Skip to content
Agentmart
Submit a tool

agent-browser

Browser automation CLI for AI agents.

Skill by MylesMCook

IndexedMIT
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/mylesmcook-agent-browser

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a website", "fill out a form", "click a button", "take a screenshot", "scrape data from a page", "test this web app", "login to a site", "automate browser actions", or any task requiring programmatic web interaction.

Use cases: frontend, testing, data, productivity

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found
Runs as
Shell scripts

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/agent-browser

Tools it pre-approves 3

From allowed-tools in SKILL.md: the agent may use these without asking you each time.

  • Bash(npx
  • agent-browser:*),
  • Bash(agent-browser:*)

What's inside

Skill name
agent-browser
Description
Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a website", "fill out a form", "click a button", "take a screenshot", "scrape data from a page", "test this web app", "login to a site", "automate browser actions", or any task requiring programmatic web interaction.
License
MIT
Allowed tools
Bash(npx, agent-browser:*),, Bash(agent-browser:*)
Instructions
110 lines in SKILL.md · view on GitHub

Before installing

  • Ships executable scripts
  • Pre-approves some shell commands

Files (17)

  • SKILL.md
  • authentication.md · references/
  • commands.md · references/
  • evidence-triage.md · references/
  • local-dev-e2e.md · references/
  • profiling.md · references/
  • proxy-support.md · references/
  • session-management.md · references/
  • snapshot-refs.md · references/
  • video-recording.md · references/
  • windows-powershell.md · references/
  • authenticated-session.sh · templates/
  • capture-workflow.sh · templates/
  • crud-scenario.md · templates/
  • form-automation.sh · templates/
  • local-dev-smoke.ps1 · templates/
  • powershell-session.ps1 · templates/

SKILL.md (first lines)

# Browser Automation with agent-browser

Use `agent-browser` when the browser itself is the work surface: local dev QA, auth flows, CRUD verification, scraping, screenshots, or reproducing UI issues.

On this machine:

- install or update with `vp install -g agent-browser@latest`
- verify with `agent-browser --version`
- run `agent-browser install` after install or upgrade to pin Chrome for Testing locally
- prefer the PowerShell-safe patterns in [references/windows-powershell.md](references/windows-powershell.md)

## Fast Rules

1. Give every task or scenario its own `AGENT_BROWSER_SESSION`. Name it `<run-id>-<scenario>`.
2. Refs are session-local and page-local. Never reuse `@e...` across sessions. Re-snapshot after navigation or DOM changes.
3. In PowerShell, quote refs: `agent-browser click '@e1'`. Do not use Bash-style `&&`.
4. Use this locator ladder: fresh refs, re-snapshot, semantic `find`, CSS selector, keyboard fallback.
5. For dialogs, popovers, comboboxes, sheets, and menus: open first, then snapshot the live subtree before the next action.
6. For mutations, clear the network log, perform the action, wait, inspect requests, then capture URL, title, errors, and a screenshot.
7. Prefer URL or element waits. Use `wait --load networkidle` only on pages whose activity actually settles.
8. Run interactive QA serially on one machine. Parallel sessions are fine for light scraping, but they add noise to mutation-heavy testing.
9. Use `--session-name` or `AGENT_BROWSER_SESSION_NAME` only when you intentionally want cookies and localStorage restored across runs.

## PowerShell Quick Start

```powershell
$runId = "probe-$(Get-Date -Format yyyyMMdd-HHmmss)"
$env:AGENT_BROWSER_SESSION = "$runId-login"

agent-browser open https://example.com/login
agent-browser snapshot -i --json
agent-browser fill '@e1' 'user@example.com'
agent-browser fill '@e2' 'password'
agent-browser click '@e3'
agent-browser wait 1500
agent-browser get url
agent-browser screenshot
agent-browser errors
```

## Standard Workflow

1. Open the route or page you actually want to validate.
2. Take `snapshot -i --json` if the next action depends on refs.
3. Interact using fresh refs.
4. Re-snapshot after anything that changes the DOM or active route.
5. For mutations, verify with `network requests`, `errors`, `console`, and a screenshot.
6. Restore baseline state before leaving the scenario unless the data is intentionally disposable.

## Local Dev Apps

For local web-app testing, use `agent-browser` as the primary browser executor and keep the app bootstrap outside the browser.

- start the app first, then test it
- use one session per scenario such as `cresa-ab-<run-id>-events`
- capture at least one screenshot per scenario
- record final URL, title, and page errors
- validate route correctness before treating a blank page as an agent-browser failure

Use [references/local-dev-e2e.md](references/local-dev-e2e.md) for the full pattern.

## Locator Strategy

Start with refs from a fresh snapshot. If the click or fill no-ops:

1. re-snapshot
2. retry with the new ref
3. use `find role`, `find text`, or `find label`
4. fall back to a CSS selector when the semantic target is stable
5. if a dialog submit still no-ops, try `focus` plus `press Enter`

Use [references/snapshot-refs.md](references/snapshot-refs.md) and [references/evidence-triage.md](references/evidence-triage.md) for the deeper workflow.

## Auth Choices

Use the simplest auth pattern that fits:

- one-off reuse of an already logged-in browser: state export or auto-connect
- recurring automation: `--profile` or `--session-name`
- secure stored credentials: auth vault

Details live in [references/authentication.md](references/authentication.md).

## Deep-Dive References

| Reference | Use When |
| --- | --- |
| [references/windows-powershell.md](references/windows-powershell.md) | PowerShell quoting, session env vars, and Windows shell behavior |
| [references/local-dev-e2e.md](references/local-dev-e2e.md) | Local app smoke tests, CRUD passes, mobile passes, and artifact capture |
| [references/evidence-triage.md](references/evidence-triage.md) | Verifying mutations, diagnosing no-op clicks, and collecting evidence |
| [references/snapshot-refs.md](references/snapshot-refs.md) | Ref lifecycle, dynamic widgets, and stale-ref recovery |
| [references/session-management.md](references/session-management.md) | Session isolation, persistence, and cleanup |
| [references/authentication.md](references/authentication.md) | Login flows, state reuse, OAuth, and 2FA handling |
| [references/commands.md](references/commands.md) | Command reference and less-common CLI options |
| [references/video-recording.md](references/video-recording.md) | Recording workflows for demos and bug reports |
| [references/profiling.md](references/profiling.md) | Performance and trace capture |
| [references/proxy-support.md](references/proxy-support.md) | Proxy and geo-testing configuration |

## Templates

| Template | Description |
| --- | --- |
| [templates/powershell-session.ps1](templates/powershell-session.ps1) | Create an isolated PowerShell session and run a small command block |
| [templates/local-dev-smoke.ps1](templates/local-dev-smoke.ps1) | Open a local app route, snapshot it, and capture evidence |
| [templates/crud-scenario.md](templates/crud-scenario.md) | Reusable CRUD scenario checklist with restore steps |
| [templates/form-automation.sh](templates/form-automation.sh) | Basic form filling flow |
| [templates/authenticated-session.sh](templates/authenticated-session.sh) | Login once and reuse saved state |
| [templates/capture-workflow.sh](templates/capture-workflow.sh) | Content extraction with screenshots |

From MylesMCook/mcook-skills, MIT.

More from MylesMCook/mcook-skills

Signals

Stars
0
Forks
0
Open issues
0
Last commit (this folder)
Apr 6, 2026 (6 months ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from MylesMCook

Cite this page

agent-browser skill. Agent Mart. https://agentmart.dev/skills/mylesmcook-agent-browser (last updated 2026-10-24)