skilld 0.0.1

package/dist/index.mjs

@@ -0,0 +1,181 @@
+import { mkdirSync, writeFileSync } from "node:fs";
+import { join as join$1 } from "node:path";
+async function generateSkill(config, onProgress) {
+	const { url, outputDir = ".skilld", chunkSize = 1e3, chunkOverlap = 200, maxPages = 100, skipLlmsTxt = false, model = "Xenova/bge-small-en-v1.5" } = config;
+	const siteName = getSiteName(url);
+	const skillDir = join$1(outputDir, siteName);
+	const referencesDir = join$1(skillDir, "references");
+	const dbPath = join$1(skillDir, "search.db");
+	mkdirSync(referencesDir, { recursive: true });
+	let docs;
+	let skillContent;
+	if (!skipLlmsTxt) {
+		const llmsResult = await fetchFromLlmsTxt(url, maxPages, onProgress);
+		if (llmsResult) {
+			docs = llmsResult.docs;
+			skillContent = llmsResult.llmsContent;
+		} else docs = await crawlSite(url, maxPages, onProgress);
+	} else docs = await crawlSite(url, maxPages, onProgress);
+	if (docs.length === 0) throw new Error("No documents found to index");
+	const skillPath = join$1(skillDir, "SKILL.md");
+	if (skillContent) writeFileSync(skillPath, skillContent);
+	const { splitText } = await import("./split-text.mjs");
+	const { sqliteVec } = await import("retriv/db/sqlite-vec");
+	const { transformers } = await import("retriv/embeddings/transformers");
+	const documents = [];
+	for (const doc of docs) {
+		const chunks = splitText(doc.content, {
+			chunkSize,
+			chunkOverlap
+		});
+		for (const chunk of chunks) {
+			const section = extractSection(chunk.text);
+			const docId = chunks.length > 1 ? `${doc.url}#chunk-${chunk.index}` : doc.url;
+			const prefix = [doc.title, section].filter(Boolean).join(" > ");
+			const content = prefix ? `${prefix}\n\n${chunk.text}` : chunk.text;
+			documents.push({
+				id: docId,
+				content,
+				metadata: {
+					source: doc.url,
+					title: doc.title,
+					...section && { section },
+					...chunks.length > 1 && {
+						chunkIndex: chunk.index,
+						chunkTotal: chunks.length
+					}
+				}
+			});
+			writeFileSync(join$1(referencesDir, sanitizeFilename(docId) + ".md"), formatReferenceFile(docId, doc, section, chunk, chunks.length));
+		}
+	}
+	onProgress?.({
+		url: "embedding",
+		count: documents.length,
+		phase: "index"
+	});
+	const db = await sqliteVec({
+		path: dbPath,
+		embeddings: transformers({ model })
+	});
+	await db.index(documents);
+	await db.close?.();
+	return {
+		siteName,
+		skillPath,
+		referencesDir,
+		dbPath,
+		chunkCount: documents.length
+	};
+}
+function getSiteName(url) {
+	return new URL(url).hostname.replace(/^www\./, "");
+}
+function sanitizeFilename(id) {
+	return id.replace(/^https?:\/\//, "").replace(/[#?]/g, "-").replace(/[^a-z0-9.-]/gi, "-").replace(/-+/g, "-").replace(/^-|-$/g, "").slice(0, 100);
+}
+function extractSection(text) {
+	const headings = [];
+	for (const line of text.split("\n")) {
+		const match = line.match(/^(#{1,6}) ([^\n]+)$/);
+		if (match) {
+			const level = match[1].length;
+			const heading = match[2].trim();
+			headings.length = level - 1;
+			headings[level - 1] = heading;
+		}
+	}
+	return headings.filter(Boolean).join(" > ") || void 0;
+}
+function formatReferenceFile(docId, doc, section, chunk, totalChunks) {
+	const frontmatter = [
+		"---",
+		`id: "${docId}"`,
+		`source: "${doc.url}"`,
+		`title: "${doc.title}"`
+	];
+	if (section) frontmatter.push(`section: "${section}"`);
+	if (totalChunks > 1) frontmatter.push(`chunk: ${chunk.index + 1}/${totalChunks}`);
+	frontmatter.push("---", "");
+	const prefix = [doc.title, section].filter(Boolean).join(" > ");
+	return frontmatter.join("\n") + (prefix ? `${prefix}\n\n` : "") + chunk.text;
+}
+async function fetchFromLlmsTxt(baseUrl, maxPages, onProgress) {
+	const llmsUrl = `${new URL(baseUrl).origin}/llms.txt`;
+	const res = await fetch(llmsUrl, { headers: { "User-Agent": "skilld/1.0" } }).catch(() => null);
+	if (!res?.ok) return null;
+	const llmsContent = await res.text();
+	if (llmsContent.length < 50) return null;
+	const links = parseLinks(llmsContent);
+	const docs = [];
+	let count = 0;
+	for (const { title, url } of links.slice(0, maxPages)) {
+		count++;
+		onProgress?.({
+			url,
+			count,
+			phase: "fetch"
+		});
+		const content = await fetchMarkdown(url);
+		if (content && content.length >= 50) docs.push({
+			url,
+			title,
+			content
+		});
+	}
+	return {
+		docs,
+		llmsContent
+	};
+}
+function parseLinks(content) {
+	const links = [];
+	const linkRegex = /\[([^\]]+)\]\(([^)]+)\)/g;
+	let match;
+	while ((match = linkRegex.exec(content)) !== null) {
+		const [, title, url] = match;
+		if (url.includes("/raw/") || url.endsWith(".md")) links.push({
+			title,
+			url
+		});
+	}
+	return links;
+}
+async function fetchMarkdown(url) {
+	const res = await fetch(url, { headers: { "User-Agent": "skilld/1.0" } }).catch(() => null);
+	if (!res?.ok) return null;
+	return res.text();
+}
+async function crawlSite(url, maxPages, onProgress) {
+	const { htmlToMarkdown } = await import("mdream");
+	const { crawlAndGenerate } = await import("@mdream/crawl");
+	const { tmpdir } = await import("node:os");
+	const { join } = await import("node:path");
+	const docs = [];
+	let count = 0;
+	const outputDir = join(tmpdir(), `skilld-crawl-${Date.now()}`);
+	await crawlAndGenerate({
+		urls: [url],
+		outputDir,
+		maxRequestsPerCrawl: maxPages,
+		followLinks: true,
+		onPage: async ({ url: pageUrl, html, title }) => {
+			count++;
+			onProgress?.({
+				url: pageUrl,
+				count,
+				phase: "fetch"
+			});
+			const markdown = htmlToMarkdown(html, { origin: new URL(pageUrl).origin });
+			if (markdown && markdown.length >= 50) docs.push({
+				url: pageUrl,
+				title: title || pageUrl,
+				content: markdown
+			});
+		}
+	});
+	return docs;
+}
+export { generateSkill };
+
+//# sourceMappingURL=index.mjs.map

