Test and Break

Systematically test a deployed application by going through user flows, trying edge cases, and attempting to break things. Outputs structured bug reports that can be converted to user stories for autonomous fixing.

---

Prerequisites

• agent-browser installed (npm install -g agent-browser && agent-browser install)
• App deployed to a URL (staging/preview)
• Basic understanding of what the app should do (read the PRD)

---

The Job

1. Read the PRD to understand what the app should do
2. Open the deployed app in agent-browser
3. Go through each major user flow
4. Try to break things at each step
5. Document all bugs and issues found
6. Output structured bug reports

---

Testing Process

Step 1: Understand the App

Read tasks/prd.md and tasks/architecture.md to understand:

• What user flows exist
• What the app should do
• What the expected behavior is

Step 2: Open the App

agent-browser open [DEPLOYMENT_URL]
agent-browser snapshot -i

Step 3: Test Each User Flow

For each major feature/flow in the PRD:

A. Happy Path Testing

1. Go through the flow as a normal user would
2. Verify each step works as expected
3. Check that success states appear correctly

B. Edge Case Testing

Try these at each input/interaction point:

Input Edge Cases:

• Empty inputs (submit with nothing)
• Very long text (500+ characters)
• Special characters (<script>alert('xss')</script>, '; DROP TABLE users;--)
• Unicode/emojis (🎉, 中文, العربية)
• Negative numbers where positive expected
• Zero where non-zero expected
• Future dates, past dates, invalid dates
• Invalid email formats
• Spaces only
• Leading/trailing whitespace

Interaction Edge Cases:

• Double-click buttons rapidly
• Click back button during operations
• Refresh page mid-flow
• Open same page in multiple tabs
• Submit form twice quickly

State Edge Cases:

• Log out mid-operation (if auth exists)
• Let session expire
• Navigate directly to URLs that require prior steps
• Use browser back/forward buttons

Visual/UX Issues:

• Check mobile responsiveness (resize browser)
• Look for overlapping elements
• Check loading states exist
• Verify error messages are helpful
• Look for console errors

Step 4: Document Each Bug

For each issue found, document:

## BUG-XXX: [Short descriptive title]

**Severity:** Critical | High | Medium | Low
**Type:** Functional | UI/UX | Security | Performance | Accessibility

**Steps to Reproduce:**
1. Go to [URL]
2. Do [action]
3. Enter [input]
4. Click [button]

**Expected Behavior:**
[What should happen]

**Actual Behavior:**
[What actually happens]

**Screenshot:** [if applicable]

**Console Errors:** [if any]

**Notes:** [any additional context]

---

Severity Guidelines

Severity	Definition	Examples
Critical	App broken, data loss, security issue	Crash, XSS vulnerability, data not saving
High	Major feature broken, bad UX	Can't complete main flow, confusing errors
Medium	Feature works but has issues	Minor validation missing, UI glitches
Low	Polish/minor issues	Typos, slight misalignment, minor UX

---

Output Format

Save bug report to tasks/bug-report-[date].md:

# Bug Report: [App Name]
**Tested:** [Date]
**URL:** [Deployment URL]
**Tester:** Claude (Automated)

## Summary
- Total bugs found: X
- Critical: X
- High: X
- Medium: X
- Low: X

## Critical Bugs
[List critical bugs first]

## High Priority Bugs
[List high bugs]

## Medium Priority Bugs
[List medium bugs]

## Low Priority Bugs
[List low bugs]

## Positive Findings
[List things that worked well - important for context]

## Recommendations
[Overall suggestions for improvement]

---

Converting Bugs to User Stories

After generating the bug report, convert each bug to a user story format:

{
  "id": "BUG-001",
  "title": "Fix: [Bug title]",
  "description": "As a user, I expect [expected behavior] but currently [actual behavior].",
  "acceptanceCriteria": [
    "Specific fix criterion 1",
    "Specific fix criterion 2",
    "Regression test: [original bug steps] no longer reproduces",
    "Typecheck passes"
  ],
  "priority": 1,
  "passes": false,
  "notes": "Original bug: [reference]"
}

Priority mapping:

• Critical bugs → priority 1-2
• High bugs → priority 3-5
• Medium bugs → priority 6-10
• Low bugs → priority 11+

---

Integration with Ralph

After generating bug stories, they can be:

1. Added to existing prd.json - Append bug fixes to current project
2. Create new prd.json - Start a bug-fix-only Ralph run

To add to existing prd.json:

# Read current max priority
MAX_PRIORITY=$(cat prd.json | jq '[.userStories[].priority] | max')

# Add bug stories starting after max priority
# (Claude should do this programmatically)

---

Example Testing Session

# 1. Open the app
agent-browser open https://my-app-staging.vercel.app

# 2. Take initial snapshot
agent-browser snapshot -i

# 3. Test login flow
agent-browser fill @e1 "test@example.com"
agent-browser fill @e2 "password123"
agent-browser click @e3
agent-browser wait --load networkidle
agent-browser snapshot -i

# 4. Try to break it
agent-browser fill @e1 ""  # empty email
agent-browser click @e3    # submit anyway
agent-browser snapshot -i  # check error handling

# 5. Try XSS
agent-browser fill @e1 "<script>alert('xss')</script>"
agent-browser snapshot -i

# Continue testing other flows...

---

Checklist

Before finishing testing:

• Tested all user flows from PRD
• Tried empty inputs on all forms
• Tried special characters/XSS on all inputs
• Checked mobile responsiveness
• Looked for console errors
• Verified error messages are helpful
• Documented all bugs with reproduction steps
• Assigned severity to each bug
• Saved bug report to tasks/bug-report-[date].md