSource {\n  const id = options.id ?? 'irs-publications'\n  const includeIndex = options.includeIndex ?? true\n  return {\n    id,\n    name: 'IRS Publications',\n    description:\n      'Internal Revenue Service publications index and individual publication landing pages from irs.gov.',\n    async fetch(opts: FetchOpts): Promise<KnowledgeFragment[]> {\n      const out: KnowledgeFragment[] = []\n      const limit = opts.limit ?? Number.POSITIVE_INFINITY\n\n      if (includeIndex && out.length < limit) {\n        out.push(await fetchIndex(id, opts))\n      }\n      for (const slug of options.publications ?? []) {\n        if (out.length >= limit) break\n        out.push(await fetchPublication(id, slug, opts))\n      }\n      for (const path of options.revenueProcedures ?? []) {\n        if (out.length >= limit) break\n        out.push(await fetchRevenueProcedure(id, path, opts))\n      }\n      return out\n    },\n  }\n}\n\nasync function fetchIndex(sourceId: string, opts: FetchOpts): Promise<KnowledgeFragment> {\n  const response = await politeFetch(INDEX_URL, { signal: opts.signal, cacheDir: opts.cacheDir })\n  const tablePattern = /<table[\\s\\S]*?<\\/table>/gi\n  const matches = response.body.match(tablePattern) ?? []\n  // Extract the table that lists current-year publications. IRS publishes\n  // one table per year on the index; the most recent table is always the\n  // first that mentions a year ≥ current.\n  const tables = matches.map((t) => htmlToText(t))\n  const body = tables\n    .filter((t) => /Publication\\s*\\d+/i.test(t))\n    .join('\\n\\n')\n    .slice(0, 200_000)\n\n  const verifiable = response.verifiable && body.length > 200\n  return {\n    id: 'index',\n    title: 'IRS Publications Index',\n    body,\n    bodyHash: sha256(body),\n    provenance: {\n      url: INDEX_URL,\n      sourceUpdatedAt: response.sourceUpdatedAt,\n      fetchedAt: response.fetchedAt,\n      jurisdiction: 'US-FED',\n      verifiable,\n      unverifiableReason:\n        response.unverifiableReason ?? (verifiable ? undefined : 'no publication rows extracted'),\n    },\n    dimensionHints: IRS_DIMENSION_HINTS,\n    metadata: { sourceId, status: response.status, fromCache: response.fromCache, kind: 'index' },\n  }\n}\n\nasync function fetchPublication(\n  sourceId: string,\n  slug: string,\n  opts: FetchOpts,\n): Promise<KnowledgeFragment> {\n  const url = `${BASE_URL}/publications/${slug.replace(/^\\/+/, '')}`\n  const response = await politeFetch(url, { signal: opts.signal, cacheDir: opts.cacheDir })\n\n  const title = extractTitle(response.body, `IRS Publication ${slug}`)\n  const body = extractMainContent(response.body)\n  const verifiable = response.verifiable && body.length > 200\n\n  return {\n    id: `publication:${slug}`,\n    title,\n    body,\n    bodyHash: sha256(body),\n    provenance: {\n      url,\n      sourceUpdatedAt: extractRevisionDate(response.body) ?? response.sourceUpdatedAt,\n      fetchedAt: response.fetchedAt,\n      jurisdiction: 'US-FED',\n      verifiable,\n      unverifiableReason:\n        response.unverifiableReason ?? (verifiable ? undefined : 'no publication body extracted'),\n    },\n    dimensionHints: IRS_DIMENSION_HINTS,\n    metadata: {\n      sourceId,\n      status: response.status,\n      fromCache: response.fromCache,\n      kind: 'publication',\n      slug,\n    },\n  }\n}\n\nasync function fetchRevenueProcedure(\n  sourceId: string,\n  path: string,\n  opts: FetchOpts,\n): Promise<KnowledgeFragment> {\n  const url = `${BASE_URL}${path.startsWith('/') ? path : `/${path}`}`\n  const response = await politeFetch(url, { signal: opts.signal, cacheDir: opts.cacheDir })\n  const body = extractMainContent(response.body)\n  const verifiable = response.verifiable && body.length > 200\n  return {\n    id: `rev-proc:${path}`,\n    title: extractTitle(response.body, `IRS Revenue Procedure ${path}`),\n    body,\n    bodyHash: sha256(body),\n    provenance: {\n      url,\n      sourceUpdatedAt: response.sourceUpdatedAt,\n      fetchedAt: response.fetchedAt,\n      jurisdiction: 'US-FED',\n      verifiable,\n      unverifiableReason:\n        response.unverifiableReason ??\n        (verifiable ? undefined : 'no revenue-procedure body extracted'),\n    },\n    dimensionHints: [...IRS_DIMENSION_HINTS, 'procedural_currency'],\n    metadata: {\n      sourceId,\n      status: response.status,\n      fromCache: response.fromCache,\n      kind: 'rev-proc',\n      path,\n    },\n  }\n}\n\nfunction extractTitle(html: string, fallback: string): string {\n  const og = /<meta\\s+property=[\"']og:title[\"']\\s+content=[\"']([^\"']+)[\"']/i.exec(html)?.[1]\n  if (og) return decodeHtml(og)\n  const title = /<title>([\\s\\S]*?)<\\/title>/i.exec(html)?.[1]\n  if (title) return htmlToText(title).split(' | ')[0] ?? fallback\n  return fallback\n}\n\nfunction extractMainContent(html: string): string {\n  // IRS uses Drupal — the main publication body is inside <main role=\"main\">\n  // or under .field--name-body. We try main first; on miss, body.\n  const main = /<main\\b[\\s\\S]*?<\\/main>/i.exec(html)?.[0]\n  if (main) {\n    const noNav = main\n      .replace(/<nav[\\s\\S]*?<\\/nav>/gi, '')\n      .replace(/<header[\\s\\S]*?<\\/header>/gi, '')\n      .replace(/<footer[\\s\\S]*?<\\/footer>/gi, '')\n    return htmlToText(noNav).slice(0, 200_000)\n  }\n  const body = /<body\\b[\\s\\S]*?<\\/body>/i.exec(html)?.[0]\n  return body ? htmlToText(body).slice(0, 200_000) : htmlToText(html).slice(0, 200_000)\n}\n\nfunction extractRevisionDate(html: string): string | undefined {\n  // IRS publication pages typically show \"Publication X (YYYY)\" in the title;\n  // pulling the year gives a stable revision marker.\n  const m = /Publication\\s+\\S+\\s*\\((\\d{4})\\)/i.exec(html)\n  if (m?.[1]) {\n    const year = Number.parseInt(m[1], 10)\n    if (Number.isFinite(year) && year >= 2000 && year <= new Date().getUTCFullYear() + 1) {\n      return new Date(Date.UTC(year, 0, 1)).toISOString()\n    }\n  }\n  return undefined\n}\n\nfunction decodeHtml(value: string): string {\n  return htmlToText(value)\n}\n","import { sha256 } from '../ids'\nimport { htmlToText } from './html'\nimport { politeFetch } from './http'\nimport type { FetchOpts, KnowledgeFragment, KnowledgeSource } from './types'\n\n/**\n * Generic Secretary-of-State source.\n *\n * Every US state SOS surfaces LLC/Corp formation requirements differently\n * (CA via static forms pages, DE via division of corporations pages, TX\n * via SOSDirect content pages). Rather than baking 50 state-specific\n * parsers into this package, the source takes a config that names the URL\n * pattern + CSS-equivalent selector + jurisdiction tag. Callers supply one\n * config per state they need tracked.\n *\n * The selector is interpreted as a substring/regex of an HTML element id\n * or class — see `StateSosSourceConfig` for the contract. This is\n * intentionally minimal; richer extraction belongs in a state-specific\n * adapter the consumer authors.\n *\n * @experimental Interface will likely grow as we add more state coverage.\n */\n\nexport interface StateSosEntity {\n  /** Stable id for this fragment within the state (e.g. 'llc-formation', 'corp-formation'). */\n  id: string\n  /** Path under the configured `baseUrl` for this entity. */\n  path: string\n  /**\n   * Extraction selector. Choose one:\n   *   - `{ kind: 'id', value: 'main-content' }` — innermost match of element with that id\n   *   - `{ kind: 'class', value: 'field--name-body' }` — innermost match of element with that class\n   *   - `{ kind: 'regex', value: /<article[\\s\\S]*?<\\/article>/i }` — raw regex\n   *   - `{ kind: 'whole' }` — full body, tags stripped (fallback for unstructured pages)\n   */\n  selector:\n    | { kind: 'id'; value: string }\n    | { kind: 'class'; value: string }\n    | { kind: 'regex'; value: RegExp }\n    | { kind: 'whole' }\n  title: string\n  /** Eval dimensions this entity feeds. */\n  dimensionHints?: string[]\n}\n\nexport interface StateSosSourceConfig {\n  /** US state postal code, e.g. 'CA', 'DE', 'TX'. */\n  state: string\n  /** Base URL for the state SOS — e.g. 'https://www.sos.ca.gov'. */\n  baseUrl: string\n  /** Entities this state exposes (LLC, Corp, etc). */\n  entities: StateSosEntity[]\n  /** Source id; default `state-sos:<state>`. */\n  id?: string\n  /** Display name; default `<state> Secretary of State`. */\n  name?: string\n}\n\nexport function createStateSosSource(config: StateSosSourceConfig): KnowledgeSource {\n  const id = config.id ?? `state-sos:${config.state.toLowerCase()}`\n  const name = config.name ?? `${config.state} Secretary of State`\n  return {\n    id,\n    name,\n    description: `${config.state} Secretary of State filings and formation guidance pages.