# Automate browser workflows with guided sessions

Manual browser tasks are slow and difficult to repeat. This skill provides session-based commands for consistent navigation, interaction, capture, and verification.

## Install

```bash
npx skillstore add 101-skills/agent-browser
```

## Metadata

- Status: approved
- Slug: 101-skills-agent-browser
- Skillstore revision: r4
- Version status: missing
- Tree hash: 9217592db0c540dbb41db26b2a752f111a2b3008d156a9cfe4edc7f56b71639d
- Author: 101-skills
- GitHub username: 101-skills
- License: MIT
- Repository: https://github.com/101-skills/skills/tree/main/tools/utilities/agent-browser/
- Ref: 7014d51800335b375ec9e0068a94267aaef4015b
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: allowed
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands, network, filesystem
- Quality score: 74
- Public page: https://skillstore.pages.dev/skills/101-skills-agent-browser
- Manifest: https://skillstore.pages.dev/api/skills/101-skills-agent-browser/manifest

## Capabilities

- Open browser sessions and navigate to specified web pages.
- Inspect page state through snapshots and element references.
- Click, fill, type, select, drag, upload, and scroll through referenced elements.
- Capture screenshots and optional session video recordings.
- Run documented page JavaScript actions within an active browser session.
- Configure documented authenticated, proxy, form, and session workflows.

## Use Cases

- Test a web form: Open a staging page, complete a form, and verify the resulting page state.
- Collect public page details: Navigate approved public pages and capture structured page information for research.
- Document a support workflow: Record a repeatable browser session that demonstrates an approved customer support process.

## Prompt Templates

### Beginner: Open a page

```
Open https://example.com in a new browser session and summarize the page title and available interactive elements.
```

### Intermediate: Check a form

```
Open the approved staging URL, inspect the contact form, fill the supplied test values, and stop before submission.
```

### Advanced: Verify a workflow

```
Use a new session to follow this approved test path. Take a snapshot after each page change and report where the expected state differs.
```

### Advanced: Capture evidence

```
Record an authorized browser test session, capture screenshots at the listed checkpoints, and return a concise result summary.
```

## Limitations

- Requires the inference.sh belt CLI and a working browser service.
- Cannot bypass login controls, CAPTCHAs, access rules, or website terms.
- Results depend on page structure, permissions, network access, and browser state.
- Users must review actions before submitting forms or handling sensitive data.

## Best Practices

- Use only approved URLs and test accounts.
- Inspect a fresh snapshot before each page-changing action.
- Close sessions and protect recordings that may contain sensitive content.

## Anti Patterns

- Do not automate actions that violate website rules or user consent.
- Do not place passwords, tokens, or personal data in prompts or logs.
- Do not submit irreversible forms without explicit review and approval.

## Security Audit

- Audited at: 2026-07-27T11:19:58.78\+00:00
- Summary: All 265 static findings are false positives in documentation or transparent shell templates. The skill documents user-directed browser automation and contains no evidence of prompt injection, credential exfiltration, hidden network activity, or malicious system access.

## Stats

- Views: 0
- Downloads: 1
- Favorites: 0
- Popularity score: 0