illuma-agents 1.0.30 → 1.0.32
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/cjs/tools/BrowserTools.cjs +43 -33
- package/dist/cjs/tools/BrowserTools.cjs.map +1 -1
- package/dist/esm/tools/BrowserTools.mjs +43 -33
- package/dist/esm/tools/BrowserTools.mjs.map +1 -1
- package/dist/types/tools/BrowserTools.d.ts +0 -23
- package/package.json +1 -1
- package/src/tools/BrowserTools.ts +45 -58
|
@@ -111,29 +111,26 @@ const BrowserKeypressSchema = zod.z.object({
|
|
|
111
111
|
/**
|
|
112
112
|
* Format browser action result for LLM consumption
|
|
113
113
|
*/
|
|
114
|
-
function formatResultForLLM(result, action
|
|
114
|
+
function formatResultForLLM(result, action) {
|
|
115
115
|
if (!result.success && result.error) {
|
|
116
|
-
|
|
117
|
-
return `${errorType}Action "${action}" failed: ${result.error}`;
|
|
116
|
+
return `Browser action "${action}" failed: ${result.error}`;
|
|
118
117
|
}
|
|
119
118
|
const parts = [];
|
|
120
|
-
|
|
121
|
-
|
|
122
|
-
|
|
123
|
-
|
|
124
|
-
|
|
125
|
-
|
|
126
|
-
|
|
127
|
-
|
|
128
|
-
|
|
119
|
+
if (result.url != null && result.url !== '') {
|
|
120
|
+
parts.push(`**Current URL:** ${result.url}`);
|
|
121
|
+
}
|
|
122
|
+
if (result.title != null && result.title !== '') {
|
|
123
|
+
parts.push(`**Page Title:** ${result.title}`);
|
|
124
|
+
}
|
|
125
|
+
if (result.elementList != null && result.elementList !== '') {
|
|
126
|
+
parts.push(`\n**Interactive Elements:**\n${result.elementList}`);
|
|
127
|
+
}
|
|
128
|
+
if (result.screenshot != null && result.screenshot !== '') {
|
|
129
|
+
parts.push('\n[Screenshot captured and displayed to user]');
|
|
130
|
+
}
|
|
131
|
+
if (parts.length === 0) {
|
|
132
|
+
return `Browser action "${action}" completed successfully.`;
|
|
129
133
|
}
|
|
130
|
-
// Page state
|
|
131
|
-
if (result.url)
|
|
132
|
-
parts.push(`URL: ${result.url}`);
|
|
133
|
-
if (result.title)
|
|
134
|
-
parts.push(`Title: ${result.title}`);
|
|
135
|
-
if (result.elementList)
|
|
136
|
-
parts.push(`\nElements:\n${result.elementList}`);
|
|
137
134
|
return parts.join('\n');
|
|
138
135
|
}
|
|
139
136
|
/**
|
|
@@ -176,40 +173,45 @@ function createBrowserTools(options) {
|
|
|
176
173
|
// The marker is still returned initially via SSE, but we wait for the callback
|
|
177
174
|
try {
|
|
178
175
|
const result = await waitForResult(action, args, toolCallId);
|
|
179
|
-
return formatResultForLLM(result, action
|
|
176
|
+
return formatResultForLLM(result, action);
|
|
180
177
|
}
|
|
181
178
|
catch (error) {
|
|
182
179
|
const errorMessage = error instanceof Error ? error.message : String(error);
|
|
183
|
-
return
|
|
184
|
-
|
|
185
|
-
**REQUIRED**: Analyze why the action failed and try an alternative approach.`;
|
|
180
|
+
return `Browser action "${action}" failed: ${errorMessage}`;
|
|
186
181
|
}
|
|
187
182
|
};
|
|
188
183
|
};
|
|
189
184
|
// browser_click
|
|
190
185
|
tools$1.push(tools.tool(createToolFunction('click'), {
|
|
191
186
|
name: EBrowserTools.CLICK,
|
|
192
|
-
description: `Click an element
|
|
193
|
-
|
|
187
|
+
description: `Click an element on the current web page by its index number.
|
|
188
|
+
The element list shows clickable items like: [0]<button>Submit</button> [1]<a href="/home">Home</a>
|
|
189
|
+
Use the index number in brackets to click that element.
