control-in-app-browser

Control the session-scoped in-app Browser for navigation, inspection, interaction, and uploads.

1.3k|142|Updated Jun 1, 2026
One-click install
npx skills add https://github.com/MiniMax-AI/minimax-code --skill control-in-app-browser-minimax-ai
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: control-in-app-browser
Source: https://github.com/MiniMax-AI/minimax-code/tree/main/packages/local-runtime-v2/assets/agents/mavis/skills/control-in-app-browser
Command: npx skills add https://github.com/MiniMax-AI/minimax-code --skill control-in-app-browser-minimax-ai

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? It gives the agent a governed way to operate the chat session's built-in Browser provider, handling page inspection, clicking, typing, file uploads, and diagnostics while enforcing strict safety rules around authentication and irreversible actions. ## Core Features & Use Cases - Structured Page Inspection: Inspect actionable elements via opaque refs, paginate truncated snapshots, and read page content with text, DOM, semantic, console, and network queries. - Safe Interaction Loop: Click, fill, type, paste, scroll, hover, and upload current-turn attachments or workspace files, with post-action visual evidence and effect verification. - Authentication and Confirmation Boundaries: Pause for user takeover at login forms and require explicit ask_user confirmation before publish, send, submit, delete, or purchase actions. - Use Case: A user asks the agent to open a web app in the right-side FilePanel Browser, fill a draft form with an attached file, and stop for confirmation before the final submit. ## Quick Start Ask the agent to open a site in the in-app Browser, inspect the page, and fill in a form while stopping before any final submit action.

Frequently Asked Questions about control-in-app-browser

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

FAQPage Schema
How do I control the in-app browser from the chat agent?

Explicitly request the in-app, right-side, or current Browser in your message, and the agent loads this skill and uses the native browser tool. It inspects the page first, then acts through opaque element refs one action at a time.

How does browser page inspection work with truncated snapshots?

The inspect action returns actionable elements with opaque refs, and truncated results include a continuation with snapshotId and offset. The agent copies that continuation unchanged to page through the same snapshot instead of restarting inspection.

Can the agent upload local files through the browser?

Yes, upload_files accepts exact paths from current-turn attachments or regular files inside the active workspace. The action itself validates existence, type, authorization, count, and size, so no separate filesystem preflight is performed.

Does the in-app browser handle login pages automatically?

No, the agent never reads or types credentials, codes, or CAPTCHAs. On a visible provider it pauses with ask_user so you take over the same tab; a headless provider without a takeover surface stops and reports that interactive login is required.

When should I use a connector or API instead of the browser?

When a linked resource has a purpose-built skill, connector, API, or CLI that can perform the operation, that path is preferred. The browser is used only when no such path exists, lacks the capability, or you explicitly request the in-app Browser.

Why does the agent stop before submitting a form in the browser?

Final publish, send, submit, delete, purchase, or account-change actions require explicit confirmation immediately before execution. The agent prepares the draft, states the exact action and content, and waits for your approval via ask_user.