Browser Tool

L'outil navigateur fournit aux agents PRX des capacites completes d'automatisation web -- navigating pages, filling forms, clicking elements, extracting content, and capturing screenshots. It uses a pluggable backend architecture supporting three automation engines and applique domain restrictions pour empecher unrestricted web access.

Les outils navigateur sont proteges par un feature gate et necessitent browser.enabled = true dans la configuration. Lorsqu'active, PRX registers browser et browser_open in le registre d'outils. The browser tool supports complex multi-step web workflows, tandis que browser_open provides a simpler interface for opening a URL et extracting its content.

PRX also includes vision-related tools (screenshot, image, image_info) that complement the browser tool for visual tasks. Screenshots captured par le browser tool peut etre passed to vision-capable LLMs for visual reasoning.

Configuration

[browser]
enabled = true
backend = "agent_browser"       # "agent_browser" | "rust_native" | "computer_use"
allowed_domains = ["github.com", "docs.rs", "*.openprx.dev", "stackoverflow.com"]
session_name = "default"        # Named browser session for persistent state

Backend Options

Backend	Description	Dependencies	Best For
agent_browser	Calls the agent-browser CLI, an external headless browser tool	agent-browser binary in PATH	General web automation, JavaScript-heavy sites
rust_native	Built-in Rust browser implementation using headless Chrome/Chromium	Chromium installed	Lightweight automation, no external dependencies
computer_use	Computer-use sidecar for full desktop interaction	Anthropic computer-use sidecar	OS-level interactions, complex GUI workflows

Restrictions de domaine

La liste allowed_domains controle quels domaines le navigateur peut acceder. La correspondance de domaine prend en charge:

• Exact match: "github.com" matches only github.com
• Subdomain wildcard: "*.openprx.dev" matches docs.openprx.dev, api.openprx.dev, etc.
• Non wildcard: An empty list bloque all browser navigation

[browser]
allowed_domains = [
  "github.com",
  "*.github.com",
  "docs.rs",
  "crates.io",
  "stackoverflow.com",
  "*.openprx.dev"
]

Utilisation

browser Tool

The main browser tool supports multiple actions for complex web workflows:

Navigate vers un URL:

{
  "name": "browser",
  "arguments": {
    "action": "navigate",
    "url": "https://github.com/openprx/prx"
  }
}

Fill a form field:

{
  "name": "browser",
  "arguments": {
    "action": "fill",
    "selector": "#search-input",
    "value": "PRX documentation"
  }
}

Click an element:

{
  "name": "browser",
  "arguments": {
    "action": "click",
    "selector": "button[type='submit']"
  }
}

Take a screenshot:

{
  "name": "browser",
  "arguments": {
    "action": "screenshot"
  }
}

Extract page content:

{
  "name": "browser",
  "arguments": {
    "action": "content"
  }
}

browser_open Tool

A simplified tool for opening a URL and retouring its content:

{
  "name": "browser_open",
  "arguments": {
    "url": "https://docs.rs/tokio/latest/tokio/"
  }
}

Multi-Step Workflow Example

A typical research workflow might chain multiple browser actions:

1. Navigate vers un search engine
2. Fill the search box avec un query
3. Click the search button
4. Extract results depuis le page
5. Navigate vers un relevant result
6. Extract the detailed content
7. Take a screenshot for visual reference

Parametres

browser Parameters

Parametre	Type	Requis	Defaut	Description
action	string	Oui	--	Action to perform: "navigate", "fill", "click", "screenshot", "content", "scroll", "wait", "back", "forward"
url	string	Conditional	--	URL to navigate to (required for "navigate" action)
selector	string	Conditional	--	CSS selector pour le target element (required for "fill", "click")
value	string	Conditional	--	Valeur to fill (required for "fill" action)
timeout_ms	integer	Non	30000	Maximum wait time pour le action to complete

browser_open Parameters

Parametre	Type	Requis	Defaut	Description
url	string	Oui	--	URL to open and extract content from

Vision Tool Parameters

screenshot:

Parametre	Type	Requis	Defaut	Description
target	string	Non	"screen"	What to capture: "screen" ou un window identifier

image:

Parametre	Type	Requis	Defaut	Description
action	string	Oui	--	Image operation: "resize", "crop", "convert"
path	string	Oui	--	Path vers le image file

image_info:

Parametre	Type	Requis	Defaut	Description
path	string	Oui	--	Path vers le image file to inspect

Backend Details

agent-browser

Le backend agent_browser delegue au agent-browser CLI tool, which provides a headless Chrome-based automation environment. Communication happens over stdio with JSON-RPC messages.

Advantages:

• Full JavaScript execution
• Cookie and session persistence
• Extension support

rust_native

The rust_native backend uses Rust bindings to control a local Chromium/Chrome installation directly. It communique via the Chrome DevTools Protocol (CDP).

Advantages:

• Non external binary dependency (beyond Chromium)
• Lower latency than lancement de a subprocess
• Tighter integration with PRX internals

computer_use

Le backend computer_use exploite le sidecar computer-use d'Anthropic pour effectuer des interactions au niveau du systeme d'exploitation including mouse movement, keyboard input, and screen capture. This goes beyond browser automation to full desktop control.

Advantages:

• Can interact with native applications, not just browsers
• Supports complex GUI workflows
• Handles popups, file dialogs, and system prompts

Securite

Domain Allowlist

The browser tool applique a strict domain allowlist. Before navigating to any URL:

1. L'URL est analysee et le nom d'hote est extrait
2. L'hotename est verifie par rapport a allowed_domains
3. Si aucun match is found, the navigation est bloque and an error is retournes

Cela empeche l'agent depuis unccessing arbitrary websites, which could expose it to malicious content ou trigger unintended actions on authenticated sessions.

Session Isolation

Browser sessions sont isoles by name. Different session d'agents or sub-agents can use separate browser contexts pour empecher state leakage (cookies, localStorage, session data).

Content Extraction Limites

Page content extraction est soumis a the web_search.fetch_max_chars limit pour empecher memory exhaustion from excessively large pages.

Moteur de politiques

Browser appels d'outils pass via le moteur de politiques de securite. L'outil peut etre denied entirely, or supervised to require approval for each navigation:

[security.tool_policy.tools]
browser = "supervised"
browser_open = "allow"

Credential Safety

The browser tool ne fait pas inject credentials ou authentication tokens into browser sessions. If l'agent doit authenticate sur un website, it must use the browser tool to fill login forms explicitly, which est soumis a supervision policies.

Voir aussi

• Web Search -- search the web without browser automation
• HTTP Request -- programmatic HTTP requests to APIs
• Shell Execution -- alternative for CLI-based web interactions (curl, wget)
• Security Sandbox -- process isolation for execution d'outil
• Configuration Reference -- [browser] configuration fields
• Tools Overview -- tous les outils et systeme de registre