<< All versions

Skill v1.0.0

currentLLM-judged scan90/100
anakin-inc/kiro-power-anakin/anakin-browser
──Details
PublishedOctober 23, 2026 at 12:40 AM
Content Hashsha256:99010cbe635aab79...
Git SHAc3d986cd1b11
──Files
Files (1 file, 3.2 KB)
SKILL.md3.2 KBactive
SKILL.md · 80 lines · 3.2 KB

version: "1.0.0" name: anakin-browser description: Use when a task needs a real browser driven step by step — multi-step flows, clicking and typing, complex navigation that scrape cannot handle — or when reaching login-protected content with a saved browser session. Covers browser_task, session_list, session_delete, and how sessionId works with scrape, crawl, and monitors.


Anakin: browser automation and saved sessions

Two related capabilities: driving a real cloud browser with natural language, and reusing saved login state so any tool can reach protected pages.

browser_task — the last resort, not the first

An AI agent drives a real cloud browser: navigating, clicking, typing, scrolling, extracting. Async, up to ~5 minutes, polls to completion.

Check cheaper options first, in this order:

  1. scrape — one page, static content
  2. scrape with useBrowser: true — one page, JS-rendered
  3. wire_discover — is there a pre-built action? **Wire actions are faster and

cheaper than driving a browser**

  1. browser_task — only when the task is genuinely multi-step or interactive

and nothing above covers it

Use it for: multi-step flows, form filling, interactions behind clicks, and navigation too complex to express as a URL.

Parameters

  • prompt (required) — the task in natural language. Be specific about what to

extract or do.

  • url — where to start.
  • output_schema — supply this whenever you know the shape you want. It is

the difference between prose you must re-parse and structured JSON.

  • session_id — for login-protected tasks (below).
  • max_steps, timeout_ms — bound long-running tasks.

Returns the result plus run metadata (steps taken, duration, run_id).

Two hard rules

  • Never put passwords in the `prompt`. Use session_id. A prompt is not a

credential store.

  • It does not execute payments or transfer funds; such tasks are refused. Do not

attempt to work around this.

Tell the user before starting — several minutes of silence otherwise reads as a hang.

Saved browser sessions

A session is encrypted login state captured through the Anakin dashboard or Browser API. Its id is what you pass as sessionId to reach protected content from any of these:

ToolParameter
scrape, crawlsessionId (pair with useBrowser: true)
monitor_createsessionId
browser_tasksession_id
  • session_list — your saved sessions, optionally filtered by domain. Call

this first to find an id.

  • session_delete — permanently deletes a session and its encrypted login data.

When no session exists

You cannot create one from here. The user must log in once interactively in the Anakin dashboard — that flow handles 2FA and captchas. Say so plainly and point them there rather than attempting a login through browser_task.

Before deleting

session_delete is irreversible. The user must log in again through the dashboard to recreate it, and any monitors or requests referencing that `sessionId` lose authenticated access — a monitor may keep running and silently start recording logged-out content. Check monitor_list for dependents and confirm with the user before deleting.

All versions