|
|
190
|
+
After clicking, you receive an updated element list showing the new page state.`,
|
|
194
191
|
schema: BrowserClickSchema,
|
|
195
192
|
}));
|
|
196
193
|
// browser_type
|
|
197
194
|
tools$1.push(tools.tool(createToolFunction('type'), {
|
|
198
195
|
name: EBrowserTools.TYPE,
|
|
199
|
-
description: `Type text into an input element
|
|
200
|
-
|
|
196
|
+
description: `Type text into an input element on the page.
|
|
197
|
+
Find the input element in the list by its index (e.g., [5]<input placeholder="Search">).
|
|
198
|
+
Set pressEnter: true to submit forms after typing.
|
|
199
|
+
After typing, you receive an updated element list.`,
|
|
201
200
|
schema: BrowserTypeSchema,
|
|
202
201
|
}));
|
|
203
202
|
// browser_navigate
|
|
204
203
|
tools$1.push(tools.tool(createToolFunction('navigate'), {
|
|
205
204
|
name: EBrowserTools.NAVIGATE,
|
|
206
|
-
description: `Navigate to a URL
|
|
205
|
+
description: `Navigate to a URL. Always include the full URL with https://.
|
|
206
|
+
After navigation, you receive the new page's element list.`,
|
|
207
207
|
schema: BrowserNavigateSchema,
|
|
208
208
|
}));
|
|
209
209
|
// browser_scroll
|
|
210
210
|
tools$1.push(tools.tool(createToolFunction('scroll'), {
|
|
211
211
|
name: EBrowserTools.SCROLL,
|
|
212
|
-
description: `Scroll the page
|
|
212
|
+
description: `Scroll the page to reveal more content.
|
|
213
|
+
Use 'down' to scroll down, 'up' to scroll up.
|
|
214
|
+
After scrolling, you receive an updated element list with newly visible elements.`,
|
|
213
215
|
schema: BrowserScrollSchema,
|
|
214
216
|
}));
|
|
215
217
|
// browser_extract
|
|
@@ -244,19 +246,27 @@ After going back, you receive the previous page's element list.`,
|
|
|
244
246
|
// browser_screenshot
|
|
245
247
|
tools$1.push(tools.tool(createToolFunction('screenshot'), {
|
|
246
248
|
name: EBrowserTools.SCREENSHOT,
|
|
247
|
-
description: `Capture a screenshot
|
|
249
|
+
description: `Capture a screenshot of the current page.
|
|
250
|
+
Returns the page state with a note that screenshot was displayed to the user.
|
|
251
|
+
Use browser_get_page_state to get the element list for automation.`,
|
|
248
252
|
schema: BrowserScreenshotSchema,
|
|
249
253
|
}));
|
|
250
254
|
// browser_get_page_state
|
|
251
255
|
tools$1.push(tools.tool(createToolFunction('get_page_state'), {
|
|
252
256
|
name: EBrowserTools.GET_PAGE_STATE,
|
|
253
|
-
description: `Get current page URL, title, and interactive elements
|
|
257
|
+
description: `Get the current page state including URL, title, and all interactive elements.
|
|
258
|
+
Use this at the start of a task to see what elements are available.
|
|
259
|
+
Returns a text list of elements with their index numbers for interaction.`,
|
|
254
260
|
schema: BrowserGetPageStateSchema,
|
|
255
261
|
}));
|
|
256
|
-
// browser_select_option
|
|
262
|
+
// browser_select_option - Skyvern-inspired for robust dropdown handling
|
|
257
263
|
tools$1.push(tools.tool(createToolFunction('select_option'), {
|
|
258
264
|
name: EBrowserTools.SELECT_OPTION,
|
|
259
|
-
description: `Select an option from a dropdown
|
|
265
|
+
description: `Select an option from a dropdown or select element.
|
|
266
|
+
For native <select> elements: finds and selects the option by value/label.
|
|
267
|
+
For custom dropdowns: clicks to open, then clicks the matching option.
|
|
268
|
+
Use this instead of click for dropdowns - it handles both native and custom selects.
