skill-tools

A Node.js utility skill that equips AI agents with file-gathering and knowledge-processing capabilities.

When to Use This Skill

Use this skill when you need to:

• Search the web for PDF documents or images without an API key
• Download files safely (streaming, MIME validation, auto-extension correction)
• Convert a PDF into clean, readable Markdown suitable for LLM consumption
• Generate speech from text entirely offline using the Supertonic v2 ONNX model

Setup

Install dependencies from the skill directory:

npm install

Import any function in an ESM script:

import { searchPDFs, downloadFile, convertPdfToMarkdown, generateSpeech } from './index.js';

Available Tools

searchPDFs(query, limit = 10)

Searches DuckDuckGo for PDF files. Returns { title, url }[]. Includes a 1–3 s random delay to avoid bot blocks.

const pdfs = await searchPDFs('machine learning tutorial', 5);

searchImages(query, limit = 10)

Searches Google Images via google-img-scrap. No API key needed. Returns { title, url }[].

const images = await searchImages('space wallpaper hd', 3);

downloadFile(url, outputDir, expectedType)

Stream-downloads a file to disk using minimal memory. Validates the real MIME type with magic-number detection and fixes the extension automatically.

• expectedType: 'pdf' | 'image'
• Returns the saved file path, or null on failure.

const path = await downloadFile(pdfs[0].url, './downloads', 'pdf');

convertPdfToMarkdown(pdfPath, options?)

Converts a PDF to structured Markdown using Mozilla PDF.js (unpdf v1.4.0 — ESM native, zero dependencies).

Options (all default true except includeMetadata):

const md = await convertPdfToMarkdown('./report.pdf', { includeMetadata: true });

generateSpeech(text, outputPath, options?)

Generates a WAV audio file from text using the Supertonic v2 ONNX model on-device via onnxruntime-node.

Requires model: Download supertonic.onnx from HuggingFace — supertone-inc/supertonic and place it at ./models/supertonic.onnx.

await generateSpeech('Hello!', './out.wav', { voice: 'M2', speed: 1.2, steps: 30 });

End-to-End Example

import { searchPDFs, downloadFile, convertPdfToMarkdown, generateSpeech } from './index.js';
import fs from 'fs/promises';

await fs.mkdir('./downloads', { recursive: true });

// 1. Find a PDF
const [first] = await searchPDFs('javascript guide', 3);

// 2. Download it
const pdfPath = await downloadFile(first.url, './downloads', 'pdf');

// 3. Convert to Markdown
const markdown = await convertPdfToMarkdown(pdfPath, { includeMetadata: true });
await fs.writeFile(pdfPath.replace('.pdf', '.md'), markdown);

// 4. Narrate a summary (requires ONNX model)
await generateSpeech(markdown.slice(0, 400), './summary.wav', { voice: 'F1' });

Notes

• All functions are async and return null on recoverable errors.
• TTS requires the Supertonic ONNX model (not bundled — too large for git).
• No API keys are required for search or download.