/browser - Agentic Browser Control

Control a browser directly from your CODITECT session. Navigate to pages, interact with elements, take screenshots, and extract data using snapshot-based element targeting with 93% token reduction compared to raw HTML.

Usage

# Navigation
/browser navigate "https://example.com"
/browser back
/browser forward
/browser reload

# Snapshot (get interactive elements with refs)
/browser snapshot              # Full page snapshot
/browser snapshot -i           # Interactive elements only (recommended)
/browser snapshot -s "#main"   # Scoped to CSS selector

# Interaction (use @eN refs from snapshot)
/browser click @e1
/browser fill @e2 "search query"
/browser type @e3 "Hello"     # Types character by character
/browser select @e4 "option"
/browser hover @e5

# Screenshots
/browser screenshot                      # Full page
/browser screenshot /tmp/page.png        # Save to path
/browser screenshot --element @e3        # Element only

# Data extraction
/browser gettext @e1                     # Get element text
/browser eval "document.title"           # Run JavaScript
/browser url                             # Get current URL
/browser title                           # Get page title

# Session management
/browser launch                          # Start new browser
/browser launch --headed                 # Visible browser window
/browser close                           # Close browser
/browser session list                    # List active sessions

# State persistence
/browser state save /tmp/auth.json       # Save cookies/localStorage
/browser state load /tmp/auth.json       # Restore saved state

System Prompt

EXECUTION DIRECTIVE: When the user invokes /browser, you MUST:

1. Check daemon status — Verify agent-browser daemon is running
2. Auto-launch if needed — Start daemon on first use (headless by default)
3. Execute the action — Run the appropriate agent-browser command
4. Display results — Show snapshot output, screenshot path, or extracted data
5. Cache state — Store latest snapshot in context for /cxq queries

Daemon Management:

# Check if daemon is running
SOCKET_PATH="${XDG_RUNTIME_DIR:-$HOME/.agent-browser}/default.sock"
if [ ! -S "$SOCKET_PATH" ]; then
    # Auto-launch daemon
    agent-browser launch
fi

Action Dispatch:

Action	agent-browser Command
navigate <url>	agent-browser navigate "<url>"
snapshot [-i] [-c] [-d N] [-s sel]	agent-browser snapshot [flags]
click <ref>	agent-browser click <ref>
fill <ref> <value>	agent-browser fill <ref> "<value>"
type <ref> <text>	agent-browser type <ref> "<text>"
screenshot [path]	agent-browser screenshot [path]
gettext <ref>	agent-browser gettext <ref>
eval <script>	agent-browser eval "<script>"
url	agent-browser url
title	agent-browser title
back	agent-browser back
forward	agent-browser forward
reload	agent-browser reload
launch [--headed]	agent-browser launch [--headed]
close	agent-browser close
state save <path>	agent-browser state_save "<path>"
state load <path>	agent-browser launch --state "<path>"
session list	List active daemon sockets

Snapshot Display:

When displaying snapshot output, format as a readable tree:

Page: https://example.com - "Example Page"

Interactive Elements:
  @e1  textbox "Email"
  @e2  textbox "Password"
  @e3  button "Sign In"
  @e4  link "Forgot password?"

Options

Option	Description
<action>	Browser action to perform (required)
-i	Interactive elements only (for snapshot)
-c	Compact mode (for snapshot)
-d N	Depth limit (for snapshot)
-s <selector>	CSS scope (for snapshot)
--headed	Show browser window (for launch)
--session <name>	Named session (default: "default")
--engine <name>	Browser engine: chromium, firefox, webkit
--json	Output raw JSON response

Examples

Quick Page Inspection

/browser navigate "https://example.com"
/browser snapshot -i

Fill and Submit Form

/browser navigate "https://login.example.com"
/browser snapshot -i
/browser fill @e1 "user@example.com"
/browser fill @e2 "password"
/browser click @e3

Screenshot for Debugging

/browser navigate "https://app.example.com/dashboard"
/browser screenshot /tmp/dashboard.png

Extract Page Data

/browser navigate "https://example.com/pricing"
/browser eval "JSON.stringify([...document.querySelectorAll('.plan')].map(p => ({name: p.querySelector('h2').textContent, price: p.querySelector('.price').textContent})))"

Success Output

Browser: navigated to https://example.com
Snapshot: 12 interactive elements found (use @e1-@e12)
Screenshot: saved to /tmp/screenshot.png

Related

• Command: /browser-extract — Content extraction to markdown/JSON
• Command: /browser-research — Multi-page agentic extraction
• Agent: coditect-browser-agent
• Skill: browser-automation-patterns
• Skill: browser-content-extraction
• Hook: browser-auto-launch.py
• Hook: browser-screenshot-on-error.py

Principles

This command embodies:

• #3 Complete Execution — Auto-launches daemon, executes action, displays results
• #6 Clear, Understandable — Element refs simplify targeting
• #1 Self-Provisioning — Daemon auto-starts on first use

---

Command Version: 1.0.0 Created: 2026-02-08 Author: CODITECT Core Team