|
|
269
|
+
After selection, you receive an updated element list.`,
|
|
260
270
|
schema: BrowserSelectOptionSchema,
|
|
261
271
|
}));
|
|
262
272
|
// browser_upload_file - Skyvern-inspired for file input handling
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"BrowserTools.cjs","sources":["../../../src/tools/BrowserTools.ts"],"sourcesContent":["import { z } from 'zod';\nimport { tool, DynamicStructuredTool } from '@langchain/core/tools';\nimport type * as _t from '@/types';\n\n/**\n * Browser tool names - keep in sync with ranger-browser extension\n * These tools execute locally in the browser extension, NOT on the server\n */\nexport const EBrowserTools = {\n CLICK: 'browser_click',\n TYPE: 'browser_type',\n NAVIGATE: 'browser_navigate',\n SCROLL: 'browser_scroll',\n EXTRACT: 'browser_extract',\n HOVER: 'browser_hover',\n WAIT: 'browser_wait',\n BACK: 'browser_back',\n SCREENSHOT: 'browser_screenshot',\n GET_PAGE_STATE: 'browser_get_page_state',\n // Skyvern-inspired additions for robust form handling\n SELECT_OPTION: 'browser_select_option',\n UPLOAD_FILE: 'browser_upload_file',\n KEYPRESS: 'browser_keypress',\n} as const;\n\nexport type BrowserToolName =\n (typeof EBrowserTools)[keyof typeof EBrowserTools];\n\n/**\n * Callback function type for waiting on browser action results\n * This allows the server (Ranger) to provide a callback that waits for the extension\n * to POST results back to the server before returning to the LLM.\n *\n * @param action - The browser action (click, type, navigate, etc.)\n * @param args - Arguments for the action\n * @param toolCallId - Unique ID for this tool call (from config.toolCall.id)\n * @returns Promise that resolves with the actual browser result (page state, etc.)\n */\nexport type BrowserToolCallback = (\n action: string,\n args: Record<string, unknown>,\n toolCallId: string\n) => Promise<BrowserActionResult>;\n\n/**\n * Action status - matches Skyvern's ActionStatus\n */\nexport type ActionStatus = 'pending' | 'completed' | 'failed' | 'skipped';\n\n/**\n * Verification status - matches Skyvern's VerificationStatus\n * - complete: Goal achieved\n * - terminate: Goal cannot be achieved, stop\n * - continue: Goal not yet achieved, keep going\n */\nexport type VerificationStatus = 'complete' | 'terminate' | 'continue';\n\n/**\n * Result returned from browser action execution\n */\nexport interface BrowserActionResult {\n success: boolean;\n status?: ActionStatus;\n url?: string;\n title?: string;\n elementList?: string;\n error?: string;\n errorType?: string; // Typed error classification\n screenshot?: string;\n elementDescription?: string;\n // State verification (before/after comparison)\n verification?: {\n urlChanged: boolean;\n titleChanged: boolean;\n elementCountDelta: number;\n significantChange: boolean;\n dialogClosed: boolean;\n formGone: boolean;\n changeDescription: string;\n };\n}\n\n/**\n * Check if browser capability is available based on request headers or context\n * The browser extension sets these headers when connected:\n * - X-Ranger-Browser-Extension: true\n * - X-Ranger-Browser-Capable: true\n */\nexport function hasBrowserCapability(req?: {\n headers?: Record<string, string | string[] | undefined>;\n}): boolean {\n if (!req?.headers) {\n return false;\n }\n\n const browserExtension = req.headers['x-ranger-browser-extension'];\n const browserCapable = req.headers['x-ranger-browser-capable'];\n\n return browserExtension === 'true' || browserCapable === 'true';\n}\n\n// Tool schemas\nconst BrowserClickSchema = z.object({\n index: z\n .number()\n .describe(\n 'The index number [0], [1], etc. of the element to click from the page state element list'\n ),\n});\n\nconst BrowserTypeSchema = z.object({\n index: z\n .number()\n .describe('The index number of the input element to type into'),\n text: z.string().describe('The text to type into the element'),\n pressEnter: z\n .boolean()\n .optional()\n .describe('Whether to press Enter after typing (useful for search forms)'),\n});\n\nconst BrowserNavigateSchema = z.object({\n url: z\n .string()\n .describe('The full URL to navigate to (must include https://)'),\n});\n\nconst BrowserScrollSchema = z.object({\n