`,\n    async fetch(opts: FetchOpts): Promise<KnowledgeFragment[]> {\n      const limit = opts.limit ?? config.entities.length\n      const entities = config.entities.slice(0, limit)\n      const out: KnowledgeFragment[] = []\n      for (const entity of entities) {\n        out.push(await fetchEntity(id, config, entity, opts))\n      }\n      return out\n    },\n  }\n}\n\nasync function fetchEntity(\n  sourceId: string,\n  config: StateSosSourceConfig,\n  entity: StateSosEntity,\n  opts: FetchOpts,\n): Promise<KnowledgeFragment> {\n  const url = joinUrl(config.baseUrl, entity.path)\n  const response = await politeFetch(url, { signal: opts.signal, cacheDir: opts.cacheDir })\n\n  const body = response.verifiable ? extractBySelector(response.body, entity.selector) : ''\n  const verifiable = response.verifiable && body.length > 100\n\n  return {\n    id: entity.id,\n    title: entity.title,\n    body,\n    bodyHash: sha256(body),\n    provenance: {\n      url,\n      sourceUpdatedAt: response.sourceUpdatedAt,\n      fetchedAt: response.fetchedAt,\n      jurisdiction: `US-${config.state.toUpperCase()}`,\n      verifiable,\n      unverifiableReason:\n        response.unverifiableReason ?? (verifiable ? undefined : 'extracted body too short'),\n    },\n    dimensionHints: entity.dimensionHints ?? [\n      'jurisdictional_accuracy',\n      'corporate_formation',\n      'citation_hygiene',\n    ],\n    metadata: {\n      sourceId,\n      status: response.status,\n      fromCache: response.fromCache,\n      state: config.state,\n    },\n  }\n}\n\nfunction extractBySelector(html: string, selector: StateSosEntity['selector']): string {\n  if (selector.kind === 'whole') {\n    const main = /<main\\b[\\s\\S]*?<\\/main>/i.exec(html)?.[0]\n    return htmlToText(main ?? html).slice(0, 200_000)\n  }\n  if (selector.kind === 'regex') {\n    const m = selector.value.exec(html)?.[0]\n    return m ? htmlToText(m).slice(0, 200_000) : ''\n  }\n  if (selector.kind === 'id') {\n    const escaped = selector.value.replace(/[.*+?^${}()|[\\]\\\\]/g, '\\\\$&')\n    const pattern = new RegExp(\n      `<([a-z][a-z0-9]*)\\\\b[^>]*\\\\sid=[\"']${escaped}[\"'][^>]*>([\\\\s\\\\S]*?)<\\\\/\\\\1>`,\n      'i',\n    )\n    const inner = pattern.exec(html)?.[2]\n    return inner ? htmlToText(inner).slice(0, 200_000) : ''\n  }\n  const escaped = selector.value.replace(/[.*+?^${}()|[\\]\\\\]/g, '\\\\$&')\n  const pattern = new RegExp(\n    `<([a-z][a-z0-9]*)\\\\b[^>]*\\\\sclass=[\"'][^\"']*\\\\b${escaped}\\\\b[^\"']*[\"'][^>]*>([\\\\s\\\\S]*?)<\\\\/\\\\1>`,\n    'i',\n  )\n  const inner = pattern.exec(html)?.[2]\n  return inner ? htmlToText(inner).slice(0, 200_000) : ''\n}\n\nfunction joinUrl(base: string, path: string): string {\n  try {\n    return new URL(path, base.endsWith('/') ? base : `${base}/`).toString()\n  } catch {\n    return `${base.replace(/\\/+$/, '')}/${path.replace(/^\\/+/, '')}`\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;AAqBA,SAAgB,WAAW,MAAsB;CAC/C,OAAO,KACJ,QAAQ,+BAA+B,EAAE,CAAC,CAC1C,QAAQ,6BAA6B,EAAE,CAAC,CACxC,QAAQ,mCAAmC,EAAE,CAAC,CAC9C,QAAQ,sBAAsB,EAAE,CAAC,CACjC,QAAQ,mBAAmB,IAAI,CAAC,CAChC,QAAQ,yDAAyD,IAAI,CAAC,CACtE,QAAQ,YAAY,EAAE,CAAC,CACvB,QAAQ,YAAY,GAAG,CAAC,CACxB,QAAQ,WAAW,GAAG,CAAC,CACvB,QAAQ,UAAU,GAAG,CAAC,CACtB,QAAQ,UAAU,GAAG,CAAC,CACtB,QAAQ,YAAY,IAAG,CAAC,CACxB,QAAQ,WAAW,GAAG,CAAC,CACvB,QAAQ,YAAY,GAAG,CAAC,CACxB,QAAQ,aAAa,GAAG,CAAC,CACzB,QAAQ,aAAa,GAAG,CAAC,CACzB,QAAQ,cAAc,GAAG,SAAS,OAAO,cAAc,OAAO,IAAI,CAAC,CAAC,CAAC,CACrE,QAAQ,sBAAsB,GAAG,SAAS,OAAO,cAAc,OAAO,SAAS,MAAM,EAAE,CAAC,CAAC,CAAC,CAC1F,MAAM,IAAI,CAAC,CACX,KAAK,SAAS,KAAK,QAAQ,YAAY,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,CACnD,QAAQ,MAAM,KAAK,QAAQ,EAAE,SAAS,MAAM,IAAI,MAAM,OAAO,GAAG,CAAC,CACjE,KAAK,IAAI,CAAC,CACV,KAAK;AACV;;AAGA,SAAgB,WAAW,MAAc,SAAqC;CAC5E,OAAO,QAAQ,KAAK,IAAI,CAAC,GAAG,EAAE,EAAE,KAAK;AACvC;;AAGA,SAAgB,cAAc,MAAc,IAAgC;CAC1E,MAAM,UAAU,GAAG,QAAQ,uBAAuB,MAAM;CAKxD,OAAO,IAJgB,OACrB,sCAAsC,QAAQ,iCAC9C,GAEc,CAAC,CAAC,KAAK,IAAI,CAAC,GAAG;AACjC;;;;;AAMA,SAAgB,aACd,MACA,aACA,SACkC;CAClC,MAAM,MAAwC,CAAC;CAE/C,KAAK,MAAM,SAAS,KAAK,SAAS,yDAAM,GAAG;EACzC,MAAM,OAAO,MAAM;EACnB,MAAM,QAAQ,MAAM;EACpB,IAAI,CAAC,QAAQ,CAAC,OAAO;EACrB,IAAI,CAAC,YAAY,KAAK,IAAI,GAAG;EAC7B,MAAM,OAAO,WAAW,KAAK;EAC7B,IAAI,CAAC,MAAM;EACX,IAAI;GACF,IAAI,KAAK;IAAE,MAAM,IAAI,IAAI,MAAM,OAAO,CAAC,CAAC,SAAS;IAAG;GAAK,CAAC;EAC5D,QAAQ,CAER;CACF;CACA,OAAO;AACT;;;;;;;;;;;;;ACzEA,MAAa,oBACX;;AAGF,MAAa,qBAAqB;;AAGlC,MAAa,qBAAqB,IAAI,OAAO;AAE7C,MAAM,+BAAe,IAAI,IAA2B;;;;;;;;;;AAiDpD,eAAsB,YACpB,KACA,UAA8B,CAAC,GACH;CAC5B,MAAM,WAAW,QAAQ,cAAc,OAAU;CACjD,MAAM,SAAS,QAAQ,WAAW,MAAM,UAAU,QAAQ,UAAU,KAAK,QAAQ,IAAI,KAAA;CACrF,IAAI,QAAQ,OAAO;CAEnB,MAAM,OAAO,SAAS,GAAG;CACzB,MAAM,aAAa,IAAI;CAEvB,MAAM,6BAAY,IAAI,KAAK,EAAA,CAAE,YAAY;CACzC,IAAI;CACJ,IAAI;EACF,WAAW,MAAM,MAAM,KAAK;GAC1B,QAAQ,QAAQ;GAChB,UAAU;GACV,SAAS;IACP,cAAc;IACd,QAAQ;IACR,mBAAmB;IACnB,GAAI,QAAQ,WAAW,CAAC;GAC1B;EACF,CAAC;CACH,SAAS,OAAO;EACd,IAAK,MAA4B,SAAS,cAAc,MAAM;EAC9D,MAAM,SAA4B;GAChC;GACA,QAAQ;GACR,MAAM;GACN,iBAAiB;GACjB;GACA,WAAW;GACX,YAAY;GACZ,oBAAoB,kBAAmB,MAAgB;EACzD;EACA,IAAI,QAAQ,UAAU,MAAM,WAAW,QAAQ,UAAU,KAAK,MAAM;EACpE,OAAO;CACT;CAEA,MAAM,OAAO,MAAM,gBAAgB,QAAQ;CAC3C,MAAM,eAAe,SAAS,QAAQ,IAAI,eAAe;CACzD,MAAM,aAAa,SAAS,QAAQ,IAAI,MAAM;CAC9C,MAAM,kBAAkB,cAAc,YAAY,KAAK,cAAc,UAAU,KAAK;CAEpF,MAAM,SAA4B;EAChC;EACA,QAAQ,SAAS;EACjB,MAAM;EACN;EACA;EACA,WAAW;EACX,YAAY;CACd;CAEA,IAAI,SAAS,SAAS,OAAO,SAAS,UAAU,KAAK;EACnD,OAAO,aAAa;EACpB,OAAO,qBAAqB,mBAAmB,SAAS;CAC1D,OAAO,IAAI,mBAAmB,IAAI,GAAG;EACnC,OAAO,aAAa;EACpB,OAAO,qBAAqB;CAC9B,OAAO,IAAI,KAAK,SAAS,OAAO,oBAAoB,IAAI,GAAG;EACzD,OAAO,aAAa;EACpB,OAAO,qBAAqB,+BAA+B,KAAK,OAAO;CACzE;CAEA,IAAI,QAAQ,UAAU,MAAM,WAAW,QAAQ,UAAU,KAAK,MAAM;CACpE,OAAO;AACT;;AAGA,SAAgB,sBAA4B;CAC1C,aAAa,MAAM;AACrB;AAEA,SAAS,SAAS,KAAqB;CACrC,IAAI;EACF,OAAO,IAAI,IAAI,GAAG,CAAC,CAAC;CACtB,QAAQ;EACN,OAAO;CACT;AACF;AAEA,eAAe,aAAa,MAA6B;CACvD,MAAM,OAAO,aAAa,IAAI,IAAI,KAAK,QAAQ,QAAQ;CACvD,IAAI,gBAA4B,CAAC;CACjC,MAAM,OAAO,IAAI,SAAe,YAAY;EAC1C,UAAU;CACZ,CAAC;CACD,aAAa,IACX,MACA,KAAK,WAAW,IAAI,CACtB;CACA,MAAM;CACN,WAAW,SAAS,kBAAkB;AACxC;AAEA,eAAe,gBAAgB,UAAqC;CAClE,IAAI,CAAC,SAAS,MAAM,OAAO;CAC3B,MAAM,SAAS,SAAS,KAAK,UAAU;CACvC,MAAM,SAAuB,CAAC;CAC9B,IAAI,QAAQ;CACZ,OAAO,MAAM;EACX,MAAM,EAAE,MAAM,UAAU,MAAM,OAAO,KAAK;EAC1C,IAAI,MAAM;EACV,IAAI,CAAC,OAAO;EACZ,SAAS,MAAM;EACf,IAAI,QAAA,SAA4B;GAE9B,MAAM,OAAO,OAAO;GACpB;EACF;EACA,OAAO,KAAK,KAAK;CACnB;CACA,MAAM,SAAS,IAAI,WAAW,KAAK,IAAI,OAAO,kBAAkB,CAAC;CACjE,IAAI,SAAS;CACb,KAAK,MAAM,SAAS,QAAQ;EAC1B,MAAM,OAAO,KAAK,IAAI,MAAM,QAAQ,OAAO,SAAS,MAAM;EAC1D,IAAI,QAAQ,GAAG;EACf,OAAO,IAAI,MAAM,SAAS,GAAG,IAAI,GAAG,MAAM;EAC1C,UAAU;CACZ;CACA,OAAO,IAAI,YAAY,SAAS,EAAE,OAAO,MAAM,CAAC,CAAC,CAAC,OAAO,MAAM;AACjE;AAEA,SAAS,cAAc,OAA0C;CAC/D,IAAI,CAAC,OAAO,OAAO,KAAA;CACnB,MAAM,KAAK,KAAK,MAAM,KAAK;CAC3B,OAAO,OAAO,SAAS,EAAE,IAAI,IAAI,KAAK,EAAE,CAAC,CAAC,YAAY,IAAI,KAAA;AAC5D;;AAGA,SAAgB,mBAAmB,MAAuB;CACxD,IAAI,CAAC,MAAM,OAAO;CAClB,MAAM,QAAQ,KAAK,YAAY;CAY/B,KAAK,MAAM,UAAU;EAVnB;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;CAEyB,GACzB,IAAI,MAAM,SAAS,MAAM,GAAG,OAAO;CAErC,OAAO;AACT;AAEA,SAAS,oBAAoB,MAAuB;CAClD,OACE,KAAK,SAAS,iBAAiB,KAC/B,KAAK,SAAS,SAAS,KACvB,KAAK,SAAS,YAAY,KAC1B,KAAK,SAAS,iBAAiB,KAC/B,KAAK,SAAS,cAAc;AAEhC;AAEA,SAAS,UAAU,UAAkB,KAAqB;CACxD,MAAM,MAAM,OAAO,GAAG;CACtB,OAAO,KAAK,UAAU,QAAQ,GAAG,IAAI,MAAM,GAAG,CAAC,KAAK,GAAG,IAAI,MAAM;AACnE;AAEA,eAAe,UACb,UACA,KACA,OACwC;CACxC,MAAM,OAAO,UAAU,UAAU,GAAG;CACpC,IAAI;EACF,MAAM,OAAO,MAAM,KAAK,IAAI;EAC5B,IAAI,KAAK,IAAI,IAAI,KAAK,UAAU,OAAO,OAAO,KAAA;EAC9C,MAAM,MAAM,MAAM,SAAS,MAAM,MAAM;EAEvC,OAAO;GAAE,GADM,KAAK,MAAM,GACT;GAAG,WAAW;EAAK;CACtC,QAAQ;EACN;CACF;AACF;AAEA,eAAe,WAAW,UAAkB,KAAa,OAAyC;CAChG,MAAM,OAAO,UAAU,UAAU,GAAG;CACpC,MAAM,MAAM,QAAQ,IAAI,GAAG,EAAE,WAAW,KAAK,CAAC;CAC9C,MAAM,UAAU,MAAM,KAAK,UAAU,KAAK,GAAG,MAAM;AACrD;;;;;;;;;;;;;ACrPA,MAAMA,aAAW;;;;;;;;;;;;;;AA0CjB,SAAgB,uBAAuB,SAAmD;CACxF,MAAM,KAAK,QAAQ,MAAM;CACzB,OAAO;EACL;EACA,MAAM;EACN,aACE;EACF,MAAM,MAAM,MAA+C;GACzD,MAAM,QAAQ,KAAK,SAAS,QAAQ,UAAU;GAC9C,MAAM,YAAY,QAAQ,UAAU,MAAM,GAAG,KAAK;GAClD,MAAM,MAA2B,CAAC;GAClC,KAAK,MAAM,YAAY,WACrB,IAAI,KAAK,MAAM,SAAS,IAAI,UAAU,IAAI,CAAC;GAE7C,OAAO;EACT;CACF;AACF;AAEA,eAAe,SACb,UACA,UACA,MAC4B;CAC5B,MAAM,OAAO,SAAS,KAAK,QAAQ,QAAQ,EAAE;CAC7C,MAAM,MACJ,SAAS,SAAS,WAAW,GAAGA,WAAS,eAAe,SAAS,GAAGA,WAAS,OAAO;CAEtF,MAAM,WAAW,MAAM,YAAY,KAAK;EACtC,QAAQ,KAAK;EACb,UAAU,KAAK;CACjB,CAAC;CAED,MAAM,aAAa,GAAG,SAAS,KAAK,GAAG,SAAS;CAChD,MAAM,iBAAiB,SAAS,kBAAkB,sBAAsB,QAAQ;CAEhF,IAAI,CAAC,SAAS,YACZ,OAAO;EACL,IAAI;EACJ,OAAO,eAAe,SAAS,KAAK,GAAG,SAAS;EAChD,MAAM;EACN,UAAU,OAAO,EAAE;EACnB,YAAY;GACV;GACA,iBAAiB,SAAS;GAC1B,WAAW,SAAS;GACpB,cAAc;GACd,YAAY;GACZ,oBAAoB,SAAS;EAC/B;EACA;EACA,UAAU;GAAE;GAAU,QAAQ,SAAS;GAAQ,WAAW,SAAS;EAAU;CAC/E;CAGF,MAAM,OAAO,SAAS;CACtB,MAAM,QAAQC,eAAa,MAAM,QAAQ;CACzC,MAAM,OAAO,YAAY,MAAM,QAAQ;CACvC,MAAM,YAAY,qBAAqB,IAAI,KAAK,SAAS;CAEzD,MAAM,aAAa,KAAK,SAAS;CACjC,OAAO;EACL,IAAI;EACJ;EACA;EACA,UAAU,OAAO,IAAI;EACrB,YAAY;GACV;GACA,iBAAiB;GACjB,WAAW,SAAS;GACpB,cAAc;GACd;GACA,oBAAoB,aAAa,KAAA,IAAY;EAC/C;EACA;EACA,UAAU;GAAE;GAAU,QAAQ,SAAS;GAAQ,WAAW,SAAS;EAAU;CAC/E;AACF;AAEA,SAASA,eAAa,MAAc,UAAsC;CACxE,MAAM,KAAK,yDAAyD,KAAK,IAAI,CAAC,GAAG;CACjF,IAAI,IAAI,OAAO,WAAW,EAAE;CAC5B,MAAM,IAAI,8BAA8B,KAAK,IAAI,CAAC,GAAG;CACrD,IAAI,GAAG,OAAO,WAAW,CAAC,CAAC,CAAC,MAAM,KAAK,CAAC,CAAC,MAAM,eAAe,SAAS;CACvE,OAAO,eAAe,SAAS,KAAK,GAAG,SAAS;AAClD;AAEA,SAAS,YAAY,MAAc,UAAsC;CACvE,IAAI,SAAS,SAAS,UAAU;EAI9B,MAAM,OAAO,4BAA4B,KAAK,IAAI,CAAC,GAAG;EACtD,IAAI,MAAM,OAAO,WAAW,IAAI;EAChC,MAAM,MAAM,cAAc,MAAM,eAAe;EAC/C,IAAI,KAAK,OAAO,WAAW,GAAG;CAChC;CAKA,MAAM,cAAc,cAAc,MAAM,cAAc;CACtD,IAAI,aACF,OAAO,WAAW,YAAY,QAAQ,sBAAsB,EAAE,CAAC;CAEjE,MAAM,YAAY,cAAc,MAAM,mBAAmB;CACzD,IAAI,WACF,OAAO,WAAW,UAAU,QAAQ,sBAAsB,EAAE,CAAC;CAE/D,OAAO,WAAW,IAAI;AACxB;AAEA,SAAS,qBAAqB,MAAkC;CAI9D,MAAM,QAAQ,mCAAmC,KAAK,IAAI,CAAC,GAAG;CAC9D,IAAI,OAAO;EACT,MAAM,IAAI,OAAO,SAAS,OAAO,EAAE;EACnC,IAAI,OAAO,SAAS,CAAC,KAAK,IAAI,QAAQ,sBAAK,IAAI,KAAK,EAAA,CAAE,eAAe,IAAI,GACvE,OAAO,IAAI,KAAK,KAAK,IAAI,GAAG,IAAI,EAAE,CAAC,CAAC,CAAC,YAAY;CAErD;AAEF;AAEA,SAAS,sBAAsB,UAAwC;CACrE,IAAI,SAAS,SAAS,UAAU,OAAO,CAAC,2BAA2B,kBAAkB;CACrF,OAAO,CAAC,kBAAkB;AAC5B;;;;;;;;;;;;;;;;;;;;;;;ACjKA,MAAM,WAAW;AACjB,MAAM,YAAY,GAAG,SAAS;;AAoB9B,MAAa,sBAAsB;CAAC;CAAkB;CAAuB;AAAkB;AAE/F,SAAgB,4BACd,UAAwC,CAAC,GACxB;CACjB,MAAM,KAAK,QAAQ,MAAM;CACzB,MAAM,eAAe,QAAQ,gBAAgB;CAC7C,OAAO;EACL;EACA,MAAM;EACN,aACE;EACF,MAAM,MAAM,MAA+C;GACzD,MAAM,MAA2B,CAAC;GAClC,MAAM,QAAQ,KAAK,SAAS,OAAO;GAEnC,IAAI,gBAAgB,IAAI,SAAS,OAC/B,IAAI,KAAK,MAAM,WAAW,IAAI,IAAI,CAAC;GAErC,KAAK,MAAM,QAAQ,QAAQ,gBAAgB,CAAC,GAAG;IAC7C,IAAI,IAAI,UAAU,OAAO;IACzB,IAAI,KAAK,MAAM,iBAAiB,IAAI,MAAM,IAAI,CAAC;GACjD;GACA,KAAK,MAAM,QAAQ,QAAQ,qBAAqB,CAAC,GAAG;IAClD,IAAI,IAAI,UAAU,OAAO;IACzB,IAAI,KAAK,MAAM,sBAAsB,IAAI,MAAM,IAAI,CAAC;GACtD;GACA,OAAO;EACT;CACF;AACF;AAEA,eAAe,WAAW,UAAkB,MAA6C;CACvF,MAAM,WAAW,MAAM,YAAY,WAAW;EAAE,QAAQ,KAAK;EAAQ,UAAU,KAAK;CAAS,CAAC;CAO9F,MAAM,QALU,SAAS,KAAK,MAAM,2BAAY,KAAK,CAAC,EAAA,CAI/B,KAAK,MAAM,WAAW,CAAC,CAC5B,CAAC,CAChB,QAAQ,MAAM,qBAAqB,KAAK,CAAC,CAAC,CAAC,CAC3C,KAAK,MAAM,CAAC,CACZ,MAAM,GAAG,GAAO;CAEnB,MAAM,aAAa,SAAS,cAAc,KAAK,SAAS;CACxD,OAAO;EACL,IAAI;EACJ,OAAO;EACP;EACA,UAAU,OAAO,IAAI;EACrB,YAAY;GACV,KAAK;GACL,iBAAiB,SAAS;GAC1B,WAAW,SAAS;GACpB,cAAc;GACd;GACA,oBACE,SAAS,uBAAuB,aAAa,KAAA,IAAY;EAC7D;EACA,gBAAgB;EAChB,UAAU;GAAE;GAAU,QAAQ,SAAS;GAAQ,WAAW,SAAS;GAAW,MAAM;EAAQ;CAC9F;AACF;AAEA,eAAe,iBACb,UACA,MACA,MAC4B;CAC5B,MAAM,MAAM,GAAG,SAAS,gBAAgB,KAAK,QAAQ,QAAQ,EAAE;CAC/D,MAAM,WAAW,MAAM,YAAY,KAAK;EAAE,QAAQ,KAAK;EAAQ,UAAU,KAAK;CAAS,CAAC;CAExF,MAAM,QAAQ,aAAa,SAAS,MAAM,mBAAmB,MAAM;CACnE,MAAM,OAAO,mBAAmB,SAAS,IAAI;CAC7C,MAAM,aAAa,SAAS,cAAc,KAAK,SAAS;CAExD,OAAO;EACL,IAAI,eAAe;EACnB;EACA;EACA,UAAU,OAAO,IAAI;EACrB,YAAY;GACV;GACA,iBAAiB,oBAAoB,SAAS,IAAI,KAAK,SAAS;GAChE,WAAW,SAAS;GACpB,cAAc;GACd;GACA,oBACE,SAAS,uBAAuB,aAAa,KAAA,IAAY;EAC7D;EACA,gBAAgB;EAChB,UAAU;GACR;GACA,QAAQ,SAAS;GACjB,WAAW,SAAS;GACpB,MAAM;GACN;EACF;CACF;AACF;AAEA,eAAe,sBACb,UACA,MACA,MAC4B;CAC5B,MAAM,MAAM,GAAG,WAAW,KAAK,WAAW,GAAG,IAAI,OAAO,IAAI;CAC5D,MAAM,WAAW,MAAM,YAAY,KAAK;EAAE,QAAQ,KAAK;EAAQ,UAAU,KAAK;CAAS,CAAC;CACxF,MAAM,OAAO,mBAAmB,SAAS,IAAI;CAC7C,MAAM,aAAa,SAAS,cAAc,KAAK,SAAS;CACxD,OAAO;EACL,IAAI,YAAY;EAChB,OAAO,aAAa,SAAS,MAAM,yBAAyB,MAAM;EAClE;EACA,UAAU,OAAO,IAAI;EACrB,YAAY;GACV;GACA,iBAAiB,SAAS;GAC1B,WAAW,SAAS;GACpB,cAAc;GACd;GACA,oBACE,SAAS,uBACR,aAAa,KAAA,IAAY;EAC9B;EACA,gBAAgB,CAAC,GAAG,qBAAqB,qBAAqB;EAC9D,UAAU;GACR;GACA,QAAQ,SAAS;GACjB,WAAW,SAAS;GACpB,MAAM;GACN;EACF;CACF;AACF;AAEA,SAAS,aAAa,MAAc,UAA0B;CAC5D,MAAM,KAAK,gEAAgE,KAAK,IAAI,CAAC,GAAG;CACxF,IAAI,IAAI,OAAO,WAAW,EAAE;CAC5B,MAAM,QAAQ,8BAA8B,KAAK,IAAI,CAAC,GAAG;CACzD,IAAI,OAAO,OAAO,WAAW,KAAK,CAAC,CAAC,MAAM,KAAK,CAAC,CAAC,MAAM;CACvD,OAAO;AACT;AAEA,SAAS,mBAAmB,MAAsB;CAGhD,MAAM,OAAO,2BAA2B,KAAK,IAAI,CAAC,GAAG;CACrD,IAAI,MAKF,OAAO,WAJO,KACX,QAAQ,yBAAyB,EAAE,CAAC,CACpC,QAAQ,+BAA+B,EAAE,CAAC,CAC1C,QAAQ,+BAA+B,EACpB,CAAC,CAAC,CAAC,MAAM,GAAG,GAAO;CAE3C,MAAM,OAAO,2BAA2B,KAAK,IAAI,CAAC,GAAG;CACrD,OAAO,OAAO,WAAW,IAAI,CAAC,CAAC,MAAM,GAAG,GAAO,IAAI,WAAW,IAAI,CAAC,CAAC,MAAM,GAAG,GAAO;AACtF;AAEA,SAAS,oBAAoB,MAAkC;CAG7D,MAAM,IAAI,mCAAmC,KAAK,IAAI;CACtD,IAAI,IAAI,IAAI;EACV,MAAM,OAAO,OAAO,SAAS,EAAE,IAAI,EAAE;EACrC,IAAI,OAAO,SAAS,IAAI,KAAK,QAAQ,OAAQ,yBAAQ,IAAI,KAAK,EAAA,CAAE,eAAe,IAAI,GACjF,OAAO,IAAI,KAAK,KAAK,IAAI,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,YAAY;CAEtD;AAEF;AAEA,SAAS,WAAW,OAAuB;CACzC,OAAO,WAAW,KAAK;AACzB;;;ACpKA,SAAgB,qBAAqB,QAA+C;CAClF,MAAM,KAAK,OAAO,MAAM,aAAa,OAAO,MAAM,YAAY;CAE9D,OAAO;EACL;EACA,MAHW,OAAO,QAAQ,GAAG,OAAO,MAAM;EAI1C,aAAa,GAAG,OAAO,MAAM;EAC7B,MAAM,MAAM,MAA+C;GACzD,MAAM,QAAQ,KAAK,SAAS,OAAO,SAAS;GAC5C,MAAM,WAAW,OAAO,SAAS,MAAM,GAAG,KAAK;GAC/C,MAAM,MAA2B,CAAC;GAClC,KAAK,MAAM,UAAU,UACnB,IAAI,KAAK,MAAM,YAAY,IAAI,QAAQ,QAAQ,IAAI,CAAC;GAEtD,OAAO;EACT;CACF;AACF;AAEA,eAAe,YACb,UACA,QACA,QACA,MAC4B;CAC5B,MAAM,MAAM,QAAQ,OAAO,SAAS,OAAO,IAAI;CAC/C,MAAM,WAAW,MAAM,YAAY,KAAK;EAAE,QAAQ,KAAK;EAAQ,UAAU,KAAK;CAAS,CAAC;CAExF,MAAM,OAAO,SAAS,aAAa,kBAAkB,SAAS,MAAM,OAAO,QAAQ,IAAI;CACvF,MAAM,aAAa,SAAS,cAAc,KAAK,SAAS;CAExD,OAAO;EACL,IAAI,OAAO;EACX,OAAO,OAAO;EACd;EACA,UAAU,OAAO,IAAI;EACrB,YAAY;GACV;GACA,iBAAiB,SAAS;GAC1B,WAAW,SAAS;GACpB,cAAc,MAAM,OAAO,MAAM,YAAY;GAC7C;GACA,oBACE,SAAS,uBAAuB,aAAa,KAAA,IAAY;EAC7D;EACA,gBAAgB,OAAO,kBAAkB;GACvC;GACA;GACA;EACF;EACA,UAAU;GACR;GACA,QAAQ,SAAS;GACjB,WAAW,SAAS;GACpB,OAAO,OAAO;EAChB;CACF;AACF;AAEA,SAAS,kBAAkB,MAAc,UAA8C;CACrF,IAAI,SAAS,SAAS,SAAS;EAC7B,MAAM,OAAO,2BAA2B,KAAK,IAAI,CAAC,GAAG;EACrD,OAAO,WAAW,QAAQ,IAAI,CAAC,CAAC,MAAM,GAAG,GAAO;CAClD;CACA,IAAI,SAAS,SAAS,SAAS;EAC7B,MAAM,IAAI,SAAS,MAAM,KAAK,IAAI,CAAC,GAAG;EACtC,OAAO,IAAI,WAAW,CAAC,CAAC,CAAC,MAAM,GAAG,GAAO,IAAI;CAC/C;CACA,IAAI,SAAS,SAAS,MAAM;EAC1B,MAAM,UAAU,SAAS,MAAM,QAAQ,uBAAuB,MAAM;EAKpE,MAAM,QAAQ,IAJM,OAClB,sCAAsC,QAAQ,iCAC9C,GAEkB,CAAC,CAAC,KAAK,IAAI,CAAC,GAAG;EACnC,OAAO,QAAQ,WAAW,KAAK,CAAC,CAAC,MAAM,GAAG,GAAO,IAAI;CACvD;CACA,MAAM,UAAU,SAAS,MAAM,QAAQ,uBAAuB,MAAM;CAKpE,MAAM,QAAQ,IAJM,OAClB,kDAAkD,QAAQ,0CAC1D,GAEkB,CAAC,CAAC,KAAK,IAAI,CAAC,GAAG;CACnC,OAAO,QAAQ,WAAW,KAAK,CAAC,CAAC,MAAM,GAAG,GAAO,IAAI;AACvD;AAEA,SAAS,QAAQ,MAAc,MAAsB;CACnD,IAAI;EACF,OAAO,IAAI,IAAI,MAAM,KAAK,SAAS,GAAG,IAAI,OAAO,GAAG,KAAK,EAAE,CAAC,CAAC,SAAS;CACxE,QAAQ;EACN,OAAO,GAAG,KAAK,QAAQ,QAAQ,EAAE,EAAE,GAAG,KAAK,QAAQ,QAAQ,EAAE;CAC/D;AACF"}
+{"version":3,"file":"index.js","names":["BASE_URL","extractTitle"],"sources":["../../src/sources/html.ts","../../src/sources/http.ts","../../src/sources/cornell-lii.ts","../../src/sources/irs-publications.ts","../../src/sources/state-sos.ts"],"sourcesContent":["/**\n * Minimal HTML helpers used by the shipped sources.\n *\n * Deliberately not a full DOM parser: every authority we ship against\n * (Cornell LII, IRS.gov, state SOS portals) has well-behaved server-rendered\n * HTML where regex-based extraction is correct and cheap. Bringing in cheerio\n * would add a 1.5MB dependency to a package whose purpose is shipping\n * primitives, not parsing arbitrary web pages.\n *\n * If a future source needs real DOM traversal, it should depend on its own\n * parser locally rather than promoting one into the package-wide deps.\n *\n * @stable\n */\n\n/**\n * Strip HTML tags, collapse whitespace, decode common entities.\n *\n * Preserves paragraph and line breaks (`</p>`, `<br>`, `</li>`, `</div>`,\n * `</h*>`) as `\\n` so statute text retains its subsection structure.\n */\nexport function htmlToText(html: string): string {\n  return html\n    .replace(/<script[\\s\\S]*?<\\/script>/gi, '')\n    .replace(/<style[\\s\\S]*?<\\/style>/gi, '')\n    .replace(/<noscript[\\s\\S]*?<\\/noscript>/gi, '')\n    .replace(/<!--([\\s\\S]*?)-->/g, '')\n    .replace(/<\\s*br\\s*\\/?>/gi, '\\n')\n    .replace(/<\\/(p|li|div|tr|h[1-6]|blockquote|section|article)>/gi, '\\n')\n    .replace(/<[^>]+>/g, '')\n    .replace(/&nbsp;/gi, ' ')\n    .replace(/&amp;/gi, '&')\n    .replace(/&lt;/gi, '<')\n    .replace(/&gt;/gi, '>')\n    .replace(/&quot;/gi, '\"')\n    .replace(/&#39;/gi, \"'\")\n    .replace(/&sect;/gi, '§')\n    .replace(/&mdash;/gi, '—')\n    .replace(/&ndash;/gi, '–')\n    .replace(/&#(\\d+);/g, (_, code) => String.fromCodePoint(Number(code)))\n    .replace(/&#x([0-9a-f]+);/gi, (_, code) => String.fromCodePoint(Number.parseInt(code, 16)))\n    .split('\\n')\n    .map((line) => line.replace(/[\\t  ]+/g, ' ').trim())\n    .filter((line, idx, all) => !(line === '' && all[idx - 1] === ''))\n    .join('\\n')\n    .trim()\n}\n\n/** Extract the first match of a regex's first capture group, or undefined. */\nexport function firstMatch(html: string, pattern: RegExp): string | undefined {\n  return pattern.exec(html)?.[1]?.trim()\n}\n\n/** Extract the inner HTML of the first matching tag with id `id`. */\nexport function innerHtmlById(html: string, id: string): string | undefined {\n  const escaped = id.replace(/[.*+?^${}()|[\\]\\\\]/g, '\\\\$&')\n  const tagPattern = new RegExp(\n    `<([a-z][a-z0-9]*)\\\\b[^>]*\\\\sid=[\"']${escaped}[\"'][^>]*>([\\\\s\\\\S]*?)<\\\\/\\\\1>`,\n    'i',\n  )\n  return tagPattern.exec(html)?.[2]\n}\n\n/**\n * Extract every (href, text) pair matching the URL regex.\n * Returns absolute URLs by resolving against `baseUrl`.\n */\nexport function extractLinks(\n  html: string,\n  hrefPattern: RegExp,\n  baseUrl: string,\n): { href: string; text: string }[] {\n  const out: { href: string; text: string }[] = []\n  const anchor = /<a\\b[^>]*\\shref=[\"']([^\"']+)[\"'][^>]*>([\\s\\S]*?)<\\/a>/gi\n  for (const match of html.matchAll(anchor)) {\n    const href = match[1]\n    const inner = match[2]\n    if (!href || !inner) continue\n    if (!hrefPattern.test(href)) continue\n    const text = htmlToText(inner)\n    if (!text) continue\n    try {\n      out.push({ href: new URL(href, baseUrl).toString(), text })\n    } catch {\n      /* skip malformed URL */\n    }\n  }\n  return out\n}\n","import { mkdir, readFile, stat, writeFile } from 'node:fs/promises'\nimport { dirname, join } from 'node:path'\nimport { sha256 } from '../ids'\n\n/**\n * Polite HTTP fetcher shared by remote sources.