autobrowse

Iteratively run browser traces and update navigation strategies until tasks pass.

Updated Aug 23, 2026
One-click install
npx skills add https://github.com/Z5Jonathan-maker/ai-coding-os --skill autobrowse
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: autobrowse
Source: https://github.com/Z5Jonathan-maker/ai-coding-os/tree/main/claude/skills/autobrowse
Command: npx skills add https://github.com/Z5Jonathan-maker/ai-coding-os --skill autobrowse

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires dotenv, and includes scripts (resource) and references (resource) components.

What problem does it solve?

AutoBrowse eliminates the trial-and-error burden of building dependable browser instructions by running an automated browse loop that iteratively improves a site-specific strategy until it consistently passes.

Core Features & Use Cases

  • Self-improving skill building: Runs a browsing “inner agent,” reviews the resulting trace, and updates strategy.md over multiple iterations to produce a shippable skill.md playbook.
  • Local and remote execution modes: Supports local Chrome automation and a remote (Browserbase) mode for bot-protected sites.
  • Durable, inspectable artifacts: Writes traces (including summary and full trace logs) so improvements are grounded in evidence rather than guesswork.
  • Use case: Graduating a navigation skill for a specific website flow (e.g., logging in, filling a form, extracting a confirmation number) so future agents can skip costly exploration.

Quick Start

Run autobrowse in your project directory to generate or refine a graduated skill for your target site task using multiple evaluate-improve iterations.

Frequently Asked Questions about autobrowse

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

FAQPage Schema
How do I automate browser navigation for recurring website tasks without trial and error?

You can automate browser navigation by running an iterative browsing loop that evaluates execution traces and updates a site-specific strategy until the task consistently passes. This produces a durable, cached navigation playbook for recurring flows like form submissions or account logins.

Can I use browser automation to extract confirmation details from bot-protected sites?

Yes, browser automation supports remote execution via Browserbase credentials specifically for bot-protected sites. This remote mode allows the inner browsing agent to navigate and extract confirmation details without being blocked by site protections.

What is trace-driven iteration for building web navigation skills?

Trace-driven iteration is a process where an automated browsing attempt runs, generates a trace log, and the system updates a strategy file based on that evidence. It repeats this evaluate-improve cycle until a reliable site-specific navigation skill is achieved.

Do I need an Anthropic API key to build reliable website navigation instructions?

Yes, an Anthropic API key is required to power the inner agent that executes the browsing attempts and strategy updates. You also need a root skill entry, a target task file, and a strategy file in your workspace.

How do I create deterministic cached navigation for web form submission workflows?

You create deterministic cached navigation by iteratively running an automated browsing agent against a form submission flow, reviewing the resulting trace logs, and refining the strategy until the workflow passes consistently. This graduates into a shippable playbook file for future agents.