package/dist/index.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.mjs","names":["join"],"sources":["../src/index.ts"],"sourcesContent":["import type { FetchedDoc, SkillConfig, SkillResult } from './types'\nimport { existsSync, mkdirSync, writeFileSync } from 'node:fs'\nimport { join } from 'node:path'\n\nexport type { FetchedDoc, SkillConfig, SkillResult }\n\n/**\n * Generate a skill from a documentation site\n */\nexport async function generateSkill(\n  config: SkillConfig,\n  onProgress?: (info: { url: string, count: number, phase: 'fetch' | 'index' }) => void,\n): Promise<SkillResult> {\n  const {\n    url,\n    outputDir = '.skilld',\n    chunkSize = 1000,\n    chunkOverlap = 200,\n    maxPages = 100,\n    skipLlmsTxt = false,\n    model = 'Xenova/bge-small-en-v1.5',\n  } = config\n\n  const siteName = getSiteName(url)\n  const skillDir = join(outputDir, siteName)\n  const referencesDir = join(skillDir, 'references')\n  const dbPath = join(skillDir, 'search.db')\n\n  mkdirSync(referencesDir, { recursive: true })\n\n  // Fetch docs\n  let docs: FetchedDoc[]\n  let skillContent: string | undefined\n\n  if (!skipLlmsTxt) {\n    const llmsResult = await fetchFromLlmsTxt(url, maxPages, onProgress)\n    if (llmsResult) {\n      docs = llmsResult.docs\n      skillContent = llmsResult.llmsContent\n    }\n    else {\n      docs = await crawlSite(url, maxPages, onProgress)\n    }\n  }\n  else {\n    docs = await crawlSite(url, maxPages, onProgress)\n  }\n\n  if (docs.length === 0) {\n    throw new Error('No documents found to index')\n  }\n\n  // Write SKILL.md\n  const skillPath = join(skillDir, 'SKILL.md')\n  if (skillContent) {\n    writeFileSync(skillPath, skillContent)\n  }\n\n  // Chunk and index docs\n  const { splitText } = await import('./split-text')\n  const { sqliteVec } = await import('retriv/db/sqlite-vec')\n  const { transformers } = await import('retriv/embeddings/transformers')\n\n  const documents: Array<{ id: string, content: string, metadata: Record<string, any> }> = []\n\n  for (const doc of docs) {\n    const chunks = splitText(doc.content, { chunkSize, chunkOverlap })\n\n    for (const chunk of chunks) {\n      const section = extractSection(chunk.text)\n      const docId = chunks.length > 1\n        ? `${doc.url}#chunk-${chunk.index}`\n        : doc.url\n\n      // Prepend title/section for better semantic matching\n      const prefix = [doc.title, section].filter(Boolean).join(' > ')\n      const content = prefix ? `${prefix}\\n\\n${chunk.text}` : chunk.text\n\n      documents.push({\n        id: docId,\n        content,\n        metadata: {\n          source: doc.url,\n          title: doc.title,\n          ...(section && { section }),\n          ...(chunks.length > 1 && {\n            chunkIndex: chunk.index,\n            chunkTotal: chunks.length,\n          }),\n        },\n      })\n\n      // Write reference file\n      const filename = sanitizeFilename(docId) + '.md'\n      const refPath = join(referencesDir, filename)\n      writeFileSync(refPath, formatReferenceFile(docId, doc, section, chunk, chunks.length))\n    }\n  }\n\n  // Index with retriv\n  onProgress?.({ url: 'embedding', count: documents.length, phase: 'index' })\n\n  const db = await sqliteVec({\n    path: dbPath,\n    embeddings: transformers({ model }),\n  })\n\n  await db.index(documents)\n  await db.close?.()\n\n  return {\n    siteName,\n    skillPath,\n    referencesDir,\n    dbPath,\n    chunkCount: documents.length,\n  }\n}\n\nfunction getSiteName(url: string): string {\n  const urlObj = new URL(url)\n  return urlObj.hostname.replace(/^www\\./, '')\n}\n\nfunction sanitizeFilename(id: string): string {\n  return id\n    .replace(/^https?:\\/\\//, '')\n    .replace(/[#?]/g, '-')\n    .replace(/[^a-z0-9.-]/gi, '-')\n    .replace(/-+/g, '-')\n    .replace(/^-|-$/g, '')\n    .slice(0, 100)\n}\n\nfunction extractSection(text: string): string | undefined {\n  const headings: string[] = []\n  for (const line of text.split('\\n')) {\n    const match = line.match(/^(#{1,6}) ([^\\n]+)$/)\n    if (match) {\n      const level = match[1]!.length\n      const heading = match[2]!.trim()\n      headings.length = level - 1\n      headings[level - 1] = heading\n    }\n  }\n  const section = headings.filter(Boolean).join(' > ')\n  return section || undefined\n}\n\nfunction formatReferenceFile(\n  docId: string,\n  doc: FetchedDoc,\n  section: string | undefined,\n  chunk: { text: string, index: number },\n  totalChunks: number,\n): string {\n  const frontmatter = [\n    '---',\n    `id: \"${docId}\"`,\n    `source: \"${doc.url}\"`,\n    `title: \"${doc.title}\"`,\n  ]\n  if (section)\n    frontmatter.push(`section: \"${section}\"`)\n  if (totalChunks > 1)\n    frontmatter.push(`chunk: ${chunk.index + 1}/${totalChunks}`)\n  frontmatter.push('---', '')\n\n  const prefix = [doc.title, section].filter(Boolean).join(' > ')\n  return frontmatter.join('\\n') + (prefix ? `${prefix}\\n\\n` : '') + chunk.text\n}\n\nasync function fetchFromLlmsTxt(\n  baseUrl: string,\n  maxPages: number,\n  onProgress?: (info: { url: string, count: number, phase: 'fetch' | 'index' }) => void,\n): Promise<{ docs: FetchedDoc[], llmsContent: string } | null> {\n  const urlObj = new URL(baseUrl)\n  const llmsUrl = `${urlObj.origin}/llms.txt`\n\n  const res = await fetch(llmsUrl, {\n    headers: { 'User-Agent': 'skilld/1.0' },\n  }).catch(() => null)\n\n  if (!res?.ok)\n    return null\n\n  const llmsContent = await res.text()\n  if (llmsContent.length < 50)\n    return null\n\n  // Parse markdown links\n  const links = parseLinks(llmsContent)\n  const docs: FetchedDoc[] = []\n\n  let count = 0\n  for (const { title, url } of links.slice(0, maxPages)) {\n    count++\n    onProgress?.