\n *\n * Independent sources share a per-origin throttle because rate-limited sites\n * may return block pages instead of 429 responses. Responses are cached by URL\n * because many publishers omit reliable ETag and Last-Modified headers. Bodies\n * are checked even after a 2xx response because captcha and block pages often\n * use successful status codes.\n */\n\n/** User-Agent string sent on every outbound request. */\nexport const POLITE_USER_AGENT =\n  'agent-knowledge (+https://github.com/tangle-network/agent-knowledge)'\n\n/** Minimum gap between successive requests to the same origin (ms). */\nexport const MIN_REQUEST_GAP_MS = 1_000\n\n/** Maximum response body we will buffer in memory (bytes). */\nexport const MAX_RESPONSE_BYTES = 8 * 1024 * 1024\n\nconst hostThrottle = new Map<string, Promise<void>>()\n\nexport interface PoliteFetchOptions {\n  signal?: AbortSignal\n  cacheDir?: string\n  /**\n   * Cache age beyond which we re-fetch. Default 1 hour, long enough to\n   * batch a cron sweep across many selectors, short enough that hourly\n   * authoritative-page changes get picked up next tick.\n   */\n  cacheTtlMs?: number\n  /**\n   * Extra request headers. The fetcher always sets `User-Agent` and\n   * `Accept`; callers can add `Accept-Language` etc.\n   */\n  headers?: Record<string, string>\n}\n\nexport interface PoliteFetchResult {\n  url: string\n  status: number\n  /** Decoded UTF-8 body. Truncated to `MAX_RESPONSE_BYTES`. */\n  body: string\n  /**\n   * Best-effort source-attested timestamp. Reads `Last-Modified`,\n   * falling back to `Date`, falling back to fetch time. Always ISO 8601.\n   */\n  sourceUpdatedAt: string\n  fetchedAt: string\n  /** True iff the response was satisfied from disk cache. */\n  fromCache: boolean\n  /**\n   * False on: non-2xx status, captcha/block page heuristic match, or\n   * decoded body below 200 chars from a host known to serve real content\n   * (Cornell, IRS, state SOS). `unverifiableReason` carries the why.\n   */\n  verifiable: boolean\n  unverifiableReason?: string\n}\n\n/**\n * Fetch one URL with per-host throttling, on-disk cache, and block-page\n * detection. Never throws on network/HTTP failure. It returns a result with\n * `verifiable: false` and `unverifiableReason` set so the caller can decide\n * whether to skip, retry, or surface.\n *\n * Throws ONLY on `AbortError` (caller asked to stop) and on cache-write\n * failures that indicate a misconfigured filesystem.\n */\nexport async function politeFetch(\n  url: string,\n  options: PoliteFetchOptions = {},\n): Promise<PoliteFetchResult> {\n  const cacheTtl = options.cacheTtlMs ?? 60 * 60 * 1000\n  const cached = options.cacheDir ? await readCache(options.cacheDir, url, cacheTtl) : undefined\n  if (cached) return cached\n\n  const host = safeHost(url)\n  await throttleHost(host)\n\n  const fetchedAt = new Date().toISOString()\n  let response: Response\n  try {\n    response = await fetch(url, {\n      signal: options.signal,\n      redirect: 'follow',\n      headers: {\n        'User-Agent': POLITE_USER_AGENT,\n        Accept: 'text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8',\n        'Accept-Language': 'en-US,en;q=0.9',\n        ...(options.headers ?? {}),\n      },\n    })\n  } catch (error) {\n    if ((error as { name?: string }).name === 'AbortError') throw error\n    const result: PoliteFetchResult = {\n      url,\n      status: 0,\n      body: '',\n      sourceUpdatedAt: fetchedAt,\n      fetchedAt,\n      fromCache: false,\n      verifiable: false,\n      unverifiableReason: `network error: ${(error as Error).message}`,\n    }\n    if (options.cacheDir) await writeCache(options.cacheDir, url, result)\n    return result\n  }\n\n  const text = await readBoundedText(response)\n  const lastModified = response.headers.get('last-modified')\n  const dateHeader = response.headers.get('date')\n  const sourceUpdatedAt = parseHttpDate(lastModified) ?? parseHttpDate(dateHeader) ?? fetchedAt\n\n  const result: PoliteFetchResult = {\n    url,\n    status: response.status,\n    body: text,\n    sourceUpdatedAt,\n    fetchedAt,\n    fromCache: false,\n    verifiable: true,\n  }\n\n  if (response.status < 200 || response.status >= 300) {\n    result.verifiable = false\n    result.unverifiableReason = `non-2xx status: ${response.status}`\n  } else if (looksLikeBlockPage(text)) {\n    result.verifiable = false\n    result.unverifiableReason = 'block-page heuristic matched'\n  } else if (text.length < 200 && knownLargeAuthority(host)) {\n    result.verifiable = false\n    result.unverifiableReason = `body shorter than expected (${text.length} chars)`\n  }\n\n  if (options.cacheDir) await writeCache(options.cacheDir, url, result)\n  return result\n}\n\n/** Reset the in-process throttle map. Test-only. */\nexport function __resetHttpThrottle(): void {\n  hostThrottle.clear()\n}\n\nfunction safeHost(url: string): string {\n  try {\n    return new URL(url).host\n  } catch {\n    return 'unknown'\n  }\n}\n\nasync function throttleHost(host: string): Promise<void> {\n  const prev = hostThrottle.get(host) ?? Promise.resolve()\n  let release: () => void = () => {}\n  const next = new Promise<void>((resolve) => {\n    release = resolve\n  })\n  hostThrottle.set(\n    host,\n    prev.then(() => next),\n  )\n  await prev\n  setTimeout(release, MIN_REQUEST_GAP_MS)\n}\n\nasync function readBoundedText(response: Response): Promise<string> {\n  if (!response.body) return ''\n  const reader = response.body.getReader()\n  const chunks: Uint8Array[] = []\n  let total = 0\n  while (true) {\n    const { done, value } = await reader.read()\n    if (done) break\n    if (!value) continue\n    total += value.length\n    if (total > MAX_RESPONSE_BYTES) {\n      // Stop reading; release the underlying connection.\n      await reader.cancel()\n      break\n    }\n    chunks.push(value)\n  }\n  const merged = new Uint8Array(Math.min(total, MAX_RESPONSE_BYTES))\n  let offset = 0\n  for (const chunk of chunks) {\n    const take = Math.min(chunk.length, merged.length - offset)\n    if (take <= 0) break\n    merged.set(chunk.subarray(0, take), offset)\n    offset += take\n  }\n  return new TextDecoder('utf-8', { fatal: false }).decode(merged)\n}\n\nfunction parseHttpDate(value: string | null): string | undefined {\n  if (!value) return undefined\n  const ms = Date.parse(value)\n  return Number.isFinite(ms) ? new Date(ms).toISOString() : undefined\n}\n\n/** Cheap heuristic that catches CAPTCHA, WAF block pages, and \"Just a moment\" interstitials. */\nexport function looksLikeBlockPage(body: string): boolean {\n  if (!body) return false\n  const lower = body.toLowerCase()\n  const markers = [\n    'verify you are human',\n    'please enable javascript and cookies',\n    'just a moment',\n    'access denied',\n    'request unsuccessful',\n    'cf-error-details',\n    'captcha',\n    'incapsula',\n    'pardon our interruption',\n  ]\n  for (const marker of markers) {\n    if (lower.includes(marker)) return true\n  }\n  return false\n}\n\nfunction knownLargeAuthority(host: string): boolean {\n  return (\n    host.endsWith('law.cornell.edu') ||\n    host.endsWith('irs.gov') ||\n    host.endsWith('sos.ca.gov') ||\n    host.endsWith('sos.state.tx.us') ||\n    host.endsWith('sos.state.us')\n  )\n}\n\nfunction cachePath(cacheDir: string, url: string): string {\n  const key = sha256(url)\n  return join(cacheDir, 'http', `${key.slice(0, 2)}`, `${key}.json`)\n}\n\nasync function readCache(\n  cacheDir: string,\n  url: string,\n  ttlMs: number,\n): Promise<PoliteFetchResult | undefined> {\n  const path = cachePath(cacheDir, url)\n  try {\n    const info = await stat(path)\n    if (Date.now() - info.mtimeMs > ttlMs) return undefined\n    const raw = await readFile(path, 'utf8')\n    const parsed = JSON.parse(raw) as PoliteFetchResult\n    return { ...parsed, fromCache: true }\n  } catch {\n    return undefined\n  }\n}\n\nasync function writeCache(cacheDir: string, url: string, value: PoliteFetchResult): Promise<void> {\n  const path = cachePath(cacheDir, url)\n  await mkdir(dirname(path), { recursive: true })\n  await writeFile(path, JSON.stringify(value), 'utf8')\n}\n","import { sha256 } from '../ids'\nimport { htmlToText, innerHtmlById } from './html'\nimport { politeFetch } from './http'\nimport type { FetchOpts, KnowledgeFragment, KnowledgeSource } from './types'\n\n/**\n * Cornell Legal Information Institute (LII) source.\n *\n * Pulls federal US Code sections and Wex encyclopedia entries — the two\n * Cornell LII surfaces an agent typically grounds against. The Wex\n * \"non-compete\" page is the canonical test case for the Ryan-LLC v. FTC\n * vacatur drift the continuous-ingestion story is designed to catch.\n *\n * @stable\n */\n\nconst BASE_URL = 'https://www.law.cornell.edu'\n\nexport interface CornellLiiSelector {\n  /** Either 'uscode' or 'wex'. */\n  kind: 'uscode' | 'wex'\n  /**\n   * For `uscode`: `<title>/<section>` (e.g. `'18/1836'` for DTSA).\n   * For `wex`: the slug (e.g. `'non-compete'`).