chrome-devtools

name	chrome-devtools
description	Use this when the user needs to control Chrome, navigate to a page, inspect a tab, click or fill elements, take screenshots, or automate a browser flow with aeroxy/chrome-devtools-cli.
allowed-tools	["Bash(chrome-devtools:*)"]

Chrome DevTools CLI (aeroxy/chrome-devtools-cli)

Use this skill for browser automation tasks that can be solved with the chrome-devtools binary.

Install & Prerequisites

Install with one of:

brew tap aeroxy/chrome-devtools-cli && brew install chrome-devtools
cargo install chrome-devtools-cli

The binary name is chrome-devtools.

Chrome remote debugging must be enabled first at chrome://inspect/#remote-debugging.

See references/prerequisites.md.

Core Workflow (Target-Aware)

Page-level commands support both --page <index> and --target <target-id>.

For quick flows, --page is the simplest option. When a command emits a target marker, capture it and reuse --target for a stable multi-step flow.

chrome-devtools list-pages
chrome-devtools new-page https://example.com

# Quick smoke-test path
chrome-devtools --page 3 snapshot
chrome-devtools --page 3 evaluate "document.title"
chrome-devtools --page 3 screenshot --output /tmp/example.png

NAV_OUTPUT="$(chrome-devtools new-page https://example.com)"

if [[ "$NAV_OUTPUT" =~ \[target:([^]]+)\] ]]; then
  TARGET="${BASH_REMATCH[1]}"
elif [[ "$NAV_OUTPUT" =~ \(target:[[:space:]]*([^)]*)\) ]]; then
  TARGET="${BASH_REMATCH[1]}"
else
  echo "Failed to capture target from command output"
  exit 1
fi

chrome-devtools --target "$TARGET" snapshot
chrome-devtools --target "$TARGET" screenshot --output ./example.png

Supported Commands

This skill is grounded on the currently supported command set:

• navigate
• new-page
• close-page
• select-page
• list-pages
• screenshot
• snapshot
• evaluate
• click
• fill
• type-text
• press-key
• hover
• resize
• wait-for

See references/commands.md and references/workflow.md for usage patterns.

Capability Boundaries (Do Not Fake)

chrome-devtools does not directly provide prior agent-browser features such as:

• auth/session save-load commands
• credential vaults
• PDF export
• video recording
• upload/select/check helpers
• snapshot element refs like @e1
• network interception tooling
• profiling tools

When these are requested, be explicit about the gap and either:

1. Use supported CSS-selector + evaluate workarounds, or
2. Return a clear blocked/unsupported outcome.

See references/capability-gaps.md.

Templates

• templates/capture-workflow.sh
• templates/form-automation.sh
• templates/authenticated-session.sh