Dogfood

Dogfood

Systematically explore a web application to find bugs, UX issues, and quality problems. Uses browser automation to test across viewports and produce a structured bug report with reproduction steps and screenshots.

When to Activate

• User says "dogfood", "QA this", "find bugs", "test this app"
• User provides a URL and asks for quality feedback
• User says "exploratory test", "bug hunt"

Input

Required:

• Target URL: The web app to test

Optional:

• Scope: Focus area (e.g., "billing page", "onboarding flow"). Default: full app.
• Authentication: Credentials if login is required.
• Viewports: Default: desktop (1280x800), tablet (768x1024), mobile (375x812).

Execution

Step 1: Initialize

Create output directory for screenshots and report:

mkdir -p {output_dir}/screenshots {output_dir}/videos

Open the target URL in a browser session. Take an initial screenshot.

Step 2: Authenticate (if needed)

If the app requires login, fill credentials and save auth state. For OTP or email codes, ask the user and wait.

Step 3: Orient

Take an annotated screenshot of the landing page. Identify main navigation elements. Map out sections to visit.

Step 4: Explore Systematically

Work through the app section by section:

1. Navigation: Visit each top-level section. Check all links work.
2. Forms: Fill and submit forms. Test empty submissions, invalid input, boundary values.
3. Interactive elements: Click buttons, open dropdowns/modals, toggle switches.
4. Error states: Trigger error conditions. Check error messages are helpful.
5. Empty states: Check what happens with no data.
6. Responsive: Resize to tablet and mobile viewports. Check layout.
7. Console: Check browser console for JS errors and failed network requests.
8. Accessibility basics: Check focus order, alt text on images, color contrast.

At each page, take a screenshot and check the console.

Step 5: Document Issues

Document each issue as you find it. Do not batch for later.

For interactive/behavioral issues: Record a video, take step-by-step screenshots, write numbered repro steps.

For static/visual issues (typos, misalignment, broken layout): A single annotated screenshot is sufficient.

Each issue gets:

• Unique ID (ISSUE-001, ISSUE-002, ...)
• Severity: Critical / High / Medium / Low
• Category: Functional / Visual / UX / Performance / Accessibility / Console Error
• Reproduction steps
• Screenshot(s) or video reference
• Expected vs actual behavior

Step 6: Wrap Up

Aim for 5-10 well-documented issues. Depth of evidence beats quantity.

Update summary counts. Close the browser session.

Output Format

# Dogfood Report: {App Name}
**URL:** {url} | **Date:** {date} | **Tester:** AI (pm-pilot)

## Summary
- Critical: {count}
- High: {count}
- Medium: {count}
- Low: {count}
- **Total:** {count}

## Viewports Tested
- Desktop (1280x800): Yes/No
- Tablet (768x1024): Yes/No
- Mobile (375x812): Yes/No

## Issues

### ISSUE-001: {Short title}
- **Severity:** {Critical/High/Medium/Low}
- **Category:** {Functional/Visual/UX/Performance/Accessibility}
- **Page:** {URL or page name}
- **Viewport:** {Desktop/Tablet/Mobile}

**Steps to Reproduce:**
1. {Step} [screenshot: issue-001-step-1.png]
2. {Step} [screenshot: issue-001-step-2.png]

**Expected:** {what should happen}
**Actual:** {what actually happens}
**Screenshot:** {path}
**Video:** {path or N/A}

---
(Repeat for each issue)

## Pages Tested
- {Page 1}: {status}
- {Page 2}: {status}

## Console Errors
- {Error message} on {page} ({count} occurrences)

Rules

• Repro is everything: Every issue needs proof. Interactive bugs need step-by-step screenshots. Static bugs need one annotated screenshot.
• Document as you go: Write each issue immediately. Do not batch.
• Test like a user: Try realistic workflows. Enter realistic data.
• Check the console: Many issues are invisible in the UI but visible in JS errors.
• No source code: Test as a user, not a code auditor. All findings come from browser observation.
• Be efficient: Spend more time on core features, less on peripheral pages.
• Never delete output: Do not remove screenshots or report files mid-session.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: mshadmanrahman
• Source: mshadmanrahman/pm-pilot
• License: MIT
• Homepage: https://github.com/mshadmanrahman/pm-pilot#readme

Install and usage instructions live in the source repository linked above.