direction: z\n .enum(['up', 'down', 'left', 'right'])\n .describe('Direction to scroll'),\n amount: z\n .number()\n .optional()\n .describe('Pixels to scroll (default: one viewport height)'),\n});\n\nconst BrowserExtractSchema = z.object({\n query: z\n .string()\n .optional()\n .describe('Optional: specific content to extract from the page'),\n});\n\nconst BrowserHoverSchema = z.object({\n index: z.number().describe('The index number of the element to hover over'),\n});\n\nconst BrowserWaitSchema = z.object({\n duration: z\n .number()\n .optional()\n .describe('Milliseconds to wait (default: 1000)'),\n});\n\nconst BrowserBackSchema = z.object({});\n\nconst BrowserScreenshotSchema = z.object({});\n\nconst BrowserGetPageStateSchema = z.object({});\n\n// Skyvern-inspired schemas for robust form handling\nconst BrowserSelectOptionSchema = z.object({\n index: z\n .number()\n .describe('The index number of the select/dropdown element'),\n value: z\n .string()\n .optional()\n .describe('The value or label of the option to select. For native <select>, use the option text. For custom dropdowns, this is the option label to click.'),\n});\n\nconst BrowserUploadFileSchema = z.object({\n index: z\n .number()\n .describe('The index number of the file input element'),\n fileUrl: z\n .string()\n .describe('URL of the file to upload (the system will download and upload it)'),\n});\n\nconst BrowserKeypressSchema = z.object({\n keys: z\n .string()\n .describe('Key(s) to press. Single key: \"Enter\", \"Escape\", \"Tab\", \"ArrowDown\". Combo: \"Control+A\", \"Shift+Enter\"'),\n});\n\n/**\n * Browser tool response interface\n * This is what the extension returns after executing the action\n */\nexport interface BrowserToolResponse {\n requiresBrowserExecution: true;\n action: string;\n args: Record<string, unknown>;\n toolCallId?: string; // Added to help extension correlate with callback\n}\n\n/**\n * Options for creating browser tools\n */\nexport interface CreateBrowserToolsOptions {\n /**\n * Optional callback that waits for browser action results.\n * When provided, tools will await this callback to get actual results from the extension.\n * When not provided, tools return markers immediately (for non-server contexts).\n */\n waitForResult?: BrowserToolCallback;\n}\n\n/**\n * Format browser action result for LLM consumption\n */\nfunction formatResultForLLM(\n result: BrowserActionResult,\n action: string,\n actionArgs?: Record<string, unknown>\n): string {\n if (!result.success && result.error) {\n const errorType = result.errorType ? `[${result.errorType}] ` : '';\n return `${errorType}Action \"${action}\" failed: ${result.error}`;\n }\n\n const parts: string[] = [];\n\n // Verification result (Skyvern pattern)\n if (result.verification) {\n const v = result.verification;\n if (v.dialogClosed || v.formGone || v.urlChanged) {\n parts.push(`✓ ${v.changeDescription}`);\n } else if (!v.significantChange) {\n parts.push(`⚠ No change detected`);\n }\n }\n\n // Page state\n if (result.url) parts.push(`URL: ${result.url}`);\n if (result.title) parts.push(`Title: ${result.title}`);\n if (result.elementList) parts.push(`\\nElements:\\n${result.elementList}`);\n\n return parts.join('\\n');\n}\n\n/**\n * Create browser tools with optional callback for waiting on results\n *\n * When waitForResult callback is provided:\n * 1. Tool returns marker that triggers extension\n * 2. Tool then awaits callback to get actual results\n * 3. Returns real page state to LLM\n *\n * When no callback:\n * 1. Tool returns marker only (for non-server contexts)\n *\n * NOTE: These tools use TEXT-BASED element lists, NOT screenshots\n * Screenshots would be 100K+ tokens each - element lists are ~100 tokens\n */\nexport function createBrowserTools(\n options?: CreateBrowserToolsOptions\n): DynamicStructuredTool[] {\n const { waitForResult } = options || {};\n const tools: DynamicStructuredTool[] = [];\n\n /**\n * Helper to create tool function that optionally waits for results\n * The toolCallId is extracted from the RunnableConfig passed by LangChain\n */\n const createToolFunction = (action: string) => {\n return async (\n args: Record<string, unknown>,\n config?: { toolCall?: { id?: string } }\n ): Promise<string> => {\n const toolCallId =\n config?.toolCall?.id ??