markit-ai 0.3.0 → 0.4.0

package/dist/commands/convert.d.ts

@@ -2,4 +2,5 @@ import type { OutputOptions } from "../utils/output.js";
 export declare function convert(source: string, options: OutputOptions & {
     output?: string;
     prompt?: string;
+    imageDir?: string;
 }): Promise<void>;

package/dist/commands/convert.js

@@ -1,4 +1,6 @@
-import { writeFileSync } from "node:fs";
+import { mkdtempSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
 import { loadConfig } from "../config.js";
 import { Markit } from "../markit.js";
 import { loadAllPlugins } from "../plugins/loader.js";
@@ -23,6 +25,8 @@ export async function convert(source, options) {
     }
     const llmFunctions = createLlmFunctions(config, options.prompt);
     const markit = new Markit(llmFunctions, plugins);
+    // Auto-create a temp dir for images if not explicitly provided
+    const imageDir = options.imageDir || mkdtempSync(join(tmpdir(), "markit-images-"));
     try {
         let result;
         const isStdin = source === "-";
@@ -36,7 +40,7 @@ export async function convert(source, options) {
                 process.exit(EXIT_ERROR);
             }
             const buffer = await readStdin();
-            result = await markit.convert(buffer, {});
+            result = await markit.convert(buffer, { imageDir });
         }
         else if (isUrl) {
             // Progress hint for URL fetches (stderr so it doesn't pollute piped output)
@@ -46,7 +50,7 @@ export async function convert(source, options) {
             result = await markit.convertUrl(source);
         }
         else {
-            result = await markit.convertFile(source);
+            result = await markit.convertFile(source, { imageDir });
         }
         const label = isStdin ? "stdin" : source;
         // Write to file or stdout

package/dist/commands/formats.js

@@ -23,6 +23,11 @@ const BUILTIN_FORMATS = [
         extensions: [".mp3", ".wav", ".m4a", ".flac"],
         builtin: true,
     },
+    {
+        name: "GitHub",
+        extensions: ["github.com/*", "gist.github.com/*"],
+        builtin: true,
+    },
     { name: "ZIP", extensions: [".zip"], builtin: true },
     {
         name: "Plain text",

package/dist/converters/docx.d.ts

@@ -2,5 +2,5 @@ import type { ConversionResult, Converter, StreamInfo } from "../types.js";
 export declare class DocxConverter implements Converter {
     name: string;
     accepts(streamInfo: StreamInfo): boolean;
-    convert(input: Buffer, _streamInfo: StreamInfo): Promise<ConversionResult>;
+    convert(input: Buffer, streamInfo: StreamInfo): Promise<ConversionResult>;
 }

package/dist/converters/docx.js

@@ -1,3 +1,5 @@
+import { mkdirSync, writeFileSync } from "node:fs";
+import { join } from "node:path";
 import mammoth from "mammoth";
 import { createTurndown, normalizeTablesHtml } from "../utils/turndown.js";
 const EXTENSIONS = [".docx"];
@@ -16,10 +18,40 @@ export class DocxConverter {
         }
         return false;
     }
-    async convert(input, _streamInfo) {
-        const { value: html } = await mammoth.convertToHtml({ buffer: input });
+    async convert(input, streamInfo) {
+        const imageDir = streamInfo.imageDir;
+        if (imageDir) {
+            mkdirSync(imageDir, { recursive: true });
+        }
+        let imageCount = 0;
+        const convertImage = imageDir
+            ? mammoth.images.imgElement((image) => {
+                imageCount++;
+                const ext = (image.contentType?.split("/")[1] || "png").replace("jpeg", "jpg");
+                const filename = `image_${imageCount}.${ext}`;
+                const filepath = join(imageDir, filename);
+                return image.read("base64").then((base64) => {
+                    writeFileSync(filepath, Buffer.from(base64, "base64"));
+                    return { src: filepath, alt: `image_${imageCount}` };
+                });
+            })
+            : mammoth.images.imgElement((image) => {
+                imageCount++;
+                const contentType = image.contentType || "image/png";
+                return image.read("base64").then((base64) => {
+                    return {
+                        src: `data:${contentType};base64,${base64.slice(0, 0)}`,
+                        alt: `image_${imageCount}`,
+                    };
+                });
+            });
+        const { value: html } = await mammoth.convertToHtml({ buffer: input }, { convertImage });
         const turndown = createTurndown();
-        const markdown = turndown.turndown(normalizeTablesHtml(html));
+        let markdown = turndown.turndown(normalizeTablesHtml(html));
+        // Replace data URI images with comment placeholders when no imageDir
+        if (!imageDir) {
+            markdown = markdown.replace(/!\[([^\]]*)\]\(data:[^)]*\)/g, "<!-- image: $1 -->");
+        }
         return { markdown: markdown.trim() };
     }
 }

