AgentSkillsCN

test-browsy-flow

通过browsy CLI测试URL，报告页面检测的准确率。

SKILL.md
--- frontmatter
name: test-browsy-flow
description: Test a URL with the browsy CLI and report page detection accuracy

Test Browsy Flow

Test browsy's page detection and interaction capabilities against a specific URL. Useful for validating that browsy correctly parses a site before deploying agents against it.

Usage

code
/test-browsy-flow https://example.com/login

What it does

Runs the browsy CLI against a target URL and reports on detection quality:

  • •Page parsing — fetches the URL and converts it to browsy's Spatial DOM, reporting element counts and types
  • •Page classification — checks whether browsy correctly identifies the page type (Login, Search, Content, Form, Consent, Verification, Error)
  • •Suggested actions — validates that browsy suggests the right actions (e.g., Login for a sign-in page, Search for a search engine)
  • •Element inventory — counts interactive elements (links, buttons, inputs, selects, checkboxes) and checks for missing or misidentified elements
  • •Hidden content — verifies that dropdown menus, modals, accordion panels, and tab content are exposed with hidden: true
  • •Interaction testing — optionally tests clicking, typing, and form submission to verify browsy handles the page's interaction patterns

Workflow

  1. •Run browsy browse <url> to fetch and parse the page
  2. •Run browsy page-info to get detection results
  3. •Analyze the Spatial DOM output:
    • •Count elements by type (links, buttons, inputs, text)
    • •Check for landmark markers (nav, header, footer, main, form)
    • •Verify hidden elements are exposed
  4. •Optionally test interactions:
    • •Click a link or button
    • •Fill a form field
    • •Submit a search query
  5. •Report findings with pass/fail assessment

Example output

code
=== Browsy Flow Test: https://example.com/login ===

Page title: Sign In - Example App
Page type: Login [CORRECT]
Suggested actions: Login [CORRECT]

Elements detected: 12
  Links: 3 (Home, Forgot password, Sign up)
  Buttons: 1 (Sign in)
  Inputs: 2 (email [text], password [password])
  Text: 4 (heading, labels, helper text)
  Hidden: 2 (error alert, password tooltip)

Landmarks:
  [PASS] <form> detected as form landmark
  [PASS] <nav> detected as navigation landmark
  [PASS] <main> detected as main landmark

Interaction test:
  [PASS] browsy_type_text on email field — value persisted
  [PASS] browsy_type_text on password field — value persisted
  [PASS] browsy_login compound action — form submitted

Result: 9/9 checks passed. Page fully compatible with browsy.

Example with issues

code
=== Browsy Flow Test: https://spa-heavy.example.com ===

Page title: App
Page type: Content [EXPECTED: Dashboard]
Suggested actions: (none) [EXPECTED: Login or Navigation]

Elements detected: 3
  Links: 1 (logo link)
  Text: 2 (loading spinner text, noscript message)

WARNING: Only 3 elements detected. This page likely requires
JavaScript rendering. Consider using Playwright for this site.

Result: 1/5 checks passed. Page NOT compatible with browsy.
Recommendation: Keep Playwright for this JS-heavy SPA.

Requirements

  • •browsy CLI must be installed (browsy in PATH or set BROWSY_BIN env var)