\n   */\n  path: string\n  /**\n   * Optional pre-declared eval dimensions affected by this section. If\n   * omitted, defaults are chosen from `kind` + path heuristics.\n   */\n  dimensionHints?: string[]\n}\n\nexport interface CornellLiiSourceOptions {\n  /**\n   * Selectors to fetch on each `fetch()` call. The caller (a per-tenant\n   * workspace config, typically) lists exactly the authorities they need\n   * tracked. There is no auto-discovery; that would crawl Cornell at\n   * cron speed, which is what the polite-fetch contract exists to avoid.\n   */\n  selectors: CornellLiiSelector[]\n  /** Source id override; default is `'cornell-lii'`. */\n  id?: string\n}\n\n/**\n * Build a Cornell LII source for the listed selectors.\n *\n * Example: track DTSA + non-compete:\n * ```\n * createCornellLiiSource({\n *   selectors: [\n *     { kind: 'uscode', path: '18/1836' },\n *     { kind: 'wex', path: 'non-compete', dimensionHints: ['jurisdictional_accuracy'] },\n *   ],\n * })\n * ```\n */\nexport function createCornellLiiSource(options: CornellLiiSourceOptions): KnowledgeSource {\n  const id = options.id ?? 'cornell-lii'\n  return {\n    id,\n    name: 'Cornell Legal Information Institute',\n    description:\n      'Federal US Code sections (uscode/text/...) and Wex legal encyclopedia entries from law.cornell.edu.',\n    async fetch(opts: FetchOpts): Promise<KnowledgeFragment[]> {\n      const limit = opts.limit ?? options.selectors.length\n      const selectors = options.selectors.slice(0, limit)\n      const out: KnowledgeFragment[] = []\n      for (const selector of selectors) {\n        out.push(await fetchOne(id, selector, opts))\n      }\n      return out\n    },\n  }\n}\n\nasync function fetchOne(\n  sourceId: string,\n  selector: CornellLiiSelector,\n  opts: FetchOpts,\n): Promise<KnowledgeFragment> {\n  const path = selector.path.replace(/^\\/+/, '')\n  const url =\n    selector.kind === 'uscode' ? `${BASE_URL}/uscode/text/${path}` : `${BASE_URL}/wex/${path}`\n\n  const response = await politeFetch(url, {\n    signal: opts.signal,\n    cacheDir: opts.cacheDir,\n  })\n\n  const fragmentId = `${selector.kind}:${selector.path}`\n  const dimensionHints = selector.dimensionHints ?? defaultDimensionHints(selector)\n\n  if (!response.verifiable) {\n    return {\n      id: fragmentId,\n      title: `Cornell LII ${selector.kind} ${selector.path}`,\n      body: '',\n      bodyHash: sha256(''),\n      provenance: {\n        url,\n        sourceUpdatedAt: response.sourceUpdatedAt,\n        fetchedAt: response.fetchedAt,\n        jurisdiction: 'US-FED',\n        verifiable: false,\n        unverifiableReason: response.unverifiableReason,\n      },\n      dimensionHints,\n      metadata: { sourceId, status: response.status, fromCache: response.fromCache },\n    }\n  }\n\n  const html = response.body\n  const title = extractTitle(html, selector)\n  const body = extractBody(html, selector)\n  const effective = extractEffectiveDate(html) ?? response.sourceUpdatedAt\n\n  const verifiable = body.length > 50\n  return {\n    id: fragmentId,\n    title,\n    body,\n    bodyHash: sha256(body),\n    provenance: {\n      url,\n      sourceUpdatedAt: effective,\n      fetchedAt: response.fetchedAt,\n      jurisdiction: 'US-FED',\n      verifiable,\n      unverifiableReason: verifiable ? undefined : 'extracted body too short',\n    },\n    dimensionHints,\n    metadata: { sourceId, status: response.status, fromCache: response.fromCache },\n  }\n}\n\nfunction extractTitle(html: string, selector: CornellLiiSelector): string {\n  const h1 = /<h1[^>]*\\bid=[\"']page_title[\"'][^>]*>([\\s\\S]*?)<\\/h1>/i.exec(html)?.[1]\n  if (h1) return htmlToText(h1)\n  const t = /<title>([\\s\\S]*?)<\\/title>/i.exec(html)?.[1]\n  if (t) return htmlToText(t).split(' | ')[0] ?? `Cornell LII ${selector.path}`\n  return `Cornell LII ${selector.kind} ${selector.path}`\n}\n\nfunction extractBody(html: string, selector: CornellLiiSelector): string {\n  if (selector.kind === 'uscode') {\n    // The statute text lives inside a <text><div class=\"text\">…</div></text>\n    // block on US Code section pages. Prefer it; fall back to #tab_default_1\n    // which always contains the section body.\n    const text = /<text>([\\s\\S]*?)<\\/text>/i.exec(html)?.[1]\n    if (text) return htmlToText(text)\n    const tab = innerHtmlById(html, 'tab_default_1')\n    if (tab) return htmlToText(tab)\n  }\n  // Wex pages wrap the encyclopedia entry under <div id=\"main-content\"> (newer\n  // Drupal template) or directly inside <div id=\"extracted-content\"> (older\n  // template). Try both — lazy regex matching against a nested-div container\n  // returns the wrong (shorter) slice, so we anchor on the leaf containers.\n  const mainContent = innerHtmlById(html, 'main-content')\n  if (mainContent) {\n    return htmlToText(mainContent.replace(/<h1[\\s\\S]*?<\\/h1>/i, ''))\n  }\n  const extracted = innerHtmlById(html, 'extracted-content')\n  if (extracted) {\n    return htmlToText(extracted.replace(/<h1[\\s\\S]*?<\\/h1>/i, ''))\n  }\n  return htmlToText(html)\n}\n\nfunction extractEffectiveDate(html: string): string | undefined {\n  // Cornell LII includes \"Editorial Notes\" / \"Amendments\" blocks with\n  // dates; the most reliable machine-readable signal is the last\n  // amendment year embedded near the section text.\n  const amend = /Amendments[\\s\\S]{0,200}?(\\d{4})/i.exec(html)?.[1]\n  if (amend) {\n    const y = Number.parseInt(amend, 10)\n    if (Number.isFinite(y) && y > 1900 && y <= new Date().getUTCFullYear() + 1) {\n      return new Date(Date.UTC(y, 11, 31)).toISOString()\n    }\n  }\n  return undefined\n}\n\nfunction defaultDimensionHints(selector: CornellLiiSelector): string[] {\n  if (selector.kind === 'uscode') return ['jurisdictional_accuracy', 'citation_hygiene']\n  return ['citation_hygiene']\n}\n","import { sha256 } from '../ids'\nimport { htmlToText } from './html'\nimport { politeFetch } from './http'\nimport type { FetchOpts, KnowledgeFragment, KnowledgeSource } from './types'\n\n/**\n * IRS publications source.\n *\n * Two surfaces:\n *\n *   1. The publications index at https://www.irs.gov/publications enumerates\n *      every active publication with its revision year — a single fragment\n *      with the full table lets change detection notice when a publication\n *      year flips (e.g. Pub 15 (2025) → Pub 15 (2026)).\n *\n *   2. Individual publication landing pages at /publications/p<N>[<suffix>]\n *      return one fragment per publication with summary text. Callers list\n *      the publications they need tracked via `selectors`.\n *\n * Revenue procedures are fetched under their numbered URLs; the IRS does\n * not maintain a stable HTML index of rev-procs, so the caller passes the\n * specific rev-proc paths they care about.\n *\n * @stable\n */\n\nconst BASE_URL = 'https://www.irs.gov'\nconst INDEX_URL = `${BASE_URL}/publications`\n\nexport interface IrsPublicationsSourceOptions {\n  /**\n   * Specific publication slugs to fetch (e.g. `['p15', 'p17', 'p463']`).\n   * When `includeIndex` is true (default), the publications index page is\n   * also fetched as a single fragment so change detection can notice\n   * year/revision shifts across the whole catalogue.\n   */\n  publications?: string[]\n  /**\n   * Revenue procedure paths to fetch (e.g. `['/irb/2024-31_IRB']`). The\n   * caller passes the exact path; this source does not auto-discover.\n   */\n  revenueProcedures?: string[]\n  includeIndex?: boolean\n  id?: string\n}\n\n/** Default eval dimensions for IRS-sourced fragments. */\nexport const IRS_DIMENSION_HINTS = ['tax_compliance', 'regulatory_currency', 'citation_hygiene']\n\nexport function createIrsPublicationsSource(\n  options: IrsPublicationsSourceOptions = {},\n): KnowledgeSource {\n  const id = options.id ?? 'irs-publications'\n  const includeIndex = options.includeIndex ?? true\n  return {\n    id,\n    name: 'IRS Publications',\n    description:\n      'Internal Revenue Service publications index and individual publication landing pages from irs.gov.',\n    async fetch(opts: FetchOpts): Promise<KnowledgeFragment[]> {\n      const out: KnowledgeFragment[] = []\n      const limit = opts.limit ?? Number.POSITIVE_INFINITY\n\n      if (includeIndex && out.length < limit) {\n        out.push(await fetchIndex(id, opts))\n      }\n      for (const slug of options.publications ?? []) {\n        if (out.length >= limit) break\n        out.push(await fetchPublication(id, slug, opts))\n      }\n      for (const path of options.revenueProcedures ?? []) {\n        if (out.length >= limit) break\n        out.push(await fetchRevenueProcedure(id, path, opts))\n      }\n      return out\n    },\n  }\n}\n\nasync function fetchIndex(sourceId: string, opts: FetchOpts): Promise<KnowledgeFragment> {\n  const response = await politeFetch(INDEX_URL, { signal: opts.signal, cacheDir: opts.cacheDir })\n  const tablePattern = /<table[\\s\\S]*?<\\/table>/gi\n  const matches = response.body.match(tablePattern) ?? []\n  // Extract the table that lists current-year publications. IRS publishes\n  // one table per year on the index; the most recent table is always the\n  // first that mentions a year ≥ current.