package/dist/converters/epub.js

@@ -23,6 +23,7 @@ export class EpubConverter {
             ignoreAttributes: false,
             attributeNamePrefix: "@_",
             textNodeName: "#text",
+            processEntities: { maxTotalExpansions: 1_000_000 },
         });
         // Find content.opf path from container.xml
         const containerXml = await zip

package/dist/converters/github.d.ts

@@ -0,0 +1,18 @@
+import type { ConversionResult, Converter, MarkitOptions, StreamInfo } from "../types.js";
+/**
+ * Matches GitHub URLs and fetches clean markdown content directly
+ * from raw endpoints or the GitHub API — no HTML scraping needed.
+ *
+ * Supported patterns:
+ *   - Repos:   github.com/owner/repo          → raw README.md
+ *   - Files:   github.com/owner/repo/blob/…   → raw file content
+ *   - Gists:   gist.github.com/owner/id       → raw gist content
+ *   - Issues:  github.com/owner/repo/issues/N  → API (title + body)
+ *   - PRs:     github.com/owner/repo/pull/N    → API (title + body)
+ */
+export declare class GitHubConverter implements Converter {
+    name: string;
+    accepts(streamInfo: StreamInfo): boolean;
+    convertUrl(url: string, _options?: MarkitOptions): Promise<ConversionResult>;
+    convert(_input: Buffer, streamInfo: StreamInfo): Promise<ConversionResult>;
+}

