manually-validate-fix

name	manually-validate-fix
description	Validate a bug-fix PR by reproducing the bug in production and confirming the fix locally, using browser automation.
disable-model-invocation	true
argument-hint	<pr-number>

Fix Manual Validation

Validate a bug-fix PR by testing like a human — reproduce the bug in production (broken), confirm the fix works locally (working). Uses a single Selenium WebDriver script run against both environments.

PR → UNDERSTAND BUG → PLAN REPRODUCTION → WRITE SCRIPT → RUN vs PRODUCTION (bug)
                                                        → RUN vs LOCAL (fix)
                                                        → VERDICT → label PR

Environments

Environment	URL	Runs	Expected
Production	https://basecamp.github.io/lexxy/try-it.html	Latest released version (CDN)	Bug reproduces
Local	http://lexxy.localhost:3000	Current branch with the fix	Bug does NOT reproduce

Production page

The try-it page loads Lexxy from jsDelivr CDN. After JS executes:

• <lexxy-editor> with toolbar, emoji prompt on : trigger
• data-direct-upload-url and data-blob-url-template are set but uploads go to a service worker (no real backend)
• Editor behavior is fully functional for text editing, formatting, paste, tables, etc.

Local sandbox

http://lexxy.localhost:3000 runs the Rails dummy app:

• Root (/sandbox) — editor with @ mention and : emoji prompts, template switcher
• Posts CRUD (/posts) — full Action Text form, all features
• Query params on /posts/new control features: ?attachments_disabled=true, ?toolbar_disabled=true, etc.

Procedure

1. Understand the Bug

gh pr view $ARGUMENTS --json title,body,files,commits,headRefName
gh pr diff $ARGUMENTS

Extract:

• What bug it fixes — the problem statement
• How a human would trigger the bug — step-by-step
• What the fix changes — which files/behavior
• Which editor features are involved

If the PR references an issue, fetch that too.

2. Plan the Reproduction

Write a concrete step-by-step plan:

1. What page to visit (both environments share the same editor, pick the right page)
2. What content to type or paste
3. What buttons to click or keys to press
4. What the broken behavior looks like
5. What the fixed behavior looks like

The plan must map directly to WebDriver actions.

3. Ensure Selenium is Available

ls /tmp/node_modules/selenium-webdriver >/dev/null 2>&1 || {
  cd /tmp && npm install selenium-webdriver
}
which chromedriver >/dev/null 2>&1 || echo "ERROR: Install chromedriver"

4. Write the Validation Script

Write a single .mjs script in /tmp/ that accepts a URL argument and runs the reproduction steps. The same script runs against both environments — only the URL changes.

Script skeleton:

import webdriver from '/tmp/node_modules/selenium-webdriver/index.js'
import chrome from '/tmp/node_modules/selenium-webdriver/chrome.js'
import { Key } from '/tmp/node_modules/selenium-webdriver/lib/input.js'
import fs from 'fs'

const BASE_URL = process.argv[2]
if (!BASE_URL) {
  console.error('Usage: node script.mjs <url>')
  process.exit(1)
}

const ENV = BASE_URL.includes('github.io') ? 'production' : 'local'

async function sleep(ms) { return new Promise(r => setTimeout(r, ms)) }

async function screenshot(driver, name) {
  const data = await driver.takeScreenshot()
  fs.writeFileSync(`/tmp/validation-${ENV}-.png`, .(data, ))
  .()
}

 options =  chrome.()
options.(, , )
 driver =   webdriver.()
  .().(options).()

 {
  
   driver.()
   driver.(
    webdriver..(webdriver..()),
    
  )
   content =  driver.(webdriver..())
   content.()
   ()

  
  
  
  
  
  
  
  

  
   (driver, )

   html =  driver.(
    
  )
  .(, html)

  
  
  
  
  
  
  
  

}  {
   driver.()
}

Key rules for the script:

• All interactions go through real browser events (sendKeys, click). No programmatic shortcuts.
• executeScript is for reading state only — never for producing the action under test.
• Screenshots at key steps, named by environment and step.
• Clear console output showing what happened in each environment.

5. Ensure Local Server is Running

curl -s -o /dev/null -w "%{http_code}" http://lexxy.localhost:3000/ 2>/dev/null | grep -q 200 || {
  echo "Starting local server..."
  cd ~/Work/basecamp/lexxy && bin/dev &
  for i in $(seq 1 30); do
    curl -s -o /dev/null -w "%{http_code}" http://lexxy.localhost:3000/ 2>/dev/null | grep -q 200 && break
    sleep 1
  done
}

Verify the right branch is checked out:

cd ~/Work/basecamp/lexxy && git branch --show-current
# Must match the PR's head branch

6. Run Against Both Environments

# Production — expect the bug
node /tmp/validate-fix-<pr>.mjs "https://basecamp.github.io/lexxy/try-it.html"

# Local — expect the fix
node /tmp/validate-fix-<pr>.mjs "http://lexxy.localhost:3000"

Review screenshots and console output from both runs.

7. Render Verdict

Production	Local	Verdict
Bug reproduces	Bug does NOT reproduce	Validated
Bug reproduces	Bug STILL reproduces	Not Fixed
Bug does NOT reproduce	N/A	Cannot Validate

Verdict format:

**Validation Verdict: [Validated / Not Fixed / Cannot Validate]**

**PR:** #<number> — <title>
**Bug:** <one-line description>

**Production** (https://basecamp.github.io/lexxy/try-it.html):
- <what happened>
- Evidence: <screenshot paths>

**Local** (http://lexxy.localhost:3000):
- <what happened>
- Evidence: <screenshot paths>

**Steps executed:**
1. <step> — <result>
2. ...

8. Label and Comment on the PR

If Validated:

gh pr edit <number> --add-label "fix-validated" -R basecamp/lexxy
gh pr comment <number> --body "$(cat <<'EOF'
## Manual Validation: Validated

<Brief description of what was tested — written for a human reviewer. Describe the steps you performed and what you observed, e.g. "Typed a bulleted list, selected all items, applied heading format from the toolbar. Confirmed the list converted to a heading without leaving orphaned list markers.">
EOF
)"

If Not Fixed — comment on the PR:

gh pr comment <number> --body "$(cat <<'EOF'
## Manual Validation: Not Fixed

<Same human-readable description of the steps tested and what was observed, explaining why validation failed.>
EOF
)"

Important Notes

• Test like a human. Every interaction goes through real browser events.
• One script, two URLs. The reproduction is identical in both environments.
• Evidence is mandatory. Screenshots + editor state for both runs.
• Don't touch the code. This skill validates only — no source modifications.
• Production may be slow. CDN version loads from jsDelivr. Allow extra wait time.