Browser Recipes

Save and replay browser automation sequences — navigate, click, fill, extract.

Overview

Browser Recipes are pre-built automation sequences that Yukkai can run against any website without you having to spell out each individual step. Think of them as saved macros: you give Yukkai a recipe name and a target (a search query, a URL, a username), and it navigates, interacts, and extracts content on its own.

Instead of manually chaining five or six browser actions — navigate, wait, type, click, extract — you simply say "search Google for AI agents" or "scrape that GitHub repo" and Yukkai picks the right recipe, feeds it the target, and returns the results.

Yukkai ships with 10 ready-to-use recipes covering popular sites:

Recipe Purpose
google_search Run a Google search and collect results
github_repo Extract metadata, README, and stats from a GitHub repository
github_trending List currently trending repos on GitHub
hackernews Fetch top stories from Hacker News
reddit Pull posts or subreddit content
twitter Collect tweets from a user or search
youtube_search Search YouTube and return video metadata
wikipedia Retrieve and summarize a Wikipedia article
amazon_search Search Amazon products and listings
linkedin_company Extract a LinkedIn company profile


How It Works

Each recipe is a bundled sequence of low-level browser actions — the same actions available through the web_browse tool:

  1. Navigate to the target URL (e.g., google.com, github.com).
  2. Wait for the page to finish loading (network idle, selector, or text).
  3. Interact — type into search fields, click buttons, scroll through results.
  4. Extract the relevant content and structure it into a clean Markdown or JSON response.
Under the hood, Yukkai uses its native WebKit browser — the same engine as Safari — so pages render exactly as they would for a human visitor. JavaScript-heavy SPAs (React, Vue, Angular) are fully supported.

Recipes also take advantage of Yukkai's human-like interaction mode when needed:

  • human_click — moves the mouse naturally before clicking.
  • human_type — types character-by-character with realistic timing.
  • human_scroll — scrolls at variable speeds.
This helps bypass simple bot-detection mechanisms on sites that flag automated traffic.


How To Use It

You don't need to memorize recipe names. Just describe what you want in plain language and Yukkai will match it to the right recipe.

In chat

"Search Google for best espresso machines under $500."
"Get the README from the GitHub repo apple/swift-evolution."
"What's trending on GitHub right now?"
"Summarize the Wikipedia article on quantum entanglement."

Yukkai recognizes the intent, selects the appropriate recipe, and passes your query or URL as the target.

Explicit recipe call

You can also name the recipe directly:

"Run the google_search recipe for AI agent frameworks."
"Use the linkedin_company recipe on https://linkedin.com/company/openai."

Listing available recipes

At any time you can ask:

"What browser recipes are available?"

Yukkai will run list_recipes and show you the full catalogue with a short description of each.


Examples

Google Search

You: Search Google for "macOS automation tools".

Yukkai runs the google_search recipe with target "macOS automation tools". It navigates to Google, types the query, waits for results to render, and extracts the top results — titles, URLs, and snippets — all formatted as Markdown.

GitHub Repository

You: Get info on the GitHub repofacebook/react.

Yukkai runs github_repo with target facebook/react. It loads the repository page, waits for the DOM, and extracts:

  • Repository name and description
  • Stars, forks, and watcher counts
  • Primary language
  • README content
  • Recent activity

Hacker News

You: What's on the front page of Hacker News?

Yukkai runs hackernews. It navigates to news.ycombinator.com, scrapes the top 30 stories with their titles, links, points, and comment counts.

YouTube Search

You: Search YouTube for "Lo-fi beats to study to".

Yukkai runs youtube_search with the query. It loads YouTube, enters the search term, waits for results, and returns video titles, channel names, view counts, and URLs.

Amazon Product Search

You: Search Amazon for "mechanical keyboard".

Yukkai runs amazon_search. It navigates to Amazon, enters the product query, and extracts product names, prices, ratings, and image URLs from the results page.


Tips

Combine recipes with follow-up questions

Recipes return raw data. You can then ask Yukkai to analyze it:

"Search Google for best espresso machines under $500, then summarize the top 5 results."

Yukkai will run the recipe, then process the extracted content to produce a concise summary.

Use recipes as a starting point

If a recipe doesn't cover exactly what you need, ask Yukkai to run it and then continue with manual browser actions. For example, run google_search to find a page, then ask Yukkai to navigate into a specific result and extract its full content.

Recipe results can be exported

After a recipe finishes, you can ask Yukkai to export the collected data:

"Export those results as a CSV file."

Yukkai will use export_data to save the data to ~/YukkaiDocs/ in your preferred format (JSON, CSV, or Markdown).

Beware of rate limits and captchas

Recipes that interact with social platforms (Twitter, LinkedIn, Reddit) may encounter rate limits or login walls. For sites requiring authentication:

  1. Open the browser window with show.
  2. Log in manually in the visible window.
  3. Save the session with save_session.
  4. Re-run the recipe — it will reuse your saved cookies.

Recipes are site-specific

Each recipe is tailored to a specific website's DOM structure. If a site changes its layout significantly, the recipe may break. Yukkai will fall back to general-purpose browser automation (navigate + extract) when this happens, so you still get results — just potentially in a less structured format.

Ask for screenshots when unsure

If a recipe returns unexpected results, ask Yukkai to take a screenshot of the page:

"Take a screenshot of what the browser sees right now."

This helps you verify that the page rendered correctly and that no cookie banner or modal is blocking the content.

Batch requests

You can chain multiple recipe calls in a single conversation turn:

"Get the GitHub repo info for apple/swift, google/jax, and microsoft/vscode."

Yukkai will run each recipe sequentially and present all results together.