package/dist/converters/github.js

@@ -0,0 +1,148 @@
+const GITHUB_HOSTS = new Set([
+    "github.com",
+    "www.github.com",
+    "gist.github.com",
+]);
+/**
+ * Matches GitHub URLs and fetches clean markdown content directly
+ * from raw endpoints or the GitHub API — no HTML scraping needed.
+ *
+ * Supported patterns:
+ *   - Repos:   github.com/owner/repo          → raw README.md
+ *   - Files:   github.com/owner/repo/blob/…   → raw file content
+ *   - Gists:   gist.github.com/owner/id       → raw gist content
+ *   - Issues:  github.com/owner/repo/issues/N  → API (title + body)
+ *   - PRs:     github.com/owner/repo/pull/N    → API (title + body)
+ */
+export class GitHubConverter {
+    name = "github";
+    accepts(streamInfo) {
+        if (!streamInfo.url)
+            return false;
+        try {
+            const { hostname } = new URL(streamInfo.url);
+            return GITHUB_HOSTS.has(hostname);
+        }
+        catch {
+            return false;
+        }
+    }
+    async convertUrl(url, _options) {
+        const parsed = new URL(url);
+        if (parsed.hostname === "gist.github.com") {
+            return fetchGist(parsed);
+        }
+        const segments = parsed.pathname.split("/").filter(Boolean);
+        // Need at least owner/repo
+        if (segments.length < 2) {
+            throw new Error(`Unsupported GitHub URL: ${url}`);
+        }
+        const [owner, repo, type, ...rest] = segments;
+        // github.com/owner/repo/blob/ref/path → raw file
+        if (type === "blob" && rest.length >= 2) {
+            const ref = rest[0];
+            const filePath = rest.slice(1).join("/");
+            return fetchRawFile(owner, repo, ref, filePath);
+        }
+        // github.com/owner/repo/issues/N or /pull/N
+        if ((type === "issues" || type === "pull") && rest[0]) {
+            const number = Number.parseInt(rest[0], 10);
+            if (!Number.isNaN(number)) {
+                return fetchIssueOrPr(owner, repo, number);
+            }
+        }
+        // github.com/owner/repo (no subpath or tree/wiki/etc) → README
+        if (!type) {
+            return fetchReadme(owner, repo);
+        }
+        throw new Error(`Unsupported GitHub URL pattern: ${url}`);
+    }
+    async convert(_input, streamInfo) {
+        // GitHub URLs are handled entirely via convertUrl.
+        // If we end up here, the URL was already fetched by the default path —
+        // just delegate to convertUrl.
+        if (streamInfo.url) {
+            return this.convertUrl(streamInfo.url);
+        }
+        throw new Error("GitHub converter requires a URL");
+    }
+}
+// ---------------------------------------------------------------------------
+// Fetchers
+// ---------------------------------------------------------------------------
+async function fetchReadme(owner, repo) {
+    const url = `https://raw.githubusercontent.com/${owner}/${repo}/HEAD/README.md`;
+    const res = await fetch(url);
+    if (!res.ok) {
+        throw new Error(`Failed to fetch README: ${res.status} ${res.statusText}`);
+    }
+    const markdown = (await res.text()).trim();
+    const title = extractFirstHeading(markdown) ?? `${owner}/${repo}`;
+    return { markdown, title };
+}
+async function fetchRawFile(owner, repo, ref, filePath) {
+    const url = `https://raw.githubusercontent.com/${owner}/${repo}/${ref}/${filePath}`;
+    const res = await fetch(url);
+    if (!res.ok) {
+        throw new Error(`Failed to fetch file: ${res.status} ${res.statusText}`);
+    }
+    const content = (await res.text()).trim();
+    const filename = filePath.split("/").pop() ?? filePath;
+    // If it's markdown, return as-is. Otherwise wrap in a code block.
+    if (filePath.endsWith(".md") || filePath.endsWith(".mdx")) {
+        const title = extractFirstHeading(content) ?? filename;
+        return { markdown: content, title };
+    }
+    const ext = filename.includes(".") ? filename.split(".").pop() : "";
+    const markdown = `# ${filename}\n\n\`\`\`${ext}\n${content}\n\`\`\``;
+    return { markdown, title: filename };
+}
+async function fetchGist(parsed) {
+    const segments = parsed.pathname.split("/").filter(Boolean);
+    // gist.github.com/owner/id
+    const [owner, id] = segments;
+    if (!owner || !id) {
+        throw new Error(`Unsupported gist URL: ${parsed.href}`);
+    }
+    const url = `https://gist.githubusercontent.com/${owner}/${id}/raw`;
+    const res = await fetch(url);
+    if (!res.ok) {
+        throw new Error(`Failed to fetch gist: ${res.status} ${res.statusText}`);
+    }
+    const content = (await res.text()).trim();
+    const title = `gist:${id}`;
+    return { markdown: content, title };
+}
+async function fetchIssueOrPr(owner, repo, number) {
+    const url = `https://api.github.com/repos/${owner}/${repo}/issues/${number}`;
+    const res = await fetch(url, {
+        headers: { Accept: "application/vnd.github.v3+json" },
+    });
+    if (!res.ok) {
+        throw new Error(`Failed to fetch issue/PR: ${res.status} ${res.statusText}`);
+    }
+    const data = (await res.json());
+    const title = data.title ?? `#${number}`;
+    const parts = [`# ${title}`];
+    // Metadata line
+    const meta = [];
+    if (data.user?.login)
+        meta.push(`@${data.user.login}`);
+    if (data.state)
+        meta.push(data.state);
+    if (data.labels?.length) {
+        meta.push(data.labels.map((l) => l.name).join(", "));
+    }
+    if (meta.length > 0)
+        parts.push(meta.join(" · "));
+    if (data.body?.trim())
+        parts.push(data.body.trim());
+    return { markdown: parts.join("\n\n"), title };
+}
+// ---------------------------------------------------------------------------
+// Helpers
+// ---------------------------------------------------------------------------
+function extractFirstHeading(markdown) {
+    const match = markdown.match(/^#\s+(.+)$/m);
+    return match?.[1]?.trim();
+}