({ url, count, phase: 'fetch' })\n\n    const content = await fetchMarkdown(url)\n    if (content && content.length >= 50) {\n      docs.push({ url, title, content })\n    }\n  }\n\n  return { docs, llmsContent }\n}\n\nfunction parseLinks(content: string): Array<{ title: string, url: string }> {\n  const links: Array<{ title: string, url: string }> = []\n  const linkRegex = /\\[([^\\]]+)\\]\\(([^)]+)\\)/g\n  let match\n\n  while ((match = linkRegex.exec(content)) !== null) {\n    const [, title, url] = match\n    if (url.includes('/raw/') || url.endsWith('.md')) {\n      links.push({ title, url })\n    }\n  }\n\n  return links\n}\n\nasync function fetchMarkdown(url: string): Promise<string | null> {\n  const res = await fetch(url, {\n    headers: { 'User-Agent': 'skilld/1.0' },\n  }).catch(() => null)\n\n  if (!res?.ok)\n    return null\n\n  return res.text()\n}\n\nasync function crawlSite(\n  url: string,\n  maxPages: number,\n  onProgress?: (info: { url: string, count: number, phase: 'fetch' | 'index' }) => void,\n): Promise<FetchedDoc[]> {\n  const { htmlToMarkdown } = await import('mdream')\n  const { crawlAndGenerate } = await import('@mdream/crawl')\n  const { tmpdir } = await import('node:os')\n  const { join } = await import('node:path')\n\n  const docs: FetchedDoc[] = []\n  let count = 0\n  const outputDir = join(tmpdir(), `skilld-crawl-${Date.now()}`)\n\n  await crawlAndGenerate({\n    urls: [url],\n    outputDir,\n    maxRequestsPerCrawl: maxPages,\n    followLinks: true,\n    onPage: async ({ url: pageUrl, html, title }) => {\n      count++\n      onProgress?.({ url: pageUrl, count, phase: 'fetch' })\n\n      const urlObj = new URL(pageUrl)\n      const markdown = htmlToMarkdown(html, { origin: urlObj.origin })\n\n      if (markdown && markdown.length >= 50) {\n        docs.push({ url: pageUrl, title: title || pageUrl, content: markdown })\n      }\n    },\n  })\n\n  return docs\n}\n"],"mappings":";;AASA,eAAsB,cACpB,QACA,YACsB;CACtB,MAAM,EACJ,KACA,YAAY,WACZ,YAAY,KACZ,eAAe,KACf,WAAW,KACX,cAAc,OACd,QAAQ,+BACN;CAEJ,MAAM,WAAW,YAAY,IAAI;CACjC,MAAM,WAAWA,OAAK,WAAW,SAAS;CAC1C,MAAM,gBAAgBA,OAAK,UAAU,aAAa;CAClD,MAAM,SAASA,OAAK,UAAU,YAAY;AAE1C,WAAU,eAAe,EAAE,WAAW,MAAM,CAAC;CAG7C,IAAI;CACJ,IAAI;AAEJ,KAAI,CAAC,aAAa;EAChB,MAAM,aAAa,MAAM,iBAAiB,KAAK,UAAU,WAAW;AACpE,MAAI,YAAY;AACd,UAAO,WAAW;AAClB,kBAAe,WAAW;QAG1B,QAAO,MAAM,UAAU,KAAK,UAAU,WAAW;OAInD,QAAO,MAAM,UAAU,KAAK,UAAU,WAAW;AAGnD,KAAI,KAAK,WAAW,EAClB,OAAM,IAAI,MAAM,8BAA8B;CAIhD,MAAM,YAAYA,OAAK,UAAU,WAAW;AAC5C,KAAI,aACF,eAAc,WAAW,aAAa;CAIxC,MAAM,EAAE,cAAc,MAAM,OAAO;CACnC,MAAM,EAAE,cAAc,MAAM,OAAO;CACnC,MAAM,EAAE,iBAAiB,MAAM,OAAO;CAEtC,MAAM,YAAmF,EAAE;AAE3F,MAAK,MAAM,OAAO,MAAM;EACtB,MAAM,SAAS,UAAU,IAAI,SAAS;GAAE;GAAW;GAAc,CAAC;AAElE,OAAK,MAAM,SAAS,QAAQ;GAC1B,MAAM,UAAU,eAAe,MAAM,KAAK;GAC1C,MAAM,QAAQ,OAAO,SAAS,IAC1B,GAAG,IAAI,IAAI,SAAS,MAAM,UAC1B,IAAI;GAGR,MAAM,SAAS,CAAC,IAAI,OAAO,QAAQ,CAAC,OAAO,QAAQ,CAAC,KAAK,MAAM;GAC/D,MAAM,UAAU,SAAS,GAAG,OAAO,MAAM,MAAM,SAAS,MAAM;AAE9D,aAAU,KAAK;IACb,IAAI;IACJ;IACA,UAAU;KACR,QAAQ,IAAI;KACZ,OAAO,IAAI;KACX,GAAI,WAAW,EAAE,SAAS;KAC1B,GAAI,OAAO,SAAS,KAAK;MACvB,YAAY,MAAM;MAClB,YAAY,OAAO;MACpB;KACF;IACF,CAAC;AAKF,iBADgBA,OAAK,eADJ,iBAAiB,MAAM,GAAG,MACE,EACtB,oBAAoB,OAAO,KAAK,SAAS,OAAO,OAAO,OAAO,CAAC;;;AAK1F,cAAa;EAAE,KAAK;EAAa,OAAO,UAAU;EAAQ,OAAO;EAAS,CAAC;CAE3E,MAAM,KAAK,MAAM,UAAU;EACzB,MAAM;EACN,YAAY,aAAa,EAAE,OAAO,CAAC;EACpC,CAAC;AAEF,OAAM,GAAG,MAAM,UAAU;AACzB,OAAM,GAAG,SAAS;AAElB,QAAO;EACL;EACA;EACA;EACA;EACA,YAAY,UAAU;EACvB;;AAGH,SAAS,YAAY,KAAqB;AAExC,QADe,IAAI,IAAI,IAAI,CACb,SAAS,QAAQ,UAAU,GAAG;;AAG9C,SAAS,iBAAiB,IAAoB;AAC5C,QAAO,GACJ,QAAQ,gBAAgB,GAAG,CAC3B,QAAQ,SAAS,IAAI,CACrB,QAAQ,iBAAiB,IAAI,CAC7B,QAAQ,OAAO,IAAI,CACnB,QAAQ,UAAU,GAAG,CACrB,MAAM,GAAG,IAAI;;AAGlB,SAAS,eAAe,MAAkC;CACxD,MAAM,WAAqB,EAAE;AAC7B,MAAK,MAAM,QAAQ,KAAK,MAAM,KAAK,EAAE;EACnC,MAAM,QAAQ,KAAK,MAAM,sBAAsB;AAC/C,MAAI,OAAO;GACT,MAAM,QAAQ,MAAM,GAAI;GACxB,MAAM,UAAU,MAAM,GAAI,MAAM;AAChC,YAAS,SAAS,QAAQ;AAC1B,YAAS,QAAQ,KAAK;;;AAI1B,QADgB,SAAS,OAAO,QAAQ,CAAC,KAAK,MAAM,IAClC,KAAA;;AAGpB,SAAS,oBACP,OACA,KACA,SACA,OACA,aACQ;CACR,MAAM,cAAc;EAClB;EACA,QAAQ,MAAM;EACd,YAAY,IAAI,IAAI;EACpB,WAAW,IAAI,MAAM;EACtB;AACD,KAAI,QACF,aAAY,KAAK,aAAa,QAAQ,GAAG;AAC3C,KAAI,cAAc,EAChB,aAAY,KAAK,UAAU,MAAM,QAAQ,EAAE,GAAG,cAAc;AAC9D,aAAY,KAAK,OAAO,GAAG;CAE3B,MAAM,SAAS,CAAC,IAAI,OAAO,QAAQ,CAAC,OAAO,QAAQ,CAAC,KAAK,MAAM;AAC/D,QAAO,YAAY,KAAK,KAAK,IAAI,SAAS,GAAG,OAAO,QAAQ,MAAM,MAAM;;AAG1E,eAAe,iBACb,SACA,UACA,YAC6D;CAE7D,MAAM,UAAU,GADD,IAAI,IAAI,QAAQ,CACL,OAAO;CAEjC,MAAM,MAAM,MAAM,MAAM,SAAS,EAC/B,SAAS,EAAE,cAAc,cAAc,EACxC,CAAC,CAAC,YAAY,KAAK;AAEpB,KAAI,CAAC,KAAK,GACR,QAAO;CAET,MAAM,cAAc,MAAM,IAAI,MAAM;AACpC,KAAI,YAAY,SAAS,GACvB,QAAO;CAGT,MAAM,QAAQ,WAAW,YAAY;CACrC,MAAM,OAAqB,EAAE;CAE7B,IAAI,QAAQ;AACZ,MAAK,MAAM,EAAE,OAAO,SAAS,MAAM,MAAM,GAAG,SAAS,EAAE;AACrD;AACA,eAAa;GAAE;GAAK;GAAO,OAAO;GAAS,CAAC;EAE5C,MAAM,UAAU,MAAM,cAAc,IAAI;AACxC,MAAI,WAAW,QAAQ,UAAU,GAC/B,MAAK,KAAK;GAAE;GAAK;GAAO;GAAS,CAAC;;AAItC,QAAO;EAAE;EAAM;EAAa;;AAG9B,SAAS,WAAW,SAAwD;CAC1E,MAAM,QAA+C,EAAE;CACvD,MAAM,YAAY;CAClB,IAAI;AAEJ,SAAQ,QAAQ,UAAU,KAAK,QAAQ,MAAM,MAAM;EACjD,MAAM,GAAG,OAAO,OAAO;AACvB,MAAI,IAAI,SAAS,QAAQ,IAAI,IAAI,SAAS,MAAM,CAC9C,OAAM,KAAK;GAAE;GAAO;GAAK,CAAC;;AAI9B,QAAO;;AAGT,eAAe,cAAc,KAAqC;CAChE,MAAM,MAAM,MAAM,MAAM,KAAK,EAC3B,SAAS,EAAE,cAAc,cAAc,EACxC,CAAC,CAAC,YAAY,KAAK;AAEpB,KAAI,CAAC,KAAK,GACR,QAAO;AAET,QAAO,IAAI,MAAM;;AAGnB,eAAe,UACb,KACA,UACA,YACuB;CACvB,MAAM,EAAE,mBAAmB,MAAM,OAAO;CACxC,MAAM,EAAE,qBAAqB,MAAM,OAAO;CAC1C,MAAM,EAAE,WAAW,MAAM,OAAO;CAChC,MAAM,EAAE,SAAS,MAAM,OAAO;CAE9B,MAAM,OAAqB,EAAE;CAC7B,IAAI,QAAQ;CACZ,MAAM,YAAY,KAAK,QAAQ,EAAE,gBAAgB,KAAK,KAAK,GAAG;AAE9D,OAAM,iBAAiB;EACrB,MAAM,CAAC,IAAI;EACX;EACA,qBAAqB;EACrB,aAAa;EACb,QAAQ,OAAO,EAAE,KAAK,SAAS,MAAM,YAAY;AAC/C;AACA,gBAAa;IAAE,KAAK;IAAS;IAAO,OAAO;IAAS,CAAC;GAGrD,MAAM,WAAW,eAAe,MAAM,EAAE,QADzB,IAAI,IAAI,QAAQ,CACwB,QAAQ,CAAC;AAEhE,OAAI,YAAY,SAAS,UAAU,GACjC,MAAK,KAAK;IAAE,KAAK;IAAS,OAAO,SAAS;IAAS,SAAS;IAAU,CAAC;;EAG5E,CAAC;AAEF,QAAO"}

