Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.browseract.com/llms.txt

Use this file to discover all available pages before exploring further.

Skill Forge is an extension built on Browser-act CLI. It lets AI agents explore a website once, then generate a reusable skill for repeated extraction or automation.

​
The problem

Without a reusable skill, an agent has to rediscover the same website every time:
  • paths change between runs
  • failure modes differ
  • scaling is unreliable
  • token usage grows with every call
Skill Forge separates exploration from reuse.
[!NOTE] Explore once. Generate a deployable skill package. Reuse the same stable path for hundreds or thousands of inputs.

​
Installation

Skill Forge is separate from the Browser-act entry Skill. Install the entry Skill and CLI first. Ask your agent: 
Install browser-act-skill-forge.
Skill URL: https://github.com/browser-act/skills/tree/main/browser-act-skill-forge
After installation, verify that it is available.
Requirements:

​
Four-step pipeline

StepWhat happens
01. DescribeTell the agent what data or action you need
02. ExploreThe agent looks for an API-first path, then falls back to DOM automation if needed
03. GenerateThe agent creates a parameterized skill package
04. Self-testThe generated skill is tested and repaired until it works

​
01. Describe the task

Use natural language: 
Extract job title, company, salary, and URL from LinkedIn jobs.
I will later run this for 300 keywords.

​
02. Explore API-first, DOM fallback

Skill Forge prefers stable paths:
PriorityStrategyStability
1API-first through network captureMore stable when backend contracts stay the same
2DOM fallback through browser automationWorks when no useful API path is available

​
03. Generate a parameterized skill

Business variables become CLI parameters instead of hard-coded values: 
job-search --keyword "designer" --location "Berlin" --limit 100
Typical output:
  • SKILL.md
  • executable script
  • usage instructions
  • self-test evidence

​
04. Self-test and repair

The generated skill should be tested end to end. When the target site changes, the agent can re-explore and update the skill instead of rediscovering the workflow on every run.

​
Business value

For individual developers

Pay the exploration cost once, reuse the generated path for many inputs, and re-explore only when the site changes.

For teams

Move repeatable scraping work from one-off scripts toward generated, tested, reusable capabilities.
Traditional model: 
request -> engineer writes scraper -> test -> deploy -> maintain
Skill Forge model: 
request -> agent explores -> skill generated -> self-test -> deploy
The engineering work shifts from writing every scraper by hand to maintaining a system that lets agents generate reusable capabilities.

​
Use cases

Batch data collection

Run one pattern across many keywords, regions, or categories.

Cross-site price comparison

Generate skills for multiple sites and normalize output.

Periodic monitoring

Track price, inventory, or listing changes.

Internal processes

Wrap repeated browser operations into reusable skills.

​
Entry Skill vs. Skill Forge

AreaBrowser-act entry SkillSkill Forge
RoleGives the agent Browser-act CLI capabilityGenerates reusable task-specific skills
RequiredYesOptional
UsageAgent operates the browser in real timeAgent explores once and reuses generated skill
Token costPaid during each live explorationConcentrated during skill generation
Browser-act gives agents the ability to act. Skill Forge helps agents turn a successful action path into a reusable skill.

​
Learn more

Installation

Install the base CLI and entry Skill before using Skill Forge.

Anti-detection & Blocking

Use blocking tools during exploration when a site challenges automation.

Designed for Agents

Use network capture and semantic descriptions effectively.