package/dist/converters/pptx.d.ts

@@ -2,7 +2,7 @@ import type { ConversionResult, Converter, StreamInfo } from "../types.js";
 export declare class PptxConverter implements Converter {
     name: string;
     accepts(streamInfo: StreamInfo): boolean;
-    convert(input: Buffer, _streamInfo: StreamInfo): Promise<ConversionResult>;
+    convert(input: Buffer, streamInfo: StreamInfo): Promise<ConversionResult>;
     private extractText;
     private extractTable;
 }

package/dist/converters/pptx.js

@@ -1,3 +1,5 @@
+import { mkdirSync, writeFileSync } from "node:fs";
+import { join } from "node:path";
 import { XMLParser } from "fast-xml-parser";
 import JSZip from "jszip";
 const EXTENSIONS = [".pptx"];
@@ -14,12 +16,13 @@ export class PptxConverter {
             return true;
         return false;
     }
-    async convert(input, _streamInfo) {
+    async convert(input, streamInfo) {
         const zip = await JSZip.loadAsync(input);
         const parser = new XMLParser({
             ignoreAttributes: false,
             attributeNamePrefix: "@_",
             textNodeName: "#text",
+            processEntities: { maxTotalExpansions: 1_000_000 },
         });
         // Get slide order from presentation.xml
         const presXml = await zip.file("ppt/presentation.xml")?.async("string");
@@ -66,7 +69,12 @@ export class PptxConverter {
             });
             slidePaths.push(...slideFiles);
         }
+        const imageDir = streamInfo.imageDir;
+        if (imageDir) {
+            mkdirSync(imageDir, { recursive: true });
+        }
         const sections = [];
+        let imageCount = 0;
         for (let i = 0; i < slidePaths.length; i++) {
             const slideXml = await zip.file(slidePaths[i])?.async("string");
             if (!slideXml)
@@ -75,6 +83,17 @@ export class PptxConverter {
             const spTree = slide["p:sld"]?.["p:cSld"]?.["p:spTree"];
             if (!spTree)
                 continue;
+            // Parse slide-level rels for image references
+            const slideRelsPath = `${slidePaths[i].replace("slides/slide", "slides/_rels/slide")}.rels`;
+            const slideRelsXml = await zip.file(slideRelsPath)?.async("string");
+            const slideRelMap = new Map();
+            if (slideRelsXml) {
+                const slideRels = parser.parse(slideRelsXml);
+                const relItems = toList(slideRels?.Relationships?.Relationship);
+                for (const r of relItems) {
+                    slideRelMap.set(r["@_Id"], r["@_Target"]);
+                }
+            }
             const slideLines = [`<!-- Slide ${i + 1} -->`];
             const shapes = spTree["p:sp"];
             const shapeList = Array.isArray(shapes) ? shapes : shapes ? [shapes] : [];
@@ -91,6 +110,55 @@ export class PptxConverter {
                     slideLines.push(text);
                 }
             }
+            // Extract embedded images
+            const pics = toList(spTree["p:pic"]);
+            for (const pic of pics) {
+                const blipFill = pic["p:blipFill"];
+                const rEmbed = blipFill?.["a:blip"]?.["@_r:embed"];
+                if (!rEmbed)
+                    continue;
+                const target = slideRelMap.get(rEmbed);
+                if (!target)
+                    continue;
+                // Resolve relative target against slide directory
+                const imagePath = target.startsWith("/")
+                    ? target.slice(1)
+                    : `ppt/slides/${target}`;
+                // Normalize path (e.g. ppt/slides/../media/image1.png → ppt/media/image1.png)
+                const normalizedPath = imagePath
+                    .split("/")
+                    .reduce((parts, seg) => {
+                    if (seg === "..")
+                        parts.pop();
+                    else
+                        parts.push(seg);
+                    return parts;
+                }, [])
+                    .join("/");
+                const imageFile = zip.file(normalizedPath);
+                if (!imageFile)
+                    continue;
+                imageCount++;
+                const name = pic["p:nvSpPr"]?.["p:cNvPr"]?.["@_name"] ||
+                    pic["p:nvPicPr"]?.["p:cNvPr"]?.["@_name"] ||
+                    `image_${imageCount}`;
+                if (imageDir) {
+                    try {
+                        const ext = normalizedPath.split(".").pop() || "png";
+                        const filename = `slide${i + 1}_${imageCount}.${ext}`;
+                        const filepath = join(imageDir, filename);
+                        const buf = await imageFile.async("nodebuffer");
+                        writeFileSync(filepath, buf);
+                        slideLines.push(`![${name}](${filepath})`);
+                    }
+                    catch {
+                        slideLines.push(`<!-- image: ${name} (slide ${i + 1}) -->`);
+                    }
+                }
+                else {
+                    slideLines.push(`<!-- image: ${name} (slide ${i + 1}) -->`);
+                }
+            }
             // Tables
             const graphicFrames = spTree["p:graphicFrame"];
             const gfList = Array.isArray(graphicFrames)
@@ -214,3 +282,8 @@ export class PptxConverter {
         return lines.join("\n");
     }
 }
+function toList(val) {
+    if (!val)
+        return [];
+    return Array.isArray(val) ? val : [val];
+}