package/dist/npm.d.mts

@@ -0,0 +1,48 @@
+//#region src/npm.d.ts
+/**
+ * NPM package discovery and documentation resolution
+ */
+interface NpmPackageInfo {
+  name: string;
+  version?: string;
+  description?: string;
+  homepage?: string;
+  repository?: {
+    type: string;
+    url: string;
+    directory?: string;
+  };
+  readme?: string;
+}
+interface ResolvedPackage {
+  name: string;
+  version?: string;
+  description?: string;
+  docsUrl?: string;
+  llmsUrl?: string;
+  readmeUrl?: string;
+  repoUrl?: string;
+}
+/**
+ * Fetch package info from npm registry
+ */
+declare function fetchNpmPackage(packageName: string): Promise<NpmPackageInfo | null>;
+/**
+ * Resolve documentation URL for a package
+ */
+declare function resolvePackageDocs(packageName: string): Promise<ResolvedPackage | null>;
+interface LocalDependency {
+  name: string;
+  version: string;
+}
+/**
+ * Read package.json dependencies with versions
+ */
+declare function readLocalDependencies(cwd: string): Promise<LocalDependency[]>;
+/**
+ * Get installed skill version from SKILL.md
+ */
+declare function getInstalledSkillVersion(skillDir: string): Promise<string | null>;
+//#endregion
+export { LocalDependency, NpmPackageInfo, ResolvedPackage, fetchNpmPackage, getInstalledSkillVersion, readLocalDependencies, resolvePackageDocs };
+//# sourceMappingURL=npm.d.mts.map

package/dist/npm.d.mts.map

@@ -0,0 +1 @@
+{"version":3,"file":"npm.d.mts","names":[],"sources":["../src/npm.ts"],"mappings":";;AAIA;;UAAiB,cAAA;EACf,IAAA;EACA,OAAA;EACA,WAAA;EACA,QAAA;EACA,UAAA;IACE,IAAA;IACA,GAAA;IACA,SAAA;EAAA;EAEF,MAAA;AAAA;AAAA,UAGe,eAAA;EACf,IAAA;EACA,OAAA;EACA,WAAA;EACA,OAAA;EACA,OAAA;EACA,SAAA;EACA,OAAA;AAAA;;;;iBAMoB,eAAA,CAAgB,WAAA,WAAsB,OAAA,CAAQ,cAAA;;AAApE;;iBAYsB,kBAAA,CAAmB,WAAA,WAAsB,OAAA,CAAQ,eAAA;AAAA,UA0EtD,eAAA;EACf,IAAA;EACA,OAAA;AAAA;;;AA5EF;iBAkFsB,qBAAA,CAAsB,GAAA,WAAc,OAAA,CAAQ,eAAA;;;;iBA+B5C,wBAAA,CACpB,QAAA,WACC,OAAA"}

package/dist/npm.mjs

@@ -0,0 +1,90 @@
+async function fetchNpmPackage(packageName) {
+	const res = await fetch(`https://registry.npmjs.org/${packageName}/latest`, { headers: { "User-Agent": "skilld/1.0" } }).catch(() => null);
+	if (!res?.ok) return null;
+	return res.json();
+}
+async function resolvePackageDocs(packageName) {
+	const pkg = await fetchNpmPackage(packageName);
+	if (!pkg) return null;
+	const result = {
+		name: pkg.name,
+		version: pkg.version,
+		description: pkg.description
+	};
+	if (pkg.repository?.url) result.repoUrl = pkg.repository.url.replace(/^git\+/, "").replace(/\.git$/, "").replace(/^git:\/\//, "https://").replace(/^ssh:\/\/git@github\.com/, "https://github.com");
+	if (pkg.homepage && !isGitHubRepoUrl(pkg.homepage)) {
+		result.docsUrl = pkg.homepage;
+		const llmsUrl = `${pkg.homepage.replace(/\/$/, "")}/llms.txt`;
+		if (await verifyUrl(llmsUrl)) result.llmsUrl = llmsUrl;
+	}
+	if (result.repoUrl?.includes("github.com")) {
+		const match = result.repoUrl.match(/github\.com\/([^/]+)\/([^/]+)/);
+		if (match) {
+			const owner = match[1];
+			const repo = match[2];
+			const subdir = pkg.repository?.directory;
+			const unghUrl = subdir ? `https://ungh.cc/repos/${owner}/${repo}/files/main/${subdir}/README.md` : `https://ungh.cc/repos/${owner}/${repo}/readme`;
+			if ((await fetch(unghUrl, { headers: { "User-Agent": "skilld/1.0" } }).catch(() => null))?.ok) result.readmeUrl = `ungh://${owner}/${repo}${subdir ? `/${subdir}` : ""}`;
+			else {
+				const basePath = subdir ? `${subdir}/` : "";
+				for (const branch of ["main", "master"]) {
+					const readmeUrl = `https://raw.githubusercontent.com/${owner}/${repo}/${branch}/${basePath}README.md`;
+					if (await verifyUrl(readmeUrl)) {
+						result.readmeUrl = readmeUrl;
+						break;
+					}
+				}
+			}
+		}
+	}
+	if (!result.docsUrl && !result.llmsUrl && !result.readmeUrl) return null;
+	return result;
+}
+async function readLocalDependencies(cwd) {
+	const { readFileSync, existsSync } = await import("node:fs");
+	const { join } = await import("node:path");
+	const pkgPath = join(cwd, "package.json");
+	if (!existsSync(pkgPath)) throw new Error("No package.json found in current directory");
+	const pkg = JSON.parse(readFileSync(pkgPath, "utf-8"));
+	const deps = {
+		...pkg.dependencies,
+		...pkg.devDependencies
+	};
+	return Object.entries(deps).filter(([name]) => !name.startsWith("@types/") && ![
+		"typescript",
+		"eslint",
+		"prettier",
+		"vitest",
+		"jest"
+	].includes(name)).map(([name, version]) => ({
+		name,
+		version: version.replace(/^[\^~>=<]/, "")
+	}));
+}
+async function getInstalledSkillVersion(skillDir) {
+	const { readFileSync, existsSync } = await import("node:fs");
+	const { join } = await import("node:path");
+	const skillPath = join(skillDir, "SKILL.md");
+	if (!existsSync(skillPath)) return null;
+	return readFileSync(skillPath, "utf-8").match(/^version:\s*"?([^"\n]+)"?/m)?.[1] || null;
+}
+function isGitHubRepoUrl(url) {
+	try {
+		const parsed = new URL(url);
+		if (parsed.hostname === "github.com" || parsed.hostname === "www.github.com") return parsed.pathname.split("/").filter(Boolean).length <= 2;
+		return false;
+	} catch {
+		return false;
+	}
+}
+async function verifyUrl(url) {
+	const res = await fetch(url, {
+		method: "HEAD",
+		headers: { "User-Agent": "skilld/1.0" }
+	}).catch(() => null);
+	if (!res?.ok) return false;
+	return !(res.headers.get("content-type") || "").includes("text/html");
+}
+export { fetchNpmPackage, getInstalledSkillVersion, readLocalDependencies, resolvePackageDocs };
+
+//# sourceMappingURL=npm.mjs.map

