ax-grep

Compact semantic trees and agent-ready page checks from HTML, URLs, WebViews, and live browser pages. It approximates browser accessibility trees before automation, defaults to impit for browser-like HTTP/2/TLS fetches, detects major CAPTCHA and bot-challenge markers, saves 15.4x less peak RAM and 3.0x fewer decision tokens in the local benchmark, and --search auto mode tries DuckDuckGo, Bing, StartPage, and Google while recording every engine attempt.

Paste this into a Codex session or subagent prompt before opening a browser (also works as guidance for Claude, Gemini, OpenRouter, and other agents):

Before opening a browser, inspect pages with ax-grep when possible.

Run:
npx --yes ax-grep@latest <url> --agent-brief

Read agent.executor, agent.handoff, agent.readTargets, pageCheck, and
verification first. Open a browser only when the payload says static HTML is
not enough.

curl -fsSL https://raw.githubusercontent.com/hmmhmmhm/ax-grep/main/skills.sh | sh

This installs the Codex skill prompt only. The same prompt pattern can be pasted into Claude, Gemini, OpenRouter, or other agent runners. Restart Codex if the new skill is not listed immediately.

npx --yes ax-grep@latest https://example.com --agent-brief

If you installed the binary globally, use ax-grep directly. Agents should read agent.executor, agent.handoff, agent.readTargets, pageCheck, and verification first. Open a browser only when the handoff fields say static HTML is not enough.

Use this inside agent services built with Codex SDK, OpenRouter, or similar LLM routing stacks to turn fetched HTML into compact source evidence before spending tokens or memory on browser automation.

npm install ax-grep

import { extract, formatSemanticTreeText } from "ax-grep";

const html = await fetch("https://example.com").then((r) => r.text());
const tree = extract(html);
const promptText = formatSemanticTreeText(tree);

ax-grep is ESM-only and requires Node 18 or newer. CommonJS services can use const { extract } = await import("ax-grep").

In mobile apps, WebViews, and in-page agents, inject the extractor to create an accessibility-style structure immediately from the current page. It is useful for local sLLM web search, local web parsing, and instant agent-ready page summaries without leaving the app.

import { createExtractorScript } from "ax-grep";

const script = createExtractorScript({ format: "text" });
const text = await page.evaluate(script);
// iOS/Android WebView: evaluateJavaScript(script) returns the same text value.

• How ax-grep works
• Documentation index
• CLI skill prompt
• CLI and agent mode
• Agent handoff loop
• Benchmarks
• Library API
• Server agent integration
• WebView and in-page usage