# Automate Browser Workflows With Agent Browser

Manual browser tasks slow research, testing, and evidence capture. This skill lets AI agents navigate pages, use element refs, capture screenshots, and run controlled browser workflows.

## Install

```bash
npx skillstore add inferen-sh/agent-browser
```

## Metadata

- Status: approved
- Slug: inferen-sh-agent-browser
- Skillstore revision: r1
- Version status: missing
- Tree hash: 0ccd4e2786e40514eceb9c31f99a035d2dd4a50688b7af8afa08de98c23b8422
- Author: inferen-sh
- GitHub username: inferen-sh
- License: MIT
- Repository: https://github.com/inferen-sh/skills/tree/main/tools/utilities/agent-browser/
- Ref: a06681402992ceae98ba04d54cfd4ab004862696
- 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/inferen-sh-agent-browser
- Manifest: https://skillstore.pages.dev/api/skills/inferen-sh-agent-browser/manifest

## Capabilities

- Open browser sessions, navigate to URLs, and return compact element refs.
- Click, type, fill forms, select options, drag elements, and upload files.
- Capture viewport or full-page screenshots and optional session video.
- Execute JavaScript in the page to extract text, links, or structured data.
- Route browser sessions through HTTP, HTTPS, or SOCKS proxies.
- Use reusable shell templates for forms, login flows, and content capture.

## Use Cases

- Capture Test Evidence: Record browser steps, screenshots, and page state for repeatable web testing evidence.
- Collect Public Page Data: Open public pages, extract visible text and links, and save screenshots for review.
- Document Product Workflows: Generate screenshots or videos that show approved product workflows without manual browser work.

## Prompt Templates

### Open A Page

```
Use agent-browser to open [URL] and list the interactive elements. Do not enter credentials or click anything.
```

### Prepare Form Steps

```
Use agent-browser to inspect [form URL], identify refs for [fields], and prepare fill steps. Stop before submission.
```

### Capture Page Evidence

```
Use agent-browser to capture a full-page screenshot and visible text from [URL]. Store outputs in [directory].
```

### Audit An Authenticated Flow

```
Use agent-browser with an approved test account to sign in to [app] and perform [steps]. Record video only when no sensitive data appears.
```

## Limitations

- Requires the inference.sh CLI, an account, and an authenticated infsh session.
- Element refs can become stale after navigation or dynamic page updates.
- Screenshots, video, and page text can expose sensitive data if misused.
- Long recordings and complex pages can create large output files.

## Best Practices

- Use public pages or approved test accounts unless you have explicit authorization.
- Re-snapshot after navigation, form submission, or dynamic content changes.
- Disable video and screenshots when credentials or personal data may appear.

## Anti Patterns

- Do not use proxy rotation to evade rate limits or access controls.
- Do not upload local files to pages that you do not trust.
- Do not extract cookies, tokens, or private page data without approval.

## Security Audit

- Audited at: 2026-07-06T14:53:20.858\+00:00
- Summary: Most Markdown backtick, relative-link, device redirection, and hardcoded URL alerts are false positives from examples. Confirmed risks are the skill's intended external browser automation: Bash-driven infsh execution, screenshot and video capture, SOCKS proxy support, credential workflows, arbitrary JavaScript, and file upload. The skill is useful, but it needs explicit consent and stronger guardrails for sensitive browsing.

## Stats

- Views: 214
- Downloads: 10
- Favorites: 0
- Popularity score: 0