package/dist/npm.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"npm.mjs","names":[],"sources":["../src/npm.ts"],"sourcesContent":["/**\n * NPM package discovery and documentation resolution\n */\n\nexport interface NpmPackageInfo {\n  name: string\n  version?: string\n  description?: string\n  homepage?: string\n  repository?: {\n    type: string\n    url: string\n    directory?: string\n  }\n  readme?: string\n}\n\nexport interface ResolvedPackage {\n  name: string\n  version?: string\n  description?: string\n  docsUrl?: string\n  llmsUrl?: string\n  readmeUrl?: string\n  repoUrl?: string\n}\n\n/**\n * Fetch package info from npm registry\n */\nexport async function fetchNpmPackage(packageName: string): Promise<NpmPackageInfo | null> {\n  const res = await fetch(`https://registry.npmjs.org/${packageName}/latest`, {\n    headers: { 'User-Agent': 'skilld/1.0' },\n  }).catch(() => null)\n\n  if (!res?.ok) return null\n  return res.json()\n}\n\n/**\n * Resolve documentation URL for a package\n */\nexport async function resolvePackageDocs(packageName: string): Promise<ResolvedPackage | null> {\n  const pkg = await fetchNpmPackage(packageName)\n  if (!pkg) return null\n\n  const result: ResolvedPackage = {\n    name: pkg.name,\n    version: pkg.version,\n    description: pkg.description,\n  }\n\n  // Extract repo URL\n  if (pkg.repository?.url) {\n    const repoUrl = pkg.repository.url\n      .replace(/^git\\+/, '')\n      .replace(/\\.git$/, '')\n      .replace(/^git:\\/\\//, 'https://')\n      .replace(/^ssh:\\/\\/git@github\\.com/, 'https://github.com')\n    result.repoUrl = repoUrl\n  }\n\n  // Try homepage for docs (skip if it's just a GitHub repo URL)\n  if (pkg.homepage && !isGitHubRepoUrl(pkg.homepage)) {\n    result.docsUrl = pkg.homepage\n\n    // Check for llms.txt\n    const llmsUrl = `${pkg.homepage.replace(/\\/$/, '')}/llms.txt`\n    if (await verifyUrl(llmsUrl)) {\n      result.llmsUrl = llmsUrl\n    }\n  }\n\n  // GitHub README fallback via ungh (unjs GitHub API proxy)\n  if (result.repoUrl?.includes('github.com')) {\n    const match = result.repoUrl.match(/github\\.com\\/([^/]+)\\/([^/]+)/)\n    if (match) {\n      const owner = match[1]\n      const repo = match[2]\n      const subdir = pkg.repository?.directory\n\n      // Use ungh for README (auto-detects branch, returns JSON with markdown field)\n      const unghUrl = subdir\n        ? `https://ungh.cc/repos/${owner}/${repo}/files/main/${subdir}/README.md`\n        : `https://ungh.cc/repos/${owner}/${repo}/readme`\n\n      // Store as ungh:// URL for special handling in CLI\n      const unghRes = await fetch(unghUrl, {\n        headers: { 'User-Agent': 'skilld/1.0' },\n      }).catch(() => null)\n\n      if (unghRes?.ok) {\n        result.readmeUrl = `ungh://${owner}/${repo}${subdir ? `/${subdir}` : ''}`\n      }\n      else {\n        // Fallback to raw.githubusercontent.com\n        const basePath = subdir ? `${subdir}/` : ''\n        for (const branch of ['main', 'master']) {\n          const readmeUrl = `https://raw.githubusercontent.com/${owner}/${repo}/${branch}/${basePath}README.md`\n          if (await verifyUrl(readmeUrl)) {\n            result.readmeUrl = readmeUrl\n            break\n          }\n        }\n      }\n    }\n  }\n\n  // Must have at least one source\n  if (!result.docsUrl && !result.llmsUrl && !result.readmeUrl) {\n    return null\n  }\n\n  return result\n}\n\nexport interface LocalDependency {\n  name: string\n  version: string\n}\n\n/**\n * Read package.json dependencies with versions\n */\nexport async function readLocalDependencies(cwd: string): Promise<LocalDependency[]> {\n  const { readFileSync, existsSync } = await import('node:fs')\n  const { join } = await import('node:path')\n\n  const pkgPath = join(cwd, 'package.json')\n  if (!existsSync(pkgPath)) {\n    throw new Error('No package.json found in current directory')\n  }\n\n  const pkg = JSON.parse(readFileSync(pkgPath, 'utf-8'))\n  const deps: Record<string, string> = {\n    ...pkg.dependencies,\n    ...pkg.devDependencies,\n  }\n\n  return Object.entries(deps)\n    .filter(([name]) =>\n      // Skip common non-doc packages\n      !name.startsWith('@types/')\n      && !['typescript', 'eslint', 'prettier', 'vitest', 'jest'].includes(name),\n    )\n    .map(([name, version]) => ({\n      name,\n      // Clean version string (remove ^, ~, etc.)\n      version: version.replace(/^[\\^~>=<]/, ''),\n    }))\n}\n\n/**\n * Get installed skill version from SKILL.md\n */\nexport async function getInstalledSkillVersion(\n  skillDir: string,\n): Promise<string | null> {\n  const { readFileSync, existsSync } = await import('node:fs')\n  const { join } = await import('node:path')\n\n  const skillPath = join(skillDir, 'SKILL.md')\n  if (!existsSync(skillPath)) return null\n\n  const content = readFileSync(skillPath, 'utf-8')\n  const match = content.match(/^version:\\s*\"?([^\"\\n]+)\"?/m)\n  return match?.[1] || null\n}\n\n/**\n * Check if URL is a GitHub repository page (not a docs site)\n */\nfunction isGitHubRepoUrl(url: string): boolean {\n  try {\n    const parsed = new URL(url)\n    // GitHub repo URLs: github.com/owner/repo or github.com/owner/repo#readme\n    if (parsed.hostname === 'github.com' || parsed.hostname === 'www.github.com') {\n      const parts = parsed.pathname.split('/').filter(Boolean)\n      // owner/repo pattern without additional paths like /docs\n      return parts.length <= 2\n    }\n    return false\n  }\n  catch {\n    return false\n  }\n}\n\nasync function verifyUrl(url: string): Promise<boolean> {\n  const res = await fetch(url, {\n    method: 'HEAD',\n    headers: { 'User-Agent': 'skilld/1.0' },\n  }).catch(() => null)\n\n  if (!res?.ok) return false\n\n  const contentType = res.headers.get('content-type') || ''\n  // Reject HTML (likely 404 page)\n  return !contentType.includes('text/html')\n}\n"],"mappings":"AA8BA,eAAsB,gBAAgB,aAAqD;CACzF,MAAM,MAAM,MAAM,MAAM,8BAA8B,YAAY,UAAU,EAC1E,SAAS,EAAE,cAAc,cAAc,EACxC,CAAC,CAAC,YAAY,KAAK;AAEpB,KAAI,CAAC,KAAK,GAAI,QAAO;AACrB,QAAO,IAAI,MAAM;;AAMnB,eAAsB,mBAAmB,aAAsD;CAC7F,MAAM,MAAM,MAAM,gBAAgB,YAAY;AAC9C,KAAI,CAAC,IAAK,QAAO;CAEjB,MAAM,SAA0B;EAC9B,MAAM,IAAI;EACV,SAAS,IAAI;EACb,aAAa,IAAI;EAClB;AAGD,KAAI,IAAI,YAAY,IAMlB,QAAO,UALS,IAAI,WAAW,IAC5B,QAAQ,UAAU,GAAG,CACrB,QAAQ,UAAU,GAAG,CACrB,QAAQ,aAAa,WAAW,CAChC,QAAQ,4BAA4B,qBAAqB;AAK9D,KAAI,IAAI,YAAY,CAAC,gBAAgB,IAAI,SAAS,EAAE;AAClD,SAAO,UAAU,IAAI;EAGrB,MAAM,UAAU,GAAG,IAAI,SAAS,QAAQ,OAAO,GAAG,CAAC;AACnD,MAAI,MAAM,UAAU,QAAQ,CAC1B,QAAO,UAAU;;AAKrB,KAAI,OAAO,SAAS,SAAS,aAAa,EAAE;EAC1C,MAAM,QAAQ,OAAO,QAAQ,MAAM,gCAAgC;AACnE,MAAI,OAAO;GACT,MAAM,QAAQ,MAAM;GACpB,MAAM,OAAO,MAAM;GACnB,MAAM,SAAS,IAAI,YAAY;GAG/B,MAAM,UAAU,SACZ,yBAAyB,MAAM,GAAG,KAAK,cAAc,OAAO,cAC5D,yBAAyB,MAAM,GAAG,KAAK;AAO3C,QAJgB,MAAM,MAAM,SAAS,EACnC,SAAS,EAAE,cAAc,cAAc,EACxC,CAAC,CAAC,YAAY,KAAK,GAEP,GACX,QAAO,YAAY,UAAU,MAAM,GAAG,OAAO,SAAS,IAAI,WAAW;QAElE;IAEH,MAAM,WAAW,SAAS,GAAG,OAAO,KAAK;AACzC,SAAK,MAAM,UAAU,CAAC,QAAQ,SAAS,EAAE;KACvC,MAAM,YAAY,qCAAqC,MAAM,GAAG,KAAK,GAAG,OAAO,GAAG,SAAS;AAC3F,SAAI,MAAM,UAAU,UAAU,EAAE;AAC9B,aAAO,YAAY;AACnB;;;;;;AAQV,KAAI,CAAC,OAAO,WAAW,CAAC,OAAO,WAAW,CAAC,OAAO,UAChD,QAAO;AAGT,QAAO;;AAWT,eAAsB,sBAAsB,KAAyC;CACnF,MAAM,EAAE,cAAc,eAAe,MAAM,OAAO;CAClD,MAAM,EAAE,SAAS,MAAM,OAAO;CAE9B,MAAM,UAAU,KAAK,KAAK,eAAe;AACzC,KAAI,CAAC,WAAW,QAAQ,CACtB,OAAM,IAAI,MAAM,6CAA6C;CAG/D,MAAM,MAAM,KAAK,MAAM,aAAa,SAAS,QAAQ,CAAC;CACtD,MAAM,OAA+B;EACnC,GAAG,IAAI;EACP,GAAG,IAAI;EACR;AAED,QAAO,OAAO,QAAQ,KAAK,CACxB,QAAQ,CAAC,UAER,CAAC,KAAK,WAAW,UAAU,IACxB,CAAC;EAAC;EAAc;EAAU;EAAY;EAAU;EAAO,CAAC,SAAS,KAAK,CAC1E,CACA,KAAK,CAAC,MAAM,cAAc;EACzB;EAEA,SAAS,QAAQ,QAAQ,aAAa,GAAG;EAC1C,EAAE;;AAMP,eAAsB,yBACpB,UACwB;CACxB,MAAM,EAAE,cAAc,eAAe,MAAM,OAAO;CAClD,MAAM,EAAE,SAAS,MAAM,OAAO;CAE9B,MAAM,YAAY,KAAK,UAAU,WAAW;AAC5C,KAAI,CAAC,WAAW,UAAU,CAAE,QAAO;AAInC,QAFgB,aAAa,WAAW,QAAQ,CAC1B,MAAM,6BAA6B,GAC1C,MAAM;;AAMvB,SAAS,gBAAgB,KAAsB;AAC7C,KAAI;EACF,MAAM,SAAS,IAAI,IAAI,IAAI;AAE3B,MAAI,OAAO,aAAa,gBAAgB,OAAO,aAAa,iBAG1D,QAFc,OAAO,SAAS,MAAM,IAAI,CAAC,OAAO,QAAQ,CAE3C,UAAU;AAEzB,SAAO;SAEH;AACJ,SAAO;;;AAIX,eAAe,UAAU,KAA+B;CACtD,MAAM,MAAM,MAAM,MAAM,KAAK;EAC3B,QAAQ;EACR,SAAS,EAAE,cAAc,cAAc;EACxC,CAAC,CAAC,YAAY,KAAK;AAEpB,KAAI,CAAC,KAAK,GAAI,QAAO;AAIrB,QAAO,EAFa,IAAI,QAAQ,IAAI,eAAe,IAAI,IAEnC,SAAS,YAAY"}

