Verified Agent Skill record

Ego Browser

ego-browser is a Chromium-based browser designed to be friendly to both human users and AI Agents. AI Agents operate in isolated task spaces that inherit the user's login state, enabling authenticated web interaction without competing for the browser. It provides a CLI-accessible Node.js runtime with preloaded helpers for observing pages, interacting with UI elements, evaluating browser-side JavaScript, and driving real browser automation tasks.

Tasks

Open a website, observe its semantic tree with snapshotText(), and extract text content or links using annotated refs
Fill out an authenticated form or log into a site by reusing the inherited user login state within an agent task space

Inputs

URLs for navigation via openOrReuseTab or gotoAndWait
CSS selectors, xpath=..., @N / ref=N, loc=... values, or viewport coordinates [x, y] for targeting elements
Task space name, numeric id, or digit-only id string for useOrCreateTaskSpace
Absolute file paths for uploadFile

Outputs

Terminal output printed via cliLog(value)
Page text semantic tree with ref=N and loc=... annotations from snapshotText()
Screenshots from captureScreenshot()
Page metadata object { url, title, w, h, sx, sy, pw, ph } or { dialog: ... } from pageInfo()
Accumulated page events (navigation, network, etc.) from drainEvents(), which must be explicitly consumed
Resolution objects such as { done: true } or { done: false, skipped: 'user-owned' } from handOffTaskSpace or completeTaskSpace

Limitations and checks

A user is controlling error is a hard stop on the entire task; you must ask the user and wait, and must not retry or auto-takeover
handOffTaskSpace and completeTaskSpace with { keep: true } are skipped for user-owned spaces, resolving { done: false, skipped: 'user-owned' }
The Node.js runtime exits after each heredoc and retains no state, requiring explicit task space resumption calls
The js() function does not capture closures or provide an argument channel when passing a function
switchTaskSpace throws for user-owned spaces, restricting it to agent-owned spaces only
Verify that each heredoc round begins with useOrCreateTaskSpace or takeOverTaskSpace to resume the correct task space, since runtime state is not retained between rounds
Confirm that completeTaskSpace occupies its own dedicated final heredoc and runs only after prior output confirmed the task is genuinely done
Ensure all final results are printed through cliLog(), as it is the only output mechanism inside a heredoc