browser-guidance

Operate JavaScript-rendered websites and web apps through a managed browser tool.

9|Updated Mar 31, 2026
One-click install
npx skills add https://github.com/tezra-io/fermix --skill browser-guidance-tezra-io
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: browser-guidance
Source: https://github.com/tezra-io/fermix/tree/main/apps/fermix_core/priv/skills/browser_guidance
Command: npx skills add https://github.com/tezra-io/fermix --skill browser-guidance-tezra-io

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Choosing the wrong web tool wastes turns and produces empty results: static fetches fail on JavaScript-rendered pages, and browser automation breaks on stale refs, unread pages, and blocked hosts. This Skill provides the decision rules and operating loop for driving websites correctly. ## Core Features & Use Cases - Tool Routing: Decide between web_search, web_fetch, browser, and computer_use based on whether the page is JavaScript-rendered, interactive, or part of the user's own live screen. - Browser Operating Loop: Open pages, read returned snapshots instead of re-snapshotting, fill forms with fill_form, submit, and verify actions through page receipts and ready_state. - Attached Tab Control: Drive one tab of the user's own signed-in browser via the Fermix extension, handling grant, detach, and unsupported-action cases. - Use Case: You need to log into a JavaScript-heavy dashboard, fill a multi-field search form, and extract the results. The Skill tells you to use the browser tool, snapshot once, fill all fields in one fill_form call, submit, and read the changed page from the action receipt. ## Quick Start Ask the agent to open a JavaScript-rendered website in the browser, fill in the search form, and report the results back to you.

Frequently Asked Questions about browser-guidance

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I automate a JavaScript-rendered website with an AI agent?▼

Use the browser tool rather than web_fetch, since static fetches return empty or partial content on dynamic pages. Open the page, read the snapshot returned with the result, then act with current refs and verify through the page receipts each action reports.

When should I use web_fetch versus a browser tool?▼

Use web_fetch for a single known URL with readable server-rendered HTML, and web_search for facts without a known URL. Switch to the browser tool for JavaScript-rendered pages, forms, clicks, logins, and interactive data.

How do I fill multiple form fields in one browser action?▼

Use fill_form with a fields array of up to 12 ref and text pairs taken from a single snapshot; it fills them in order with one value receipt each. It only fills, so follow it with submit or a click to send the form.

Can the browser tool control a tab I already have open?▼

Yes, with profile set to selected_tab after you click the Fermix browser extension on that tab, which hands over one signed-in tab. Open, close, focus, cookies, and download are refused there and must run in the managed profile instead.

Why does the browser refuse to read certain pages or hosts?▼

Reads are checked against a URL policy: file:, data:, chrome: and similar documents are refused, and hosts ending .internal, .local, or .localhost are blocked unless listed in allowed_hosts in config.toml. Loopback addresses stay allowed under every setting.

What should I do when browser automation hits a CAPTCHA or 2FA?▼

Stop and report to the user. The Skill defines explicit stop conditions covering CAPTCHA, 2FA, payment confirmation, missing credentials, URL-policy blocks, unsafe dialogs, and repeated failures with no new information.