package/dist/split-text.d.mts

@@ -0,0 +1,24 @@
+//#region src/split-text.d.ts
+/**
+ * Recursive markdown text splitter (LangChain-style)
+ */
+interface SplitTextOptions {
+  chunkSize?: number;
+  chunkOverlap?: number;
+  separators?: string[];
+}
+interface TextChunk {
+  text: string;
+  index: number;
+  /** Character range [start, end] in original text */
+  range: [number, number];
+  /** Line range [startLine, endLine] (1-indexed) */
+  lines: [number, number];
+}
+/**
+ * Split text recursively using markdown-aware separators
+ */
+declare function splitText(text: string, options?: SplitTextOptions): TextChunk[];
+//#endregion
+export { SplitTextOptions, TextChunk, splitText };
+//# sourceMappingURL=split-text.d.mts.map

package/dist/split-text.d.mts.map

@@ -0,0 +1 @@
+{"version":3,"file":"split-text.d.mts","names":[],"sources":["../src/split-text.ts"],"mappings":";;AAoBA;;UAAiB,gBAAA;EACf,SAAA;EACA,YAAA;EACA,UAAA;AAAA;AAAA,UAGe,SAAA;EACf,IAAA;EACA,KAAA;EAFwB;EAIxB,KAAA;EAJwB;EAMxB,KAAA;AAAA;;;;iBAkBc,SAAA,CACd,IAAA,UACA,OAAA,GAAS,gBAAA,GACR,SAAA"}

