# Test Web Apps With Repro Evidence

Web app issues are hard to fix when reports lack proof. This skill explores the app in a browser and creates reproducible evidence for each finding.

## Install

```bash
npx skillstore add vercel-labs/dogfood
```

## Metadata

- Status: approved
- Slug: vercel-labs-dogfood
- Skillstore revision: r1
- Version status: missing
- Tree hash: 086fdf24d0b190eaa911742e376e3210f7fce0d21cd7fe4391374f28b37ab4a3
- Author: vercel-labs
- GitHub username: vercel-labs
- License: MIT
- Repository: https://github.com/vercel-labs/agent-browser/tree/main/skills/dogfood/
- Ref: b959ebfd1043a07c0c4378ab94ca4342cb0259f5
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: confirmation\_required
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands, network, filesystem
- Quality score: 38
- Quality tier: warning
- Public page: https://skillstore.pages.dev/skills/vercel-labs-dogfood
- Manifest: https://skillstore.pages.dev/api/skills/vercel-labs-dogfood/manifest

## Capabilities

- Opens a target web app with agent-browser and creates a named test session.
- Captures annotated screenshots, browser snapshots, console output, and browser errors.
- Records repro videos for interactive issues and step-by-step screenshots.
- Uses a severity taxonomy for visual, functional, UX, content, performance, console, and accessibility issues.
- Writes a structured Markdown report with issue counts, repro steps, and evidence paths.
- Saves browser state after login when authenticated testing is required.

## Use Cases

- Pre-release exploratory QA: Run a structured pass across launch-critical flows and produce evidence that engineers can reproduce quickly.
- Regression bug hunting: Check important pages after a release candidate changes navigation, forms, or stateful workflows.
- UX quality review: Find confusing flows, missing feedback, visual defects, and accessibility issues from a user viewpoint.

## Prompt Templates

### Quick app check

```
Dogfood http://localhost:3000 and create a report in ./dogfood-output.
```

### Focus on one workflow

```
Dogfood https://staging.example.com with focus on onboarding. Capture evidence for every issue and summarize severity counts.
```

### Authenticated product review

```
Dogfood https://app.example.com. Sign in with the test account I provide, then test account settings and team management.
```

### Release readiness audit

```
Dogfood https://preview.example.com. Prioritize billing, invitations, and project creation. Record videos for interactive failures and use clear repro steps.
```

## Limitations

- Requires the agent-browser CLI and browser automation access.
- Tests observed browser behavior, not the target app source code.
- Authenticated testing depends on user-provided credentials or OTP codes.
- Evidence files can contain sensitive page data and need careful handling.

## Best Practices

- Use test accounts and seed data that can be safely changed.
- Review screenshots and videos for sensitive information before sharing them.
- Keep the scope clear so exploration focuses on important workflows.

## Anti Patterns

- Running the skill against production with real customer data without approval.
- Treating the generated report as a substitute for automated regression tests.
- Sharing the output directory before removing secrets, tokens, or private user data.

## Security Audit

- Audited at: 2026-07-09T17:08:12.647\+00:00
- Summary: Most static findings are Markdown backtick false positives in the taxonomy and instructional prose. Confirmed risks remain where the skill runs agent-browser and shell commands with user-supplied values, captures screenshots or videos, and stores authenticated browser state locally. No prompt injection text was found in the reviewed files.

## Stats

- Views: 160
- Downloads: 17
- Favorites: 0
- Popularity score: 0