\n  const tables = matches.map((t) => htmlToText(t))\n  const body = tables\n    .filter((t) => /Publication\\s*\\d+/i.test(t))\n    .join('\\n\\n')\n    .slice(0, 200_000)\n\n  const verifiable = response.verifiable && body.length > 200\n  return {\n    id: 'index',\n    title: 'IRS Publications Index',\n    body,\n    bodyHash: sha256(body),\n    provenance: {\n      url: INDEX_URL,\n      sourceUpdatedAt: response.sourceUpdatedAt,\n      fetchedAt: response.fetchedAt,\n      jurisdiction: 'US-FED',\n      verifiable,\n      unverifiableReason:\n        response.unverifiableReason ?? (verifiable ? undefined : 'no publication rows extracted'),\n    },\n    dimensionHints: IRS_DIMENSION_HINTS,\n    metadata: { sourceId, status: response.status, fromCache: response.fromCache, kind: 'index' },\n  }\n}\n\nasync function fetchPublication(\n  sourceId: string,\n  slug: string,\n  opts: FetchOpts,\n): Promise<KnowledgeFragment> {\n  const url = `${BASE_URL}/publications/${slug.replace(/^\\/+/, '')}`\n  const response = await politeFetch(url, { signal: opts.signal, cacheDir: opts.cacheDir })\n\n  const title = extractTitle(response.body, `IRS Publication ${slug}`)\n  const body = extractMainContent(response.body)\n  const verifiable = response.verifiable && body.length > 200\n\n  return {\n    id: `publication:${slug}`,\n    title,\n    body,\n    bodyHash: sha256(body),\n    provenance: {\n      url,\n      sourceUpdatedAt: extractRevisionDate(response.body) ?? response.sourceUpdatedAt,\n      fetchedAt: response.fetchedAt,\n      jurisdiction: 'US-FED',\n      verifiable,\n      unverifiableReason:\n        response.unverifiableReason ?? (verifiable ? undefined : 'no publication body extracted'),\n    },\n    dimensionHints: IRS_DIMENSION_HINTS,\n    metadata: {\n      sourceId,\n      status: response.status,\n      fromCache: response.fromCache,\n      kind: 'publication',\n      slug,\n    },\n  }\n}\n\nasync function fetchRevenueProcedure(\n  sourceId: string,\n  path: string,\n  opts: FetchOpts,\n): Promise<KnowledgeFragment> {\n  const url = `${BASE_URL}${path.startsWith('/') ? path : `/${path}`}`\n  const response = await politeFetch(url, { signal: opts.signal, cacheDir: opts.cacheDir })\n  const body = extractMainContent(response.body)\n  const verifiable = response.verifiable && body.length > 200\n  return {\n    id: `rev-proc:${path}`,\n    title: extractTitle(response.body, `IRS Revenue Procedure ${path}`),\n    body,\n    bodyHash: sha256(body),\n    provenance: {\n      url,\n      sourceUpdatedAt: response.sourceUpdatedAt,\n      fetchedAt: response.fetchedAt,\n      jurisdiction: 'US-FED',\n      verifiable,\n      unverifiableReason:\n        response.unverifiableReason ??\n        (verifiable ? undefined : 'no revenue-procedure body extracted'),\n    },\n    dimensionHints: [...IRS_DIMENSION_HINTS, 'procedural_currency'],\n    metadata: {\n      sourceId,\n      status: response.status,\n      fromCache: response.fromCache,\n      kind: 'rev-proc',\n      path,\n    },\n  }\n}\n\nfunction extractTitle(html: string, fallback: string): string {\n  const og = /<meta\\s+property=[\"']og:title[\"']\\s+content=[\"']([^\"']+)[\"']/i.exec(html)?.[1]\n  if (og) return decodeHtml(og)\n  const title = /<title>([\\s\\S]*?)<\\/title>/i.exec(html)?.[1]\n  if (title) return htmlToText(title).split(' | ')[0] ?? fallback\n  return fallback\n}\n\nfunction extractMainContent(html: string): string {\n  // IRS uses Drupal — the main publication body is inside <main role=\"main\">\n  // or under .field--name-body. We try main first; on miss, body.\n  const main = /<main\\b[\\s\\S]*?<\\/main>/i.exec(html)?.[0]\n  if (main) {\n    const noNav = main\n      .replace(/<nav[\\s\\S]*?<\\/nav>/gi, '')\n      .replace(/<header[\\s\\S]*?<\\/header>/gi, '')\n      .replace(/<footer[\\s\\S]*?<\\/footer>/gi, '')\n    return htmlToText(noNav).slice(0, 200_000)\n  }\n  const body = /<body\\b[\\s\\S]*?<\\/body>/i.exec(html)?.[0]\n  return body ? htmlToText(body).slice(0, 200_000) : htmlToText(html).slice(0, 200_000)\n}\n\nfunction extractRevisionDate(html: string): string | undefined {\n  // IRS publication pages typically show \"Publication X (YYYY)\" in the title;\n  // pulling the year gives a stable revision marker.\n  const m = /Publication\\s+\\S+\\s*\\((\\d{4})\\)/i.exec(html)\n  if (m?.[1]) {\n    const year = Number.parseInt(m[1], 10)\n    if (Number.isFinite(year) && year >= 2000 && year <= new Date().getUTCFullYear() + 1) {\n      return new Date(Date.UTC(year, 0, 1)).toISOString()\n    }\n  }\n  return undefined\n}\n\nfunction decodeHtml(value: string): string {\n  return htmlToText(value)\n}\n","import { sha256 } from '../ids'\nimport { htmlToText } from './html'\nimport { politeFetch } from './http'\nimport type { FetchOpts, KnowledgeFragment, KnowledgeSource } from './types'\n\n/**\n * Generic Secretary-of-State source.\n *\n * Every US state SOS surfaces LLC/Corp formation requirements differently\n * (CA via static forms pages, DE via division of corporations pages, TX\n * via SOSDirect content pages). Rather than baking 50 state-specific\n * parsers into this package, the source takes a config that names the URL\n * pattern + CSS-equivalent selector + jurisdiction tag. Callers supply one\n * config per state they need tracked.\n *\n * The selector is interpreted as a substring/regex of an HTML element id\n * or class — see `StateSosSourceConfig` for the contract. This is\n * intentionally minimal; richer extraction belongs in a state-specific\n * adapter the consumer authors.\n *\n * @experimental Interface will likely grow as we add more state coverage.\n */\n\nexport interface StateSosEntity {\n  /** Stable id for this fragment within the state (e.g. 'llc-formation', 'corp-formation'). */\n  id: string\n  /** Path under the configured `baseUrl` for this entity. */\n  path: string\n  /**\n   * Extraction selector. Choose one:\n   *   - `{ kind: 'id', value: 'main-content' }` — innermost match of element with that id\n   *   - `{ kind: 'class', value: 'field--name-body' }` — innermost match of element with that class\n   *   - `{ kind: 'regex', value: /<article[\\s\\S]*?<\\/article>/i }` — raw regex\n   *   - `{ kind: 'whole' }` — full body, tags stripped (fallback for unstructured pages)\n   */\n  selector:\n    | { kind: 'id'; value: string }\n    | { kind: 'class'; value: string }\n    | { kind: 'regex'; value: RegExp }\n    | { kind: 'whole' }\n  title: string\n  /** Eval dimensions this entity feeds. */\n  dimensionHints?: string[]\n}\n\nexport interface StateSosSourceConfig {\n  /** US state postal code, e.g. 'CA', 'DE', 'TX'. */\n  state: string\n  /** Base URL for the state SOS — e.g. 'https://www.sos.ca.gov'. */\n  baseUrl: string\n  /** Entities this state exposes (LLC, Corp, etc). */\n  entities: StateSosEntity[]\n  /** Source id; default `state-sos:<state>`. */\n  id?: string\n  /** Display name; default `<state> Secretary of State`. */\n  name?: string\n}\n\nexport function createStateSosSource(config: StateSosSourceConfig): KnowledgeSource {\n  const id = config.id ?? `state-sos:${config.state.toLowerCase()}`\n  const name = config.name ?? `${config.state} Secretary of State`\n  return {\n    id,\n    name,\n    description: `${config.state} Secretary of State filings and formation guidance pages.`,\n    async fetch(opts: FetchOpts): Promise<KnowledgeFragment[]> {\n      const limit = opts.limit ?? config.entities.length\n      const entities = config.entities.slice(0, limit)\n      const out: KnowledgeFragment[] = []\n      for (const entity of entities) {\n        out.push(await fetchEntity(id, config, entity, opts))\n      }\n      return out\n    },\n  }\n}\n\nasync function fetchEntity(\n  sourceId: string,\n  config: StateSosSourceConfig,\n  entity: StateSosEntity,\n  opts: FetchOpts,\n): Promise<KnowledgeFragment> {\n  const url = joinUrl(config.baseUrl, entity.path)\n  const response = await politeFetch(url, { signal: opts.signal, cacheDir: opts.cacheDir })\n\n  const body = response.verifiable ? extractBySelector(response.body, entity.selector) : ''\n  const verifiable = response.verifiable && body.length > 100\n\n  return {\n    id: entity.id,\n    title: entity.title,\n    body,\n    bodyHash: sha256(body),\n    provenance: {\n      url,\n      sourceUpdatedAt: response.sourceUpdatedAt,\n      fetchedAt: response.fetchedAt,\n      jurisdiction: `US-${config.state.toUpperCase()}`,\n      verifiable,\n      unverifiableReason:\n        response.unverifiableReason ?? (verifiable ? undefined : 'extracted body too short'),\n    },\n    dimensionHints: entity.dimensionHints ?? [\n      'jurisdictional_accuracy',\n      'corporate_formation',\n      'citation_hygiene',\n    ],\n    metadata: {\n      sourceId,\n      status: response.status,\n      fromCache: response.fromCache,\n      state: config.state,\n    },\n  }\n}\n\nfunction extractBySelector(html: string, selector: StateSosEntity['selector']): string {\n  if (selector.kind === 'whole') {\n    const main = /<main\\b[\\s\\S]*?<\\/main>/i.exec(html)?.