package/dist/split-text.mjs

@@ -0,0 +1,87 @@
+const MARKDOWN_SEPARATORS = [
+	"\n## ",
+	"\n### ",
+	"\n#### ",
+	"\n##### ",
+	"\n###### ",
+	"```\n\n",
+	"\n\n***\n\n",
+	"\n\n---\n\n",
+	"\n\n___\n\n",
+	"\n\n",
+	"\n",
+	" ",
+	""
+];
+function offsetToLine(text, offset) {
+	let line = 1;
+	for (let i = 0; i < offset && i < text.length; i++) if (text[i] === "\n") line++;
+	return line;
+}
+function splitText(text, options = {}) {
+	const { chunkSize = 1e3, chunkOverlap = 200, separators = MARKDOWN_SEPARATORS } = options;
+	if (text.length <= chunkSize) {
+		const endLine = offsetToLine(text, text.length);
+		return [{
+			text,
+			index: 0,
+			range: [0, text.length],
+			lines: [1, endLine]
+		}];
+	}
+	return mergeChunks(splitRecursive(text, chunkSize, separators), chunkSize, chunkOverlap, text);
+}
+function splitRecursive(text, chunkSize, separators) {
+	if (text.length <= chunkSize || separators.length === 0) return [text];
+	const separator = separators.find((sep) => sep === "" || text.includes(sep));
+	if (!separator && separator !== "") return [text];
+	const parts = separator === "" ? [...text] : text.split(separator);
+	const results = [];
+	for (let i = 0; i < parts.length; i++) {
+		const part = parts[i];
+		const withSep = i < parts.length - 1 && separator !== "" ? part + separator : part;
+		if (withSep.length <= chunkSize) results.push(withSep);
+		else {
+			const subParts = splitRecursive(withSep, chunkSize, separators.slice(1));
+			results.push(...subParts);
+		}
+	}
+	return results;
+}
+function mergeChunks(parts, chunkSize, chunkOverlap, originalText) {
+	const chunks = [];
+	let current = "";
+	let currentStart = 0;
+	for (const part of parts) if (current.length + part.length <= chunkSize) current += part;
+	else {
+		if (current) {
+			const start = originalText.indexOf(current, currentStart);
+			const actualStart = start >= 0 ? start : currentStart;
+			const actualEnd = actualStart + current.length;
+			chunks.push({
+				text: current,
+				index: chunks.length,
+				range: [actualStart, actualEnd],
+				lines: [offsetToLine(originalText, actualStart), offsetToLine(originalText, actualEnd)]
+			});
+			currentStart = Math.max(0, actualStart + current.length - chunkOverlap);
+		}
+		if (chunkOverlap > 0 && current.length > chunkOverlap) current = current.slice(-chunkOverlap) + part;
+		else current = part;
+	}
+	if (current) {
+		const start = originalText.indexOf(current, currentStart);
+		const actualStart = start >= 0 ? start : currentStart;
+		const actualEnd = start >= 0 ? start + current.length : originalText.length;
+		chunks.push({
+			text: current,
+			index: chunks.length,
+			range: [actualStart, actualEnd],
+			lines: [offsetToLine(originalText, actualStart), offsetToLine(originalText, actualEnd)]
+		});
+	}
+	return chunks;
+}
+export { splitText };
+
+//# sourceMappingURL=split-text.mjs.map

package/dist/split-text.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"split-text.mjs","names":[],"sources":["../src/split-text.ts"],"sourcesContent":["/**\n * Recursive markdown text splitter (LangChain-style)\n */\n\nconst MARKDOWN_SEPARATORS = [\n  '\\n## ',\n  '\\n### ',\n  '\\n#### ',\n  '\\n##### ',\n  '\\n###### ',\n  '```\\n\\n',\n  '\\n\\n***\\n\\n',\n  '\\n\\n---\\n\\n',\n  '\\n\\n___\\n\\n',\n  '\\n\\n',\n  '\\n',\n  ' ',\n  '',\n]\n\nexport interface SplitTextOptions {\n  chunkSize?: number\n  chunkOverlap?: number\n  separators?: string[]\n}\n\nexport interface TextChunk {\n  text: string\n  index: number\n  /** Character range [start, end] in original text */\n  range: [number, number]\n  /** Line range [startLine, endLine] (1-indexed) */\n  lines: [number, number]\n}\n\n/**\n * Convert character offset to line number (1-indexed)\n */\nfunction offsetToLine(text: string, offset: number): number {\n  let line = 1\n  for (let i = 0; i < offset && i < text.length; i++) {\n    if (text[i] === '\\n')\n      line++\n  }\n  return line\n}\n\n/**\n * Split text recursively using markdown-aware separators\n */\nexport function splitText(\n  text: string,\n  options: SplitTextOptions = {},\n): TextChunk[] {\n  const {\n    chunkSize = 1000,\n    chunkOverlap = 200,\n    separators = MARKDOWN_SEPARATORS,\n  } = options\n\n  if (text.length <= chunkSize) {\n    const endLine = offsetToLine(text, text.length)\n    return [{ text, index: 0, range: [0, text.length], lines: [1, endLine] }]\n  }\n\n  const chunks = splitRecursive(text, chunkSize, separators)\n  return mergeChunks(chunks, chunkSize, chunkOverlap, text)\n}\n\nfunction splitRecursive(\n  text: string,\n  chunkSize: number,\n  separators: string[],\n): string[] {\n  if (text.length <= chunkSize || separators.length === 0) {\n    return [text]\n  }\n\n  const separator = separators.find(sep => sep === '' || text.includes(sep))\n  if (!separator && separator !== '') {\n    return [text]\n  }\n\n  const parts = separator === '' ? [...text] : text.split(separator)\n  const results: string[] = []\n\n  for (let i = 0; i < parts.length; i++) {\n    const part = parts[i]!\n    const withSep = i < parts.length - 1 && separator !== ''\n      ? part + separator\n      : part\n\n    if (withSep.length <= chunkSize) {\n      results.push(withSep)\n    }\n    else {\n      // Recurse with remaining separators\n      const subParts = splitRecursive(withSep, chunkSize, separators.slice(1))\n      results.push(...subParts)\n    }\n  }\n\n  return results\n}\n\nfunction mergeChunks(\n  parts: string[],\n  chunkSize: number,\n  chunkOverlap: number,\n  originalText: string,\n): TextChunk[] {\n  const chunks: TextChunk[] = []\n  let current = ''\n  let currentStart = 0\n\n  for (const part of parts) {\n    if (current.length + part.length <= chunkSize) {\n      current += part\n    }\n    else {\n      if (current) {\n        const start = originalText.indexOf(current, currentStart)\n        const actualStart = start >= 0 ? start : currentStart\n        const actualEnd = actualStart + current.length\n        chunks.push({\n          text: current,\n          index: chunks.length,\n          range: [actualStart, actualEnd],\n          lines: [offsetToLine(originalText, actualStart), offsetToLine(originalText, actualEnd)],\n        })\n        currentStart = Math.max(0, actualStart + current.length - chunkOverlap)\n      }\n\n      // Start new chunk, possibly with overlap from previous\n      if (chunkOverlap > 0 && current.length > chunkOverlap) {\n        const overlap = current.slice(-chunkOverlap)\n        current = overlap + part\n      }\n      else {\n        current = part\n      }\n    }\n  }\n\n  // Don't forget the last chunk\n  if (current) {\n    const start = originalText.indexOf(current, currentStart)\n    const actualStart = start >= 0 ? start : currentStart\n    const actualEnd = start >= 0 ? start + current.length : originalText.length\n    chunks.push({\n      text: current,\n      index: chunks.length,\n      range: [actualStart, actualEnd],\n      lines: [offsetToLine(originalText, actualStart), offsetToLine(originalText, actualEnd)],\n    })\n  }\n\n  return chunks\n}\n"],"mappings":"AAIA,MAAM,sBAAsB;CAC1B;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACD;AAoBD,SAAS,aAAa,MAAc,QAAwB;CAC1D,IAAI,OAAO;AACX,MAAK,IAAI,IAAI,GAAG,IAAI,UAAU,IAAI,KAAK,QAAQ,IAC7C,KAAI,KAAK,OAAO,KACd;AAEJ,QAAO;;AAMT,SAAgB,UACd,MACA,UAA4B,EAAE,EACjB;CACb,MAAM,EACJ,YAAY,KACZ,eAAe,KACf,aAAa,wBACX;AAEJ,KAAI,KAAK,UAAU,WAAW;EAC5B,MAAM,UAAU,aAAa,MAAM,KAAK,OAAO;AAC/C,SAAO,CAAC;GAAE;GAAM,OAAO;GAAG,OAAO,CAAC,GAAG,KAAK,OAAO;GAAE,OAAO,CAAC,GAAG,QAAQ;GAAE,CAAC;;AAI3E,QAAO,YADQ,eAAe,MAAM,WAAW,WAAW,EAC/B,WAAW,cAAc,KAAK;;AAG3D,SAAS,eACP,MACA,WACA,YACU;AACV,KAAI,KAAK,UAAU,aAAa,WAAW,WAAW,EACpD,QAAO,CAAC,KAAK;CAGf,MAAM,YAAY,WAAW,MAAK,QAAO,QAAQ,MAAM,KAAK,SAAS,IAAI,CAAC;AAC1E,KAAI,CAAC,aAAa,cAAc,GAC9B,QAAO,CAAC,KAAK;CAGf,MAAM,QAAQ,cAAc,KAAK,CAAC,GAAG,KAAK,GAAG,KAAK,MAAM,UAAU;CAClE,MAAM,UAAoB,EAAE;AAE5B,MAAK,IAAI,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK;EACrC,MAAM,OAAO,MAAM;EACnB,MAAM,UAAU,IAAI,MAAM,SAAS,KAAK,cAAc,KAClD,OAAO,YACP;AAEJ,MAAI,QAAQ,UAAU,UACpB,SAAQ,KAAK,QAAQ;OAElB;GAEH,MAAM,WAAW,eAAe,SAAS,WAAW,WAAW,MAAM,EAAE,CAAC;AACxE,WAAQ,KAAK,GAAG,SAAS;;;AAI7B,QAAO;;AAGT,SAAS,YACP,OACA,WACA,cACA,cACa;CACb,MAAM,SAAsB,EAAE;CAC9B,IAAI,UAAU;CACd,IAAI,eAAe;AAEnB,MAAK,MAAM,QAAQ,MACjB,KAAI,QAAQ,SAAS,KAAK,UAAU,UAClC,YAAW;MAER;AACH,MAAI,SAAS;GACX,MAAM,QAAQ,aAAa,QAAQ,SAAS,aAAa;GACzD,MAAM,cAAc,SAAS,IAAI,QAAQ;GACzC,MAAM,YAAY,cAAc,QAAQ;AACxC,UAAO,KAAK;IACV,MAAM;IACN,OAAO,OAAO;IACd,OAAO,CAAC,aAAa,UAAU;IAC/B,OAAO,CAAC,aAAa,cAAc,YAAY,EAAE,aAAa,cAAc,UAAU,CAAC;IACxF,CAAC;AACF,kBAAe,KAAK,IAAI,GAAG,cAAc,QAAQ,SAAS,aAAa;;AAIzE,MAAI,eAAe,KAAK,QAAQ,SAAS,aAEvC,WADgB,QAAQ,MAAM,CAAC,aAAa,GACxB;MAGpB,WAAU;;AAMhB,KAAI,SAAS;EACX,MAAM,QAAQ,aAAa,QAAQ,SAAS,aAAa;EACzD,MAAM,cAAc,SAAS,IAAI,QAAQ;EACzC,MAAM,YAAY,SAAS,IAAI,QAAQ,QAAQ,SAAS,aAAa;AACrE,SAAO,KAAK;GACV,MAAM;GACN,OAAO,OAAO;GACd,OAAO,CAAC,aAAa,UAAU;GAC/B,OAAO,CAAC,aAAa,cAAc,YAAY,EAAE,aAAa,cAAc,UAAU,CAAC;GACxF,CAAC;;AAGJ,QAAO"}