package/dist/converters/xlsx.js

@@ -20,6 +20,7 @@ export class XlsxConverter {
             ignoreAttributes: false,
             attributeNamePrefix: "@_",
             textNodeName: "#text",
+            processEntities: { maxTotalExpansions: 1_000_000 },
         });
         // Parse shared strings
         const ssXml = await zip.file("xl/sharedStrings.xml")?.async("string");

package/dist/index.d.ts

@@ -3,6 +3,7 @@ export { AudioConverter } from "./converters/audio.js";
 export { CsvConverter } from "./converters/csv.js";
 export { DocxConverter } from "./converters/docx.js";
 export { EpubConverter } from "./converters/epub.js";
+export { GitHubConverter } from "./converters/github.js";
 export { HtmlConverter } from "./converters/html.js";
 export { ImageConverter } from "./converters/image.js";
 export { IpynbConverter } from "./converters/ipynb.js";

package/dist/index.js

@@ -2,6 +2,7 @@ export { AudioConverter } from "./converters/audio.js";
 export { CsvConverter } from "./converters/csv.js";
 export { DocxConverter } from "./converters/docx.js";
 export { EpubConverter } from "./converters/epub.js";
+export { GitHubConverter } from "./converters/github.js";
 export { HtmlConverter } from "./converters/html.js";
 export { ImageConverter } from "./converters/image.js";
 export { IpynbConverter } from "./converters/ipynb.js";

package/dist/main.js

@@ -18,6 +18,7 @@ program
     .option("-q, --quiet", "Raw markdown only, no decoration")
     .option("-p, --prompt <text>", "Extra instructions for image description")
     .option("-o, --output <file>", "Write to file instead of stdout")
+    .option("-i, --image-dir <dir>", "Extract images to this directory")
     .addHelpText("after", `
 Examples:
   $ markit report.pdf                  Convert a PDF to markdown
@@ -42,7 +43,8 @@ program
         json: globals.json,
         quiet: globals.quiet,
         output: opts.output,
-        prompt: opts.prompt,
+        prompt: globals.prompt,
+        imageDir: globals.imageDir,
     });
 });
 program
@@ -144,6 +146,7 @@ program.on("command:*", async (args) => {
         quiet: globals.quiet,
         output: globals.output,
         prompt: globals.prompt,
+        imageDir: globals.imageDir,
     });
 });
 // No args → show concise help

