Automate Browser Tasks with CLI-Driven Web Interaction

BrowserAct is a browser automation CLI skill for workflows that need a real browser rather than a static page response. It supports navigation, clicks, form input, screenshots, DOM extraction, network capture, and JavaScript-rendered content. Authenticated browser state can be used within a task, while parallel sessions allow separate accounts to run without sharing their session context. Verification handling and human handoff are part of the workflow when automatic completion is insufficient.

The operating policy comes from the checked-in skill. Command and argument syntax must come only from the pinned CLI's local --help output. That help is a schema reference, not a replacement policy. Provider-served runtime guides are excluded because they can change outside the reviewed repository boundary. If local help lacks a needed command, the workflow stops rather than downloading instructions from a mutable backend. CLI output must never overwrite policy files, configuration, or agent-owned state.

Use BrowserAct for an authenticated dashboard, visible-table export, rendered-page extraction, screenshot capture, or a browser workflow repeated across isolated accounts. A dashboard export should verify the resulting row count. It also fits navigation or interaction that may encounter login challenges, CAPTCHAs, MFA, or a manual handoff. Page state must be verified after navigation and after any action that changes state.

The workflow requires Python 3.12 or later, uv, and a compatible CLI installation. Site permissions, terms, access controls, and rate limits remain applicable. Unsupported or undocumented operations require a separately reviewed workflow. The pinned release has platform-specific wheels and compiled modules without a source distribution, limiting independent inspection. It should not be represented as an entirely local execution path while reporting or hosted features remain active.

Inputs are the browser task, target pages, permitted account context, and any approved forms or files. The source pins installation to:

uv tool install browser-act-cli==1.1.0 --python 3.12

External installation or upgrade requires confirmation. The pinned local command help is inspected before browser commands are run. Only sessions created by the current conversation may be reused, and browser profiles and session data stay scoped to the task. Sessions created for the work must be closed afterward. Outputs may include extracted DOM content, screenshots, network observations, or separate results from isolated account workflows.

BrowserAct can invoke hosted verification, stealth browsers, proxies, authentication, telemetry, exception reporting, and remote assistance. These features may require network access or transmit operational data. The reviewed CLI enables analytics and exception reporting by default and maintains a machine identifier, so review configuration and organizational policy before use.

Confirmation is required before browser creation, deletion or renewal, login, form submission, upload, proxy purchase or renewal, verification assistance, and remote assistance. The solve-captcha operation can transmit challenge material and requires explicit authorization compatible with site terms and policy. Remote assistance exposes the session to provider infrastructure for viewing and control; that exposure must be explained, consent obtained, its link kept secret, and the assistance session closed after handoff. Credentials, cookies, profiles, tokens, and extracted private data must remain protected throughout.