How do I install the Agent Browser skill?

Run npx skillmds@latest add keith-cy/agent-browser in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Agent Browser safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Agent Browser?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Agent Browser free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Agent Browser?

Keith-CY (@keith-cy) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Agent Browser by keith-cy — Agent Skill

Name: Agent Browser
Author: Keith-CY

File contents

agent-browser

Overview

Use agent-browser (CLI) for deterministic, scriptable headless browser automation via snapshots and stable element refs (@e1, @e2, ...).

Core Workflow

Open a page: agent-browser open <url>
Get a ref-based accessibility snapshot: agent-browser snapshot -i
Interact using refs (preferred) or selectors:
- Click: agent-browser click @e2
- Fill: agent-browser fill @e3 "text"
- Press: agent-browser press Enter
Extract info:
- Visible text: agent-browser get text @e1
- URL/title: agent-browser get url / agent-browser get title
Capture artifacts:
- Screenshot: agent-browser screenshot --full path/to.png
- Trace: agent-browser trace start / agent-browser trace stop path/to-trace.json.gz

Quick Reference

Common Flags

Reuse a session across commands: --session <name> (or AGENT_BROWSER_SESSION=<name>)
Machine-readable output: --json
Debug output: --debug
Run with a visible window (debugging): --headed

Playwright → agent-browser Mapping

Intent	Playwright concept	`agent-browser` command
Navigate	`page.goto(url)`	`agent-browser open <url>`
Snapshot	a11y snapshot / locator discovery	`agent-browser snapshot -i`
Click	`locator.click()`	`agent-browser click <sel>` or `agent-browser click @eN`
Fill	`locator.fill()`	`agent-browser fill <sel> <text>` or `agent-browser fill @eN <text>`
Type	`locator.type()`	`agent-browser type <sel> <text>`
Press key	`page.keyboard.press()`	`agent-browser press <key>`
Wait for element	`page.waitForSelector()`	`agent-browser wait <sel>`
Read text	`locator.innerText()`	`agent-browser get text <sel>`
Screenshot	`page.screenshot()`	`agent-browser screenshot [--full] [path]`
Evaluate JS	`page.evaluate()`	`agent-browser eval <js>`

Examples

Login/Form Flow (ref-first)

export AGENT_BROWSER_SESSION="acme-login"
agent-browser open "https://example.com/login"
agent-browser snapshot -i
agent-browser fill @e12 "user@example.com"
agent-browser fill @e13 "correct horse battery staple"
agent-browser click @e14
agent-browser wait @e20
agent-browser get text @e20

Auth Headers (origin-scoped)

agent-browser open "https://example.com"
agent-browser --headers '{"Authorization":"Bearer YOUR_TOKEN"}' open "https://example.com/protected"
agent-browser snapshot -i --json

Common Mistakes

Not using --session when a multi-step flow spans multiple commands.
Clicking with brittle CSS selectors when a snapshot -i ref is available.
Forgetting --json when downstream steps require structured parsing.
In sandboxed environments, the agent-browser daemon may fail to start due to restricted writes to user cache/home; rerun with the necessary filesystem permissions.
If the browser can’t launch, run agent-browser install (may require network access).

---name: agent-browserdescription: Use when a task needs headless browser automation (navigate, click, fill forms, extract text, take screenshots) and Playwright-based tooling should be avoided; prefer the agent-browser CLI.---# agent-browser## OverviewUse `agent-browser` (CLI) for deterministic, scriptable headless browser automation via snapshots and stable element refs (`@e1`, `@e2`, ...).## Core Workflow1. Open a page: `agent-browser open <url>`2. Get a ref-based accessibility snapshot: `agent-browser snapshot -i`3. Interact using refs (preferred) or selectors: - Click: `agent-browser click @e2` - Fill: `agent-browser fill @e3 "text"` - Press: `agent-browser press Enter`4. Extract info: - Visible text: `agent-browser get text @e1` - URL/title: `agent-browser get url` / `agent-browser get title`5. Capture artifacts: - Screenshot: `agent-browser screenshot --full path/to.png` - Trace: `agent-browser trace start` / `agent-browser trace stop path/to-trace.json.gz`## Quick Reference### Common Flags- Reuse a session across commands: `--session <name>` (or `AGENT_BROWSER_SESSION=<name>`)- Machine-readable output: `--json`- Debug output: `--debug`- Run with a visible window (debugging): `--headed`### Playwright → agent-browser Mapping| Intent | Playwright concept | `agent-browser` command || --- | --- | --- || Navigate | `page.goto(url)` | `agent-browser open <url>` || Snapshot | a11y snapshot / locator discovery | `agent-browser snapshot -i` || Click | `locator.click()` | `agent-browser click <sel>` or `agent-browser click @eN` || Fill | `locator.fill()` | `agent-browser fill <sel> <text>` or `agent-browser fill @eN <text>` || Type | `locator.type()` | `agent-browser type <sel> <text>` || Press key | `page.keyboard.press()` | `agent-browser press <key>` || Wait for element | `page.waitForSelector()` | `agent-browser wait <sel>` || Read text | `locator.innerText()` | `agent-browser get text <sel>` || Screenshot | `page.screenshot()` | `agent-browser screenshot [--full] [path]` || Evaluate JS | `page.evaluate()` | `agent-browser eval <js>` |## Examples### Login/Form Flow (ref-first)```bashexport AGENT_BROWSER_SESSION="acme-login"agent-browser open "https://example.com/login"agent-browser snapshot -iagent-browser fill @e12 "user@example.com"agent-browser fill @e13 "correct horse battery staple"agent-browser click @e14agent-browser wait @e20agent-browser get text @e20```### Auth Headers (origin-scoped)```bashagent-browser open "https://example.com"agent-browser --headers '{"Authorization":"Bearer YOUR_TOKEN"}' open "https://example.com/protected"agent-browser snapshot -i --json```## Common Mistakes- Not using `--session` when a multi-step flow spans multiple commands.- Clicking with brittle CSS selectors when a `snapshot -i` ref is available.- Forgetting `--json` when downstream steps require structured parsing.- In sandboxed environments, the `agent-browser` daemon may fail to start due to restricted writes to user cache/home; rerun with the necessary filesystem permissions.- If the browser can’t launch, run `agent-browser install` (may require network access).

Frequently asked questions

Agent Browser