package/dist/markit.d.ts

@@ -7,7 +7,7 @@ export declare class Markit {
     /**
      * Convert a local file to markdown.
      */
-    convertFile(path: string): Promise<ConversionResult>;
+    convertFile(path: string, extra?: Partial<StreamInfo>): Promise<ConversionResult>;
     /**
      * Convert a URL to markdown.
      */

package/dist/markit.js

@@ -4,6 +4,7 @@ import { AudioConverter } from "./converters/audio.js";
 import { CsvConverter } from "./converters/csv.js";
 import { DocxConverter } from "./converters/docx.js";
 import { EpubConverter } from "./converters/epub.js";
+import { GitHubConverter } from "./converters/github.js";
 import { HtmlConverter } from "./converters/html.js";
 import { ImageConverter } from "./converters/image.js";
 import { IpynbConverter } from "./converters/ipynb.js";
@@ -32,6 +33,7 @@ export class Markit {
             new XlsxConverter(),
             new EpubConverter(),
             new IpynbConverter(),
+            new GitHubConverter(),
             new WikipediaConverter(),
             new RssConverter(),
             new CsvConverter(),
@@ -56,12 +58,13 @@ export class Markit {
     /**
      * Convert a local file to markdown.
      */
-    async convertFile(path) {
+    async convertFile(path, extra) {
         const buffer = readFileSync(path);
         const streamInfo = {
             localPath: path,
             extension: extname(path).toLowerCase(),
             filename: basename(path),
+            ...extra,
         };
         return this.convert(buffer, streamInfo);
     }
@@ -69,6 +72,18 @@ export class Markit {
      * Convert a URL to markdown.
      */
     async convertUrl(url) {
+        // Let converters with a URL-specific hook handle it first
+        const streamInfo = { url };
+        for (const converter of this.converters) {
+            if (!converter.convertUrl || !converter.accepts(streamInfo))
+                continue;
+            try {
+                return await converter.convertUrl(url, this.options);
+            }
+            catch {
+                // Fall through to default fetch path
+            }
+        }
         const response = await fetch(url, {
             headers: {
                 Accept: "text/markdown, text/html;q=0.9, text/plain;q=0.8, */*;q=0.1",
@@ -84,13 +99,13 @@ export class Markit {
         const urlPath = new URL(url).pathname;
         const ext = extname(urlPath).toLowerCase();
         const buffer = Buffer.from(await response.arrayBuffer());
-        const streamInfo = {
+        const fetchedInfo = {
             url,
             mimetype: mimetype.trim(),
             extension: ext || undefined,
             filename: basename(urlPath) || undefined,
         };
-        return this.convert(buffer, streamInfo);
+        return this.convert(buffer, fetchedInfo);
     }
     /**
      * Convert a buffer with stream info to markdown.

package/dist/types.d.ts

@@ -25,6 +25,12 @@ export interface Converter {
     name: string;
     /** Quick check: can this converter handle the given stream? */
     accepts(streamInfo: StreamInfo): boolean;
+    /**
+     * Optional URL-first hook. When present, called before the default fetch
+     * so the converter can handle URL fetching itself (e.g. rewrite to a raw
+     * content URL or call an API).
+     */
+    convertUrl?(url: string, options?: MarkitOptions): Promise<ConversionResult>;
     /** Convert the source to markdown */
     convert(input: Buffer, streamInfo: StreamInfo, options?: MarkitOptions): Promise<ConversionResult>;
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "markit-ai",
-  "version": "0.3.0",
+  "version": "0.4.0",
   "description": "Convert anything to markdown. PDF, DOCX, PPTX, XLSX, HTML, EPUB, Jupyter, RSS, images, audio, URLs, and more. Pluggable converters, built-in LLM providers for image description and audio transcription. Works as a CLI and as a library.",
   "type": "module",
   "main": "dist/index.js",