package/dist/types.d.mts

@@ -0,0 +1,37 @@
+//#region src/types.d.ts
+interface SkillConfig {
+  /** Base URL or llms.txt URL */
+  url: string;
+  /** Output directory for skill files */
+  outputDir?: string;
+  /** Chunk size in characters */
+  chunkSize?: number;
+  /** Chunk overlap in characters */
+  chunkOverlap?: number;
+  /** Max pages to fetch */
+  maxPages?: number;
+  /** Skip llms.txt check and always crawl */
+  skipLlmsTxt?: boolean;
+  /** Embedding model */
+  model?: string;
+}
+interface SkillResult {
+  /** Site name (hostname) */
+  siteName: string;
+  /** Path to SKILL.md */
+  skillPath: string;
+  /** Path to references directory */
+  referencesDir: string;
+  /** Path to search database */
+  dbPath: string;
+  /** Number of chunks indexed */
+  chunkCount: number;
+}
+interface FetchedDoc {
+  url: string;
+  title: string;
+  content: string;
+}
+//#endregion
+export { FetchedDoc, SkillConfig, SkillResult };
+//# sourceMappingURL=types.d.mts.map

package/dist/types.d.mts.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.d.mts","names":[],"sources":["../src/types.ts"],"mappings":";UAAiB,WAAA;EAAA;EAEf,GAAA;;EAEA,SAAA;EAFA;EAIA,SAAA;EAAA;EAEA,YAAA;EAEA;EAAA,QAAA;EAIA;EAFA,WAAA;EAEK;EAAL,KAAA;AAAA;AAAA,UAGe,WAAA;EAAW;EAE1B,QAAA;EAEA;EAAA,SAAA;EAIA;EAFA,aAAA;EAIU;EAFV,MAAA;EAKe;EAHf,UAAA;AAAA;AAAA,UAGe,UAAA;EACf,GAAA;EACA,KAAA;EACA,OAAA;AAAA"}

package/dist/types.mjs

@@ -0,0 +1 @@
+export {};

package/package.json

@@ -0,0 +1,52 @@
+{
+  "name": "skilld",
+  "type": "module",
+  "version": "0.0.1",
+  "description": "Generate searchable skills from documentation sites using llms.txt and crawling.",
+  "author": {
+    "name": "Harlan Wilton",
+    "email": "harlan@harlanzw.com",
+    "url": "https://harlanzw.com/"
+  },
+  "license": "MIT",
+  "repository": {
+    "type": "git",
+    "url": "https://github.com/harlan-zw/skilld"
+  },
+  "exports": {
+    ".": {
+      "types": "./dist/index.d.mts",
+      "import": "./dist/index.mjs"
+    }
+  },
+  "main": "./dist/index.mjs",
+  "types": "./dist/index.d.mts",
+  "bin": {
+    "skilld": "./dist/cli.mjs"
+  },
+  "files": [
+    "dist"
+  ],
+  "dependencies": {
+    "@huggingface/transformers": "^3.8.1",
+    "@mdream/crawl": "^0.15.3",
+    "citty": "^0.1.6",
+    "consola": "^3.4.2",
+    "mdream": "^0.15.3",
+    "retriv": "link:../retriv"
+  },
+  "devDependencies": {
+    "@antfu/eslint-config": "^6.7.3",
+    "@types/node": "^22.10.0",
+    "obuild": "^0.4.14",
+    "typescript": "^5.9.3",
+    "vitest": "^4.0.16"
+  },
+  "scripts": {
+    "build": "obuild",
+    "dev:prepare": "obuild --stub",
+    "lint": "eslint .",
+    "typecheck": "tsc --noEmit",
+    "test": "vitest"
+  }
+}