visual-evidence

visual-evidence

Capture browser screenshots and GIF recordings using Playwright headless Chromium. Upload to cloud storage for embedding in PRs and reports.

When to Use

• After implementing a bug fix (before/after comparison)
• After implementing a feature (demo key flows)
• During PR review to verify visual changes
• Interaction/timing bugs (double-click, loading states, race conditions) → use GIF recording

When NOT to Use

• Backend-only changes, config/infra, test-only, CLI tools, APIs without UI
• Skip if Playwright install fails — visual evidence is best-effort, never blocks the pipeline

Prerequisites

Install Playwright and Chromium in the remote environment:

cd /tmp && npm install playwright 2>&1 | tail -3 && \
  npx playwright install-deps chromium 2>&1 | tail -3 && \
  npx playwright install chromium 2>&1 | tail -3 && \
  echo PW_READY

For GIF recording, also install ffmpeg:

sudo apt-get update -qq && sudo apt-get install -y -qq ffmpeg 2>&1 | tail -1 && echo FFMPEG_READY

Workflow

1. Screenshot Capture

Create /tmp/screenshot.mjs:

import { chromium } from '/tmp/node_modules/playwright/index.mjs';
const [url, output] = process.argv.slice(2);
const browser = await chromium.launch({ headless: true });
const page = await browser.newPage({ viewport: { width: 1280, height: 720 } });
console.log(`Navigating to ${url}...`);
await page.goto(url, { waitUntil: 'networkidle', timeout: 30000 });
const title = await page.title();
console.log(`Page title: ${title}`);
await page.screenshot({ path: output, fullPage: true });
console.log(`Screenshot saved: ${output}`);
await browser.close();

Usage:

node /tmp/screenshot.mjs http://localhost:3000 /tmp/evidence-home.png
node /tmp/screenshot.mjs http://localhost:3000/admin /tmp/evidence-admin.png

2. GIF Recording (Interaction Evidence)

Create /tmp/record.mjs:

import { chromium } from '/tmp/node_modules/playwright/index.mjs';
const [url, output, ...actions] = process.argv.slice(2);
const browser = await chromium.launch({ headless: true });
const context = await browser.newContext({
  viewport: { width: 1280, height: 720 },
  recordVideo: { dir: '/tmp/videos', size: { width: 1280, height: 720 } }
});
const page = await context.newPage();
console.log(`Navigating to ${url}...`);
await page.goto(url, { waitUntil: 'networkidle', timeout: 30000 });

// Execute actions passed as JSON array
if (actions[0]) {
  const steps = JSON.parse(actions[0]);
  for (const step of steps) {
    if (step.action === 'click') {
      console.log(`Clicking ${step.selector}...`);
      await page.click(step.selector);
    } else if (step.action === 'dblclick') {
      console.log(`Double-clicking ${step.selector}...`);
      await page.click(step.selector);
      await page.waitForTimeout(200);
      await page.click(step.selector);
    } else if (step.action === 'wait') {
      console.log(`Waiting ${step.ms}ms...`);
      await page.waitForTimeout(step.ms);
    } else if (step.action === 'fill') {
      console.log(`Filling ${step.selector}...`);
      await page.fill(step.selector, step.value);
    } else if (step.action === 'screenshot') {
      console.log(`Snapshot: ${step.label}`);
      await page.screenshot({ path: `/tmp/evidence-${step.label}.png` });
    }
  }
}

await page.waitForTimeout(500);
await page.close();
await context.close();
await browser.close();

// Convert to GIF
const { execSync } = await import('child_process');
const webm = execSync('ls -t /tmp/videos/*.webm | head -1').toString().trim();
console.log(`Video: ${webm}`);
execSync(`ffmpeg -y -i "${webm}" -vf "fps=10,scale=1280:-1:flags=lanczos,split[s0][s1];[s0]palettegen[p];[s1][p]paletteuse" -loop 0 "${output}" 2>&1 | tail -3`);
console.log(`GIF saved: ${output}`);

Usage:

node /tmp/record.mjs http://localhost:3000/form /tmp/evidence-before.gif \
  '[{"action":"click","selector":".submit-btn"},{"action":"wait","ms":2000}]'

3. Upload to Cloud Storage

Use the evidence-upload skill to upload screenshots and GIFs to the pylot assets backend. $PYLOT_GATEWAY_URL and $PYLOT_DISPATCH_TOKEN are already in every operator/worker environment — no static AWS keys needed.

Run /evidence-upload for each captured file and capture the returned public_url for embedding in Step 4.

4. PR Embedding

Before/After (bug fixes):

## Visual Evidence

<details>
<summary>Before / After</summary>

**Before (bug present on default branch):**
![before](<url>/before.png)

**After (fix applied):**
![after](<url>/after.png)

</details>

Interaction demo (GIF):

## Visual Evidence

<details>
<summary>Interaction demo</summary>

![demo](<url>/demo.gif)
*Caption: what this demonstrates*

</details>

Decision Table: Screenshot vs GIF

Scenario	Format
Visual layout change	Screenshot (PNG)
New page / feature	Screenshot (PNG)
Button disable on click	GIF
Loading spinner / skeleton	GIF
Double-click prevention	GIF
Form validation feedback	GIF
Transition / animation	GIF
Error → retry flow	GIF

Error Handling

Visual evidence is best-effort. Never block the pipeline.

• Playwright install fails → log, skip evidence, proceed
• Dev server won't start → log, skip evidence, proceed
• Screenshot fails (404, timeout) → log, skip evidence, proceed
• Upload fails (no creds, permission denied) → log, skip evidence, proceed
• Hard timeout: 120 seconds for the entire evidence phase. If exceeded, kill and proceed.

Critical Rules

• Never block the pipeline — evidence is best-effort
• 120 second hard timeout — kill and move on
• Always verify upload before embedding URLs in PRs
• Screenshots in /tmp/ are ephemeral — they disappear when the environment stops