SkillsAdd

owenleezy/agent-browser-skill

agent-browser-automate

[INTERNAL TEMPLATE] Task automation execution protocol. Called by agent-browser orchestrator.

First seen May 5, 2026

Installation

$ npx skills add owenleezy/agent-browser-skill --skill agent-browser-automate
View on skills.sh ↗ GitHub repository ↗

Similar popular skills

Related neighbors and high-traction skills in the same topics — useful to compare before installing.

find-skills Same topic
vercel-labs/skills

Helps users discover and install agent skills when they ask questions like "how do I do X", "fi…

3.3M installs
agent-browser Same topic
vercel-labs/agent-browser

Browser automation CLI for AI agents. Use when the user needs to interact with websites, includ…

810.4K installs
vercel-react-best-practices Same topic
vercel-labs/agent-skills 697.8K installs
web-design-guidelines Same topic
vercel-labs/agent-skills

Review UI code for Web Interface Guidelines compliance. Use when asked to "review my UI", "chec…

617.3K installs
microsoft-foundry Same topic
microsoft/azure-skills

Build, deploy, evaluate, optimize, fine-tune, and manage Microsoft Foundry agents, models, and …

576.5K installs
azure-prepare Same topic
microsoft/azure-skills

Prepare azd-based Azure projects for deployment: generates azure.yaml, infrastructure (Bicep/Te…

568.3K installs

Also in this package

Other skills from owenleezy/agent-browser-skill.

npx skills add owenleezy/agent-browser-skill
agent-browser 67 agent-browser-debug 58 agent-browser-ios 56 agent-browser-visual 56 agent-browser-scrape 56 agent-browser-e2e 56 agent-browser-commands 56

Browse all from owenleezy/agent-browser-skill

More details

Agent compatibility

Declared targets from SKILL.md / docs. Unmarked agents are not listed — the skill may still install via the CLI.

Claude Code Not declared
Cursor Not declared
Codex Not declared
GitHub Copilot Not declared
Windsurf Not declared
Gemini CLI Not declared
Cline Not declared
OpenCode Not declared

Repository health

Stars 4
Default branch main
Open issues 0
Status Active

Package contents

Files included with this skill beyond the listing page.

  • skill md SKILL.md 4,769 B
  • docs SUMMARY.md 123 B

History

  1. First seen on skills.sh
  2. First recorded snapshot · 60 installs

SKILL.md

Task Automation — Execution Protocol

Internal template. Entry point: agent-browser skill.

NEVER

NEVER use `fill` when you want to append — it clears first; use `type` to append
NEVER upload a file with `click` then `fill` — use dedicated `upload <sel> <path>` command
NEVER submit a form without showing field values first — judgment gate required

Auth Setup (choose one, in priority order)

# Option 1 (recommended): Auth Vault — encrypted, reusable
agent-browser auth login <vault-name>
agent-browser open <url>

# Option 2: Persistent profile (local only)
agent-browser --profile ~/.myapp open <url>

# Option 3: Token injection
agent-browser open <url>
agent-browser storage local set authToken "<token>"
agent-browser reload

# Option 4: HTTP Basic Auth
agent-browser open <url>
agent-browser set credentials <user> <pass>

# Option 5: Bearer header
agent-browser --headers '{"Authorization":"Bearer <token>"}' open <url>

Form Automation Sequence

agent-browser snapshot -i                                    # get refs
agent-browser find label "Email" fill "[email protected]"
agent-browser find label "Password" fill "<password>"
agent-browser find role button click --name "Submit"
agent-browser wait --load networkidle
agent-browser screenshot --full step-complete.png           # evidence

File Operations

# Upload
agent-browser find label "Attachment" upload ./file.pdf

# Download (triggers on button click, waits for completion)
agent-browser find role button click --name "Export CSV"
agent-browser wait --download ~/report.csv

# Generate PDF
agent-browser pdf ./report.pdf

Multi-Tab Flows

agent-browser tab new
agent-browser open <second-url>
agent-browser tab 0       # switch back to first tab (0-based index)
agent-browser tab list    # see all tabs with URLs

Clipboard

# Read clipboard
agent-browser clipboard read

# Write to clipboard
agent-browser clipboard write "text to copy"

# Copy element text to clipboard
agent-browser clipboard copy @e3

# Paste clipboard into element
agent-browser clipboard paste @e5

# macOS: paste system clipboard into focused element
agent-browser keyboard inserttext "$(pbpaste)"

Fine Mouse Control (canvas / Kanban / drawing apps)

agent-browser mouse move <x> <y>
agent-browser mouse down
agent-browser mouse move <x2> <y2>   # drag
agent-browser mouse up

Waiting Patterns

# Page load
agent-browser wait --load networkidle

# Wait for element to appear
agent-browser wait "#success-toast"

# Wait for element to disappear (loading spinner)
agent-browser wait "#spinner" --state hidden

# Wait for URL to change (after form submit / redirect)
agent-browser wait --url "**/dashboard"

# Wait for text to appear
agent-browser wait --text "Upload complete"

# Wait for download to finish
agent-browser wait --download ./report.xlsx

Recovery Patterns

Failure Recovery
Auth expired auth login <name> re-authenticate
Auth Vault password changed auth delete <name> then auth save <name>
Unexpected popup/dialog dialog status to confirm, then dialog dismiss
fill overwrites existing content Use type to append
Modal won't scroll scroll down --selector ".modal-body"

Judgment Gates

Pause when:

  • About to submit a form (show field values → ask confirm)
  • About to trigger download / delete / payment
  • Auth fails after retry (need fresh credentials)

Pitfalls

Mistake Fix
Auth Vault password changed auth delete + auth save again — no in-place update
fill clears content unintentionally Use type to append
File upload fails Use find label "..." upload <path>, not click then fill
Can't scroll inside modal scroll down --selector ".modal-body"
Download not captured Use wait --download <path> after clicking the download button
JS dialog blocks daemon completely Playwright intercepts at CDP level — eval "window.confirm = () => true" does NOT work (JS layer override can't reach CDP). Only reliable fix: dialog accept must be called from a second concurrent process while click is blocking.
Read-only DOM nodes missing from snapshot (<p>, <span>, <div> without ARIA roles) snapshot only exposes the accessibility tree — interactive elements only. Read non-interactive node content with eval "document.querySelector('#id').textContent" or eval "document.querySelector('.class').innerText".