| WEB_AGENT_SYSTEM_PROMPT = """ |
| You are a 'web agent' capable of performing browser-based goals. |
| You use the PinchTab browser control plane to navigate the web, perceive page state via accessibility tree snapshots, and perform actions like clicking, typing, and scrolling. |
| |
| ### Guidelines: |
| 1. **Observe first**: Always start by navigating to the target URL and taking a snapshot to understand the page structure. |
| 2. **Use References**: Use the `ref` (e.g., 'e5', 'e12') from the accessibility tree snapshot to interact with elements. |
| 3. **Be Incremental**: If a goal is complex, break it down into smaller steps (e.g., Search -> Click Result -> Extract Info). |
| 4. **Confirm Success**: After an action, take another snapshot or check the page text to verify the expected change occurred. |
| 5. **Efficiency**: Favor the accessibility tree snapshot (`browser.snapshot`) over full page text when looking for interactive elements. |
| |
| Today's date: {current_date} |
| """.strip() |
|
|
| WEB_TOOL_CONTENT = [ |
| { |
| "type": "function", |
| "function": { |
| "name": "browser.navigate", |
| "description": "Navigates the browser to a specific URL.", |
| "parameters": { |
| "type": "object", |
| "properties": { |
| "url": {"type": "string", "description": "The URL to navigate to"} |
| }, |
| "required": ["url"] |
| } |
| } |
| }, |
| { |
| "type": "function", |
| "function": { |
| "name": "browser.snapshot", |
| "description": "Captures a snapshot of the current page's accessibility tree, showing interactive elements and their references.", |
| "parameters": { |
| "type": "object", |
| "properties": { |
| "filter": { |
| "type": "string", |
| "enum": ["interactive", "all"], |
| "default": "interactive", |
| "description": "Whether to show only interactive elements or the entire tree." |
| } |
| } |
| } |
| } |
| }, |
| { |
| "type": "function", |
| "function": { |
| "name": "browser.click", |
| "description": "Clicks an element identified by its reference (ref) from the snapshot.", |
| "parameters": { |
| "type": "object", |
| "properties": { |
| "ref": {"type": "string", "description": "The element reference (e.g., 'e5')"} |
| }, |
| "required": ["ref"] |
| } |
| } |
| }, |
| { |
| "type": "function", |
| "function": { |
| "name": "browser.type", |
| "description": "Types text into an element identified by its reference (ref).", |
| "parameters": { |
| "type": "object", |
| "properties": { |
| "ref": {"type": "string", "description": "The element reference (e.g., 'e5')"}, |
| "text": {"type": "string", "description": "The text to type"} |
| }, |
| "required": ["ref", "text"] |
| } |
| } |
| }, |
| { |
| "type": "function", |
| "function": { |
| "name": "browser.press", |
| "description": "Presses a specific key (e.g., 'Enter', 'Tab').", |
| "parameters": { |
| "type": "object", |
| "properties": { |
| "key": {"type": "string", "description": "The key to press"} |
| }, |
| "required": ["key"] |
| } |
| } |
| }, |
| { |
| "type": "function", |
| "function": { |
| "name": "browser.get_text", |
| "description": "Extracts all visible text from the current page.", |
| "parameters": { |
| "type": "object", |
| "properties": {} |
| } |
| } |
| } |
| ] |
|
|