[0]\n    return htmlToText(main ?? html).slice(0, 200_000)\n  }\n  if (selector.kind === 'regex') {\n    const m = selector.value.exec(html)?.[0]\n    return m ? htmlToText(m).slice(0, 200_000) : ''\n  }\n  if (selector.kind === 'id') {\n    const escaped = selector.value.replace(/[.*+?^${}()|[\\]\\\\]/g, '\\\\$&')\n    const pattern = new RegExp(\n      `<([a-z][a-z0-9]*)\\\\b[^>]*\\\\sid=[\"']${escaped}[\"'][^>]*>([\\\\s\\\\S]*?)<\\\\/\\\\1>`,\n      'i',\n    )\n    const inner = pattern.exec(html)?.[2]\n    return inner ? htmlToText(inner).slice(0, 200_000) : ''\n  }\n  const escaped = selector.value.replace(/[.*+?^${}()|[\\]\\\\]/g, '\\\\$&')\n  const pattern = new RegExp(\n    `<([a-z][a-z0-9]*)\\\\b[^>]*\\\\sclass=[\"'][^\"']*\\\\b${escaped}\\\\b[^\"']*[\"'][^>]*>([\\\\s\\\\S]*?)<\\\\/\\\\1>`,\n    'i',\n  )\n  const inner = pattern.exec(html)?.[2]\n  return inner ? htmlToText(inner).slice(0, 200_000) : ''\n}\n\nfunction joinUrl(base: string, path: string): string {\n  try {\n    return new URL(path, base.endsWith('/') ? base : `${base}/`).toString()\n  } catch {\n    return `${base.replace(/\\/+$/, '')}/${path.replace(/^\\/+/, '')}`\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;AAqBA,SAAgB,WAAW,MAAsB;CAC/C,OAAO,KACJ,QAAQ,+BAA+B,EAAE,CAAC,CAC1C,QAAQ,6BAA6B,EAAE,CAAC,CACxC,QAAQ,mCAAmC,EAAE,CAAC,CAC9C,QAAQ,sBAAsB,EAAE,CAAC,CACjC,QAAQ,mBAAmB,IAAI,CAAC,CAChC,QAAQ,yDAAyD,IAAI,CAAC,CACtE,QAAQ,YAAY,EAAE,CAAC,CACvB,QAAQ,YAAY,GAAG,CAAC,CACxB,QAAQ,WAAW,GAAG,CAAC,CACvB,QAAQ,UAAU,GAAG,CAAC,CACtB,QAAQ,UAAU,GAAG,CAAC,CACtB,QAAQ,YAAY,IAAG,CAAC,CACxB,QAAQ,WAAW,GAAG,CAAC,CACvB,QAAQ,YAAY,GAAG,CAAC,CACxB,QAAQ,aAAa,GAAG,CAAC,CACzB,QAAQ,aAAa,GAAG,CAAC,CACzB,QAAQ,cAAc,GAAG,SAAS,OAAO,cAAc,OAAO,IAAI,CAAC,CAAC,CAAC,CACrE,QAAQ,sBAAsB,GAAG,SAAS,OAAO,cAAc,OAAO,SAAS,MAAM,EAAE,CAAC,CAAC,CAAC,CAC1F,MAAM,IAAI,CAAC,CACX,KAAK,SAAS,KAAK,QAAQ,YAAY,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,CACnD,QAAQ,MAAM,KAAK,QAAQ,EAAE,SAAS,MAAM,IAAI,MAAM,OAAO,GAAG,CAAC,CACjE,KAAK,IAAI,CAAC,CACV,KAAK;AACV;;AAGA,SAAgB,WAAW,MAAc,SAAqC;CAC5E,OAAO,QAAQ,KAAK,IAAI,CAAC,GAAG,EAAE,EAAE,KAAK;AACvC;;AAGA,SAAgB,cAAc,MAAc,IAAgC;CAC1E,MAAM,UAAU,GAAG,QAAQ,uBAAuB,MAAM;CAKxD,OAAO,IAJgB,OACrB,sCAAsC,QAAQ,iCAC9C,GAEc,CAAC,CAAC,KAAK,IAAI,CAAC,GAAG;AACjC;;;;;AAMA,SAAgB,aACd,MACA,aACA,SACkC;CAClC,MAAM,MAAwC,CAAC;CAE/C,KAAK,MAAM,SAAS,KAAK,SAAS,yDAAM,GAAG;EACzC,MAAM,OAAO,MAAM;EACnB,MAAM,QAAQ,MAAM;EACpB,IAAI,CAAC,QAAQ,CAAC,OAAO;EACrB,IAAI,CAAC,YAAY,KAAK,IAAI,GAAG;EAC7B,MAAM,OAAO,WAAW,KAAK;EAC7B,IAAI,CAAC,MAAM;EACX,IAAI;GACF,IAAI,KAAK;IAAE,MAAM,IAAI,IAAI,MAAM,OAAO,CAAC,CAAC,SAAS;IAAG;GAAK,CAAC;EAC5D,QAAQ,CAER;CACF;CACA,OAAO;AACT;;;;;;;;;;;;;ACzEA,MAAa,oBACX;;AAGF,MAAa,qBAAqB;;AAGlC,MAAa,qBAAqB;AAElC,MAAM,+BAAe,IAAI,IAA2B;;;;;;;;;;AAiDpD,eAAsB,YACpB,KACA,UAA8B,CAAC,GACH;CAC5B,MAAM,WAAW,QAAQ,cAAc;CACvC,MAAM,SAAS,QAAQ,WAAW,MAAM,UAAU,QAAQ,UAAU,KAAK,QAAQ,IAAI,KAAA;CACrF,IAAI,QAAQ,OAAO;CAEnB,MAAM,OAAO,SAAS,GAAG;CACzB,MAAM,aAAa,IAAI;CAEvB,MAAM,6BAAY,IAAI,KAAK,EAAA,CAAE,YAAY;CACzC,IAAI;CACJ,IAAI;EACF,WAAW,MAAM,MAAM,KAAK;GAC1B,QAAQ,QAAQ;GAChB,UAAU;GACV,SAAS;IACP,cAAc;IACd,QAAQ;IACR,mBAAmB;IACnB,GAAI,QAAQ,WAAW,CAAC;GAC1B;EACF,CAAC;CACH,SAAS,OAAO;EACd,IAAK,MAA4B,SAAS,cAAc,MAAM;EAC9D,MAAM,SAA4B;GAChC;GACA,QAAQ;GACR,MAAM;GACN,iBAAiB;GACjB;GACA,WAAW;GACX,YAAY;GACZ,oBAAoB,kBAAmB,MAAgB;EACzD;EACA,IAAI,QAAQ,UAAU,MAAM,WAAW,QAAQ,UAAU,KAAK,MAAM;EACpE,OAAO;CACT;CAEA,MAAM,OAAO,MAAM,gBAAgB,QAAQ;CAC3C,MAAM,eAAe,SAAS,QAAQ,IAAI,eAAe;CACzD,MAAM,aAAa,SAAS,QAAQ,IAAI,MAAM;CAC9C,MAAM,kBAAkB,cAAc,YAAY,KAAK,cAAc,UAAU,KAAK;CAEpF,MAAM,SAA4B;EAChC;EACA,QAAQ,SAAS;EACjB,MAAM;EACN;EACA;EACA,WAAW;EACX,YAAY;CACd;CAEA,IAAI,SAAS,SAAS,OAAO,SAAS,UAAU,KAAK;EACnD,OAAO,aAAa;EACpB,OAAO,qBAAqB,mBAAmB,SAAS;CAC1D,OAAO,IAAI,mBAAmB,IAAI,GAAG;EACnC,OAAO,aAAa;EACpB,OAAO,qBAAqB;CAC9B,OAAO,IAAI,KAAK,SAAS,OAAO,oBAAoB,IAAI,GAAG;EACzD,OAAO,aAAa;EACpB,OAAO,qBAAqB,+BAA+B,KAAK,OAAO;CACzE;CAEA,IAAI,QAAQ,UAAU,MAAM,WAAW,QAAQ,UAAU,KAAK,MAAM;CACpE,OAAO;AACT;;AAGA,SAAgB,sBAA4B;CAC1C,aAAa,MAAM;AACrB;AAEA,SAAS,SAAS,KAAqB;CACrC,IAAI;EACF,OAAO,IAAI,IAAI,GAAG,CAAC,CAAC;CACtB,QAAQ;EACN,OAAO;CACT;AACF;AAEA,eAAe,aAAa,MAA6B;CACvD,MAAM,OAAO,aAAa,IAAI,IAAI,KAAK,QAAQ,QAAQ;CACvD,IAAI,gBAA4B,CAAC;CACjC,MAAM,OAAO,IAAI,SAAe,YAAY;EAC1C,UAAU;CACZ,CAAC;CACD,aAAa,IACX,MACA,KAAK,WAAW,IAAI,CACtB;CACA,MAAM;CACN,WAAW,SAAS,kBAAkB;AACxC;AAEA,eAAe,gBAAgB,UAAqC;CAClE,IAAI,CAAC,SAAS,MAAM,OAAO;CAC3B,MAAM,SAAS,SAAS,KAAK,UAAU;CACvC,MAAM,SAAuB,CAAC;CAC9B,IAAI,QAAQ;CACZ,OAAO,MAAM;EACX,MAAM,EAAE,MAAM,UAAU,MAAM,OAAO,KAAK;EAC1C,IAAI,MAAM;EACV,IAAI,CAAC,OAAO;EACZ,SAAS,MAAM;EACf,IAAI,QAAA,SAA4B;GAE9B,MAAM,OAAO,OAAO;GACpB;EACF;EACA,OAAO,KAAK,KAAK;CACnB;CACA,MAAM,SAAS,IAAI,WAAW,KAAK,IAAI,OAAO,kBAAkB,CAAC;CACjE,IAAI,SAAS;CACb,KAAK,MAAM,SAAS,QAAQ;EAC1B,MAAM,OAAO,KAAK,IAAI,MAAM,QAAQ,OAAO,SAAS,MAAM;EAC1D,IAAI,QAAQ,GAAG;EACf,OAAO,IAAI,MAAM,SAAS,GAAG,IAAI,GAAG,MAAM;EAC1C,UAAU;CACZ;CACA,OAAO,IAAI,YAAY,SAAS,EAAE,OAAO,MAAM,CAAC,CAAC,CAAC,OAAO,MAAM;AACjE;AAEA,SAAS,cAAc,OAA0C;CAC/D,IAAI,CAAC,OAAO,OAAO,KAAA;CACnB,MAAM,KAAK,KAAK,MAAM,KAAK;CAC3B,OAAO,OAAO,SAAS,EAAE,IAAI,IAAI,KAAK,EAAE,CAAC,CAAC,YAAY,IAAI,KAAA;AAC5D;;AAGA,SAAgB,mBAAmB,MAAuB;CACxD,IAAI,CAAC,MAAM,OAAO;CAClB,MAAM,QAAQ,KAAK,YAAY;CAY/B,KAAK,MAAM,UAAU;EAVnB;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;CAEyB,GACzB,IAAI,MAAM,SAAS,MAAM,GAAG,OAAO;CAErC,OAAO;AACT;AAEA,SAAS,oBAAoB,MAAuB;CAClD,OACE,KAAK,SAAS,iBAAiB,KAC/B,KAAK,SAAS,SAAS,KACvB,KAAK,SAAS,YAAY,KAC1B,KAAK,SAAS,iBAAiB,KAC/B,KAAK,SAAS,cAAc;AAEhC;AAEA,SAAS,UAAU,UAAkB,KAAqB;CACxD,MAAM,MAAM,OAAO,GAAG;CACtB,OAAO,KAAK,UAAU,QAAQ,GAAG,IAAI,MAAM,GAAG,CAAC,KAAK,GAAG,IAAI,MAAM;AACnE;AAEA,eAAe,UACb,UACA,KACA,OACwC;CACxC,MAAM,OAAO,UAAU,UAAU,GAAG;CACpC,IAAI;EACF,MAAM,OAAO,MAAM,KAAK,IAAI;EAC5B,IAAI,KAAK,IAAI,IAAI,KAAK,UAAU,OAAO,OAAO,KAAA;EAC9C,MAAM,MAAM,MAAM,SAAS,MAAM,MAAM;EAEvC,OAAO;GAAE,GADM,KAAK,MAAM,GACT;GAAG,WAAW;EAAK;CACtC,QAAQ;EACN;CACF;AACF;AAEA,eAAe,WAAW,UAAkB,KAAa,OAAyC;CAChG,MAAM,OAAO,UAAU,UAAU,GAAG;CACpC,MAAM,MAAM,QAAQ,IAAI,GAAG,EAAE,WAAW,KAAK,CAAC;CAC9C,MAAM,UAAU,MAAM,KAAK,UAAU,KAAK,GAAG,MAAM;AACrD;;;;;;;;;;;;;ACrPA,MAAMA,aAAW;;;;;;;;;;;;;;AA0CjB,SAAgB,uBAAuB,SAAmD;CACxF,MAAM,KAAK,QAAQ,MAAM;CACzB,OAAO;EACL;EACA,MAAM;EACN,aACE;EACF,MAAM,MAAM,MAA+C;GACzD,MAAM,QAAQ,KAAK,SAAS,QAAQ,UAAU;GAC9C,MAAM,YAAY,QAAQ,UAAU,MAAM,GAAG,KAAK;GAClD,MAAM,MAA2B,CAAC;GAClC,KAAK,MAAM,YAAY,WACrB,IAAI,KAAK,MAAM,SAAS,IAAI,UAAU,IAAI,CAAC;GAE7C,OAAO;EACT;CACF;AACF;AAEA,eAAe,SACb,UACA,UACA,MAC4B;CAC5B,MAAM,OAAO,SAAS,KAAK,QAAQ,QAAQ,EAAE;CAC7C,MAAM,MACJ,SAAS,SAAS,WAAW,GAAGA,WAAS,eAAe,SAAS,GAAGA,WAAS,OAAO;CAEtF,MAAM,WAAW,MAAM,YAAY,KAAK;EACtC,QAAQ,KAAK;EACb,UAAU,KAAK;CACjB,CAAC;CAED,MAAM,aAAa,GAAG,SAAS,KAAK,GAAG,SAAS;CAChD,MAAM,iBAAiB,SAAS,kBAAkB,sBAAsB,QAAQ;CAEhF,IAAI,CAAC,SAAS,YACZ,OAAO;EACL,IAAI;EACJ,OAAO,eAAe,SAAS,KAAK,GAAG,SAAS;EAChD,MAAM;EACN,UAAU,OAAO,EAAE;EACnB,YAAY;GACV;GACA,iBAAiB,SAAS;GAC1B,WAAW,SAAS;GACpB,cAAc;GACd,YAAY;GACZ,oBAAoB,SAAS;EAC/B;EACA;EACA,UAAU;GAAE;GAAU,QAAQ,SAAS;GAAQ,WAAW,SAAS;EAAU;CAC/E;CAGF,MAAM,OAAO,SAAS;CACtB,MAAM,QAAQC,eAAa,MAAM,QAAQ;CACzC,MAAM,OAAO,YAAY,MAAM,QAAQ;CACvC,MAAM,YAAY,qBAAqB,IAAI,KAAK,SAAS;CAEzD,MAAM,aAAa,KAAK,SAAS;CACjC,OAAO;EACL,IAAI;EACJ;EACA;EACA,UAAU,OAAO,IAAI;EACrB,YAAY;GACV;GACA,iBAAiB;GACjB,WAAW,SAAS;GACpB,cAAc;GACd;GACA,oBAAoB,aAAa,KAAA,IAAY;EAC/C;EACA;EACA,UAAU;GAAE;GAAU,QAAQ,SAAS;GAAQ,WAAW,SAAS;EAAU;CAC/E;AACF;AAEA,SAASA,eAAa,MAAc,UAAsC;CACxE,MAAM,KAAK,yDAAyD,KAAK,IAAI,CAAC,GAAG;CACjF,IAAI,IAAI,OAAO,WAAW,EAAE;CAC5B,MAAM,IAAI,8BAA8B,KAAK,IAAI,CAAC,GAAG;CACrD,IAAI,GAAG,OAAO,WAAW,CAAC,CAAC,CAAC,MAAM,KAAK,CAAC,CAAC,MAAM,eAAe,SAAS;CACvE,OAAO,eAAe,SAAS,KAAK,GAAG,SAAS;AAClD;AAEA,SAAS,YAAY,MAAc,UAAsC;CACvE,IAAI,SAAS,SAAS,UAAU;EAI9B,MAAM,OAAO,4BAA4B,KAAK,IAAI,CAAC,GAAG;EACtD,IAAI,MAAM,OAAO,WAAW,IAAI;EAChC,MAAM,MAAM,cAAc,MAAM,eAAe;EAC/C,IAAI,KAAK,OAAO,WAAW,GAAG;CAChC;CAKA,MAAM,cAAc,cAAc,MAAM,cAAc;CACtD,IAAI,aACF,OAAO,WAAW,YAAY,QAAQ,sBAAsB,EAAE,CAAC;CAEjE,MAAM,YAAY,cAAc,MAAM,mBAAmB;CACzD,IAAI,WACF,OAAO,WAAW,UAAU,QAAQ,sBAAsB,EAAE,CAAC;CAE/D,OAAO,WAAW,IAAI;AACxB;AAEA,SAAS,qBAAqB,MAAkC;CAI9D,MAAM,QAAQ,mCAAmC,KAAK,IAAI,CAAC,GAAG;CAC9D,IAAI,OAAO;EACT,MAAM,IAAI,OAAO,SAAS,OAAO,EAAE;EACnC,IAAI,OAAO,SAAS,CAAC,KAAK,IAAI,QAAQ,sBAAK,IAAI,KAAK,EAAA,CAAE,eAAe,IAAI,GACvE,OAAO,IAAI,KAAK,KAAK,IAAI,GAAG,IAAI,EAAE,CAAC,CAAC,CAAC,YAAY;CAErD;AAEF;AAEA,SAAS,sBAAsB,UAAwC;CACrE,IAAI,SAAS,SAAS,UAAU,OAAO,CAAC,2BAA2B,kBAAkB;CACrF,OAAO,CAAC,kBAAkB;AAC5B;;;;;;;;;;;;;;;;;;;;;;;ACjKA,MAAM,WAAW;AACjB,MAAM,YAAY,GAAG,SAAS;;AAoB9B,MAAa,sBAAsB;CAAC;CAAkB;CAAuB;AAAkB;AAE/F,SAAgB,4BACd,UAAwC,CAAC,GACxB;CACjB,MAAM,KAAK,QAAQ,MAAM;CACzB,MAAM,eAAe,QAAQ,gBAAgB;CAC7C,OAAO;EACL;EACA,MAAM;EACN,aACE;EACF,MAAM,MAAM,MAA+C;GACzD,MAAM,MAA2B,CAAC;GAClC,MAAM,QAAQ,KAAK,SAAS,OAAO;GAEnC,IAAI,gBAAgB,IAAI,SAAS,OAC/B,IAAI,KAAK,MAAM,WAAW,IAAI,IAAI,CAAC;GAErC,KAAK,MAAM,QAAQ,QAAQ,gBAAgB,CAAC,GAAG;IAC7C,IAAI,IAAI,UAAU,OAAO;IACzB,IAAI,KAAK,MAAM,iBAAiB,IAAI,MAAM,IAAI,CAAC;GACjD;GACA,KAAK,MAAM,QAAQ,QAAQ,qBAAqB,CAAC,GAAG;IAClD,IAAI,IAAI,UAAU,OAAO;IACzB,IAAI,KAAK,MAAM,sBAAsB,IAAI,MAAM,IAAI,CAAC;GACtD;GACA,OAAO;EACT;CACF;AACF;AAEA,eAAe,WAAW,UAAkB,MAA6C;CACvF,MAAM,WAAW,MAAM,YAAY,WAAW;EAAE,QAAQ,KAAK;EAAQ,UAAU,KAAK;CAAS,CAAC;CAO9F,MAAM,QALU,SAAS,KAAK,MAAM,2BAAY,KAAK,CAAC,EAAA,CAI/B,KAAK,MAAM,WAAW,CAAC,CAC5B,CAAC,CAChB,QAAQ,MAAM,qBAAqB,KAAK,CAAC,CAAC,CAAC,CAC3C,KAAK,MAAM,CAAC,CACZ,MAAM,GAAG,GAAO;CAEnB,MAAM,aAAa,SAAS,cAAc,KAAK,SAAS;CACxD,OAAO;EACL,IAAI;EACJ,OAAO;EACP;EACA,UAAU,OAAO,IAAI;EACrB,YAAY;GACV,KAAK;GACL,iBAAiB,SAAS;GAC1B,WAAW,SAAS;GACpB,cAAc;GACd;GACA,oBACE,SAAS,uBAAuB,aAAa,KAAA,IAAY;EAC7D;EACA,gBAAgB;EAChB,UAAU;GAAE;GAAU,QAAQ,SAAS;GAAQ,WAAW,SAAS;GAAW,MAAM;EAAQ;CAC9F;AACF;AAEA,eAAe,iBACb,UACA,MACA,MAC4B;CAC5B,MAAM,MAAM,GAAG,SAAS,gBAAgB,KAAK,QAAQ,QAAQ,EAAE;CAC/D,MAAM,WAAW,MAAM,YAAY,KAAK;EAAE,QAAQ,KAAK;EAAQ,UAAU,KAAK;CAAS,CAAC;CAExF,MAAM,QAAQ,aAAa,SAAS,MAAM,mBAAmB,MAAM;CACnE,MAAM,OAAO,mBAAmB,SAAS,IAAI;CAC7C,MAAM,aAAa,SAAS,cAAc,KAAK,SAAS;CAExD,OAAO;EACL,IAAI,eAAe;EACnB;EACA;EACA,UAAU,OAAO,IAAI;EACrB,YAAY;GACV;GACA,iBAAiB,oBAAoB,SAAS,IAAI,KAAK,SAAS;GAChE,WAAW,SAAS;GACpB,cAAc;GACd;GACA,oBACE,SAAS,uBAAuB,aAAa,KAAA,IAAY;EAC7D;EACA,gBAAgB;EAChB,UAAU;GACR;GACA,QAAQ,SAAS;GACjB,WAAW,SAAS;GACpB,MAAM;GACN;EACF;CACF;AACF;AAEA,eAAe,sBACb,UACA,MACA,MAC4B;CAC5B,MAAM,MAAM,GAAG,WAAW,KAAK,WAAW,GAAG,IAAI,OAAO,IAAI;CAC5D,MAAM,WAAW,MAAM,YAAY,KAAK;EAAE,QAAQ,KAAK;EAAQ,UAAU,KAAK;CAAS,CAAC;CACxF,MAAM,OAAO,mBAAmB,SAAS,IAAI;CAC7C,MAAM,aAAa,SAAS,cAAc,KAAK,SAAS;CACxD,OAAO;EACL,IAAI,YAAY;EAChB,OAAO,aAAa,SAAS,MAAM,yBAAyB,MAAM;EAClE;EACA,UAAU,OAAO,IAAI;EACrB,YAAY;GACV;GACA,iBAAiB,SAAS;GAC1B,WAAW,SAAS;GACpB,cAAc;GACd;GACA,oBACE,SAAS,uBACR,aAAa,KAAA,IAAY;EAC9B;EACA,gBAAgB,CAAC,GAAG,qBAAqB,qBAAqB;EAC9D,UAAU;GACR;GACA,QAAQ,SAAS;GACjB,WAAW,SAAS;GACpB,MAAM;GACN;EACF;CACF;AACF;AAEA,SAAS,aAAa,MAAc,UAA0B;CAC5D,MAAM,KAAK,gEAAgE,KAAK,IAAI,CAAC,GAAG;CACxF,IAAI,IAAI,OAAO,WAAW,EAAE;CAC5B,MAAM,QAAQ,8BAA8B,KAAK,IAAI,CAAC,GAAG;CACzD,IAAI,OAAO,OAAO,WAAW,KAAK,CAAC,CAAC,MAAM,KAAK,CAAC,CAAC,MAAM;CACvD,OAAO;AACT;AAEA,SAAS,mBAAmB,MAAsB;CAGhD,MAAM,OAAO,2BAA2B,KAAK,IAAI,CAAC,GAAG;CACrD,IAAI,MAKF,OAAO,WAJO,KACX,QAAQ,yBAAyB,EAAE,CAAC,CACpC,QAAQ,+BAA+B,EAAE,CAAC,CAC1C,QAAQ,+BAA+B,EACxB,CAAK,CAAC,CAAC,MAAM,GAAG,GAAO;CAE3C,MAAM,OAAO,2BAA2B,KAAK,IAAI,CAAC,GAAG;CACrD,OAAO,OAAO,WAAW,IAAI,CAAC,CAAC,MAAM,GAAG,GAAO,IAAI,WAAW,IAAI,CAAC,CAAC,MAAM,GAAG,GAAO;AACtF;AAEA,SAAS,oBAAoB,MAAkC;CAG7D,MAAM,IAAI,mCAAmC,KAAK,IAAI;CACtD,IAAI,IAAI,IAAI;EACV,MAAM,OAAO,OAAO,SAAS,EAAE,IAAI,EAAE;EACrC,IAAI,OAAO,SAAS,IAAI,KAAK,QAAQ,OAAQ,yBAAQ,IAAI,KAAK,EAAA,CAAE,eAAe,IAAI,GACjF,OAAO,IAAI,KAAK,KAAK,IAAI,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,YAAY;CAEtD;AAEF;AAEA,SAAS,WAAW,OAAuB;CACzC,OAAO,WAAW,KAAK;AACzB;;;ACpKA,SAAgB,qBAAqB,QAA+C;CAClF,MAAM,KAAK,OAAO,MAAM,aAAa,OAAO,MAAM,YAAY;CAE9D,OAAO;EACL;EACA,MAHW,OAAO,QAAQ,GAAG,OAAO,MAAM;EAI1C,aAAa,GAAG,OAAO,MAAM;EAC7B,MAAM,MAAM,MAA+C;GACzD,MAAM,QAAQ,KAAK,SAAS,OAAO,SAAS;GAC5C,MAAM,WAAW,OAAO,SAAS,MAAM,GAAG,KAAK;GAC/C,MAAM,MAA2B,CAAC;GAClC,KAAK,MAAM,UAAU,UACnB,IAAI,KAAK,MAAM,YAAY,IAAI,QAAQ,QAAQ,IAAI,CAAC;GAEtD,OAAO;EACT;CACF;AACF;AAEA,eAAe,YACb,UACA,QACA,QACA,MAC4B;CAC5B,MAAM,MAAM,QAAQ,OAAO,SAAS,OAAO,IAAI;CAC/C,MAAM,WAAW,MAAM,YAAY,KAAK;EAAE,QAAQ,KAAK;EAAQ,UAAU,KAAK;CAAS,CAAC;CAExF,MAAM,OAAO,SAAS,aAAa,kBAAkB,SAAS,MAAM,OAAO,QAAQ,IAAI;CACvF,MAAM,aAAa,SAAS,cAAc,KAAK,SAAS;CAExD,OAAO;EACL,IAAI,OAAO;EACX,OAAO,OAAO;EACd;EACA,UAAU,OAAO,IAAI;EACrB,YAAY;GACV;GACA,iBAAiB,SAAS;GAC1B,WAAW,SAAS;GACpB,cAAc,MAAM,OAAO,MAAM,YAAY;GAC7C;GACA,oBACE,SAAS,uBAAuB,aAAa,KAAA,IAAY;EAC7D;EACA,gBAAgB,OAAO,kBAAkB;GACvC;GACA;GACA;EACF;EACA,UAAU;GACR;GACA,QAAQ,SAAS;GACjB,WAAW,SAAS;GACpB,OAAO,OAAO;EAChB;CACF;AACF;AAEA,SAAS,kBAAkB,MAAc,UAA8C;CACrF,IAAI,SAAS,SAAS,SAAS;EAC7B,MAAM,OAAO,2BAA2B,KAAK,IAAI,CAAC,GAAG;EACrD,OAAO,WAAW,QAAQ,IAAI,CAAC,CAAC,MAAM,GAAG,GAAO;CAClD;CACA,IAAI,SAAS,SAAS,SAAS;EAC7B,MAAM,IAAI,SAAS,MAAM,KAAK,IAAI,CAAC,GAAG;EACtC,OAAO,IAAI,WAAW,CAAC,CAAC,CAAC,MAAM,GAAG,GAAO,IAAI;CAC/C;CACA,IAAI,SAAS,SAAS,MAAM;EAC1B,MAAM,UAAU,SAAS,MAAM,QAAQ,uBAAuB,MAAM;EAKpE,MAAM,QAAQ,IAJM,OAClB,sCAAsC,QAAQ,iCAC9C,GAEkB,CAAC,CAAC,KAAK,IAAI,CAAC,GAAG;EACnC,OAAO,QAAQ,WAAW,KAAK,CAAC,CAAC,MAAM,GAAG,GAAO,IAAI;CACvD;CACA,MAAM,UAAU,SAAS,MAAM,QAAQ,uBAAuB,MAAM;CAKpE,MAAM,QAAQ,IAJM,OAClB,kDAAkD,QAAQ,0CAC1D,GAEkB,CAAC,CAAC,KAAK,IAAI,CAAC,GAAG;CACnC,OAAO,QAAQ,WAAW,KAAK,CAAC,CAAC,MAAM,GAAG,GAAO,IAAI;AACvD;AAEA,SAAS,QAAQ,MAAc,MAAsB;CACnD,IAAI;EACF,OAAO,IAAI,IAAI,MAAM,KAAK,SAAS,GAAG,IAAI,OAAO,GAAG,KAAK,EAAE,CAAC,CAAC,SAAS;CACxE,QAAQ;EACN,OAAO,GAAG,KAAK,QAAQ,QAAQ,EAAE,EAAE,GAAG,KAAK,QAAQ,QAAQ,EAAE;CAC/D;AACF"}

package/dist/viz/index.d.ts

@@ -1,38 +1,37 @@
 import { d as KnowledgeGraphNode, l as KnowledgeGraph, u as KnowledgeGraphEdge } from "../types-m2QB86fF.js";
 //#region src/viz/index.d.ts
-interface KnowledgeVizNode extends KnowledgeGraphNode {
+export interface KnowledgeVizNode extends KnowledgeGraphNode {
   degree: number;
   community: number;
 }
-interface KnowledgeVizEdge extends KnowledgeGraphEdge {
+export interface KnowledgeVizEdge extends KnowledgeGraphEdge {
   id: string;
 }
-interface KnowledgeCommunity {
+export interface KnowledgeCommunity {
   id: number;
   nodeIds: string[];
   topTitles: string[];
   cohesion: number;
 }
-interface KnowledgeVizGraph {
+export interface KnowledgeVizGraph {
   nodes: KnowledgeVizNode[];
   edges: KnowledgeVizEdge[];
   communities: KnowledgeCommunity[];
 }
-interface KnowledgeGap {
+export interface KnowledgeGap {
   type: 'isolated-node' | 'sparse-community' | 'bridge-node';
   title: string;
   nodeIds: string[];
   suggestion: string;
 }
-interface SurprisingConnection {
+export interface SurprisingConnection {
   source: KnowledgeVizNode;
   target: KnowledgeVizNode;
   score: number;
   reasons: string[];
 }
-declare function toKnowledgeVizGraph(graph: KnowledgeGraph): KnowledgeVizGraph;
-declare function detectKnowledgeGaps(graph: KnowledgeVizGraph, limit?: number): KnowledgeGap[];
-declare function findSurprisingConnections(graph: KnowledgeVizGraph, limit?: number): SurprisingConnection[];
+export declare function toKnowledgeVizGraph(graph: KnowledgeGraph): KnowledgeVizGraph;
+export declare function detectKnowledgeGaps(graph: KnowledgeVizGraph, limit?: number): KnowledgeGap[];
+export declare function findSurprisingConnections(graph: KnowledgeVizGraph, limit?: number): SurprisingConnection[];
 //#endregion
-export { KnowledgeCommunity, KnowledgeGap, KnowledgeVizEdge, KnowledgeVizGraph, KnowledgeVizNode, SurprisingConnection, detectKnowledgeGaps, findSurprisingConnections, toKnowledgeVizGraph };
 //# sourceMappingURL=index.d.ts.map

package/dist/viz/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/viz/index.ts"],"mappings":";;UAEiB,yBAAyB;EACxC;EACA;;UAGe,yBAAyB;EACxC;;UAGe;EACf;EACA;EACA;EACA;;UAGe;EACf,OAAO;EACP,OAAO;EACP,aAAa;;UAGE;EACf;EACA;EACA;EACA;;UAGe;EACf,QAAQ;EACR,QAAQ;EACR;EACA;;iBAGc,oBAAoB,OAAO,iBAAiB;iBAmB5C,oBAAoB,OAAO,mBAAmB,iBAAa;iBA+C3D,0BACd,OAAO,mBACP,iBACC"}
+{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/viz/index.ts"],"mappings":";;iBAEiB,yBAAyB;EACxC;EACA;;iBAGe,yBAAyB;EACxC;;iBAGe;EACf;EACA;EACA;EACA;;iBAGe;EACf,OAAO;EACP,OAAO;EACP,aAAa;;iBAGE;EACf;EACA;EACA;EACA;;iBAGe;EACf,QAAQ;EACR,QAAQ;EACR;EACA;;wBAGc,oBAAoB,OAAO,iBAAiB;wBAmB5C,oBAAoB,OAAO,mBAAmB,iBAAa;wBA+C3D,0BACd,OAAO,mBACP,iBACC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tangle-network/agent-knowledge",
-  "version": "17.1.10",
+  "version": "18.0.0",
   "description": "Build, search, evaluate, and improve source-backed knowledge bases.",
   "homepage": "https://github.com/tangle-network/agent-knowledge#readme",
   "repository": {
@@ -81,57 +81,32 @@
   "dependencies": {
     "@types/proper-lockfile": "4.1.4",
     "proper-lockfile": "4.1.2",
-    "zod": "4.5.4"
+    "zod": "4.6.5",
+    "@tangle-network/tcloud": ">=0.8.0 <0.9.0"
   },
   "peerDependencies": {
-    "@tangle-network/agent-eval": ">=0.182.0 <0.200.0",
-    "@tangle-network/agent-interface": "^2.0.0"
+    "@tangle-network/agent-eval": ">=0.201.0 <0.204.0",
+    "@tangle-network/agent-interface": "^2.15.0"
   },
   "devDependencies": {
     "@arethetypeswrong/cli": "^0.18.5",
-    "@biomejs/biome": "^2.5.11",
-    "@neo4j-labs/agent-memory": "0.4.1",
-    "@tangle-network/agent-eval": "0.199.0",
-    "@tangle-network/agent-interface": "2.0.0",
-    "@types/node": "^26.4.0",
-    "mem0ai": "3.1.7",
-    "oxc-parser": "0.147.0",
+    "@biomejs/biome": "^2.5.14",
+    "@neo4j-labs/agent-memory": "0.5.0",
+    "@tangle-network/agent-eval": "0.203.0",
+    "@tangle-network/agent-interface": "2.15.0",
+    "@types/node": "^26.6.2",
+    "mem0ai": "3.3.0",
+    "oxc-parser": "0.151.0",
     "publint": "^0.3.24",
-    "tsdown": "^0.22.14",
+    "tsdown": "^0.23.0",
     "typescript": "^7.0.2",
-    "vite": "8.2.2",
-    "vitest": "^4.1.11",
-    "yaml": "2.9.0"
-  },
-  "pnpm": {
-    "ignoredBuiltDependencies": [
-      "@ax-llm/ax",
-      "esbuild"
-    ],
-    "onlyBuiltDependencies": [
-      "better-sqlite3"
-    ],
-    "minimumReleaseAge": 4320,
-    "minimumReleaseAgeExclude": [
-      "@tangle-network/agent-core",
-      "@tangle-network/agent-eval",
-      "@tangle-network/agent-interface",
-      "@tangle-network/agent-trace-contract",
-      "esbuild",
-      "vite",
-      "zod@4.5.4"
-    ],
-    "overrides": {
-      "@hono/node-server": "2.0.12",
-      "esbuild": "0.28.1",
-      "hono": "4.12.32",
-      "vite": "8.2.2",
-      "ws": "8.21.1"
-    }
+    "vite": "8.3.1",
+    "vitest": "^5.0.1",
+    "yaml": "2.9.1"
   },
   "engines": {
-    "node": ">=20.19.0"
+    "node": ">=22.12.0"
   },
   "license": "MIT",
-  "packageManager": "pnpm@10.34.5"
+  "packageManager": "pnpm@12.6.0"
 }