\n `tool_${Date.now()}_${Math.random().toString(36).slice(2)}`;\n\n // Create marker for extension\n const marker: BrowserToolResponse = {\n requiresBrowserExecution: true,\n action,\n args,\n toolCallId,\n };\n\n // If no callback, return marker immediately (extension handles via SSE interception)\n if (!waitForResult) {\n return JSON.stringify(marker);\n }\n\n // With callback: wait for actual results from extension\n // The marker is still returned initially via SSE, but we wait for the callback\n try {\n const result = await waitForResult(action, args, toolCallId);\n return formatResultForLLM(result, action, args);\n } catch (error) {\n const errorMessage =\n error instanceof Error ? error.message : String(error);\n return `❌ **ACTION FAILED**: Browser action \"${action}\" failed: ${errorMessage}\n\n**REQUIRED**: Analyze why the action failed and try an alternative approach.`;\n }\n };\n };\n\n // browser_click\n tools.push(\n tool(createToolFunction('click'), {\n name: EBrowserTools.CLICK,\n description: `Click an element by its index number from the element list (e.g., [0], [1]).\nReturns state verification showing if UI changed (dialog closed, form submitted, etc.).`,\n schema: BrowserClickSchema,\n })\n );\n\n // browser_type\n tools.push(\n tool(createToolFunction('type'), {\n name: EBrowserTools.TYPE,\n description: `Type text into an input element by index. Set pressEnter: true to submit.\nReturns state verification showing if form submitted or errors appeared.`,\n schema: BrowserTypeSchema,\n })\n );\n\n // browser_navigate\n tools.push(\n tool(createToolFunction('navigate'), {\n name: EBrowserTools.NAVIGATE,\n description: `Navigate to a URL (include https://). Returns new page's element list.`,\n schema: BrowserNavigateSchema,\n })\n );\n\n // browser_scroll\n tools.push(\n tool(createToolFunction('scroll'), {\n name: EBrowserTools.SCROLL,\n description: `Scroll the page. Returns updated element list with newly visible elements.`,\n schema: BrowserScrollSchema,\n })\n );\n\n // browser_extract\n tools.push(\n tool(createToolFunction('extract'), {\n name: EBrowserTools.EXTRACT,\n description: `Extract content from the current page.\nReturns page URL, title, and element list.`,\n schema: BrowserExtractSchema,\n })\n );\n\n // browser_hover\n tools.push(\n tool(createToolFunction('hover'), {\n name: EBrowserTools.HOVER,\n description: `Hover over an element to reveal tooltips, dropdowns, or other hover-triggered content.\nAfter hovering, you receive an updated element list with any newly revealed elements.`,\n schema: BrowserHoverSchema,\n })\n );\n\n // browser_wait\n tools.push(\n tool(createToolFunction('wait'), {\n name: EBrowserTools.WAIT,\n description: `Wait for a specified duration for page content to load.\nUse this after actions that trigger async content loading.\nAfter waiting, you receive an updated element list.`,\n schema: BrowserWaitSchema,\n })\n );\n\n // browser_back\n tools.push(\n tool(createToolFunction('back'), {\n name: EBrowserTools.BACK,\n description: `Go back to the previous page in browser history.\nAfter going back, you receive the previous page's element list.`,\n schema: BrowserBackSchema,\n })\n );\n\n // browser_screenshot\n tools.push(\n tool(createToolFunction('screenshot'), {\n name: EBrowserTools.SCREENSHOT,\n description: `Capture a screenshot. Use browser_get_page_state for element list.`,\n schema: BrowserScreenshotSchema,\n })\n );\n\n // browser_get_page_state\n tools.push(\n tool(createToolFunction('get_page_state'), {\n name: EBrowserTools.GET_PAGE_STATE,\n description: `Get current page URL, title, and interactive elements list with index numbers.`,\n schema: BrowserGetPageStateSchema,\n })\n );\n\n // browser_select_option\n tools.push(\n tool(createToolFunction('select_option'), {\n name: EBrowserTools.SELECT_OPTION,\n description: `Select an option from a dropdown. Handles both native <select> and custom dropdowns.`,\n schema: BrowserSelectOptionSchema,\n })\n );\n\n // browser_upload_file - Skyvern-inspired for file input handling\n tools.push(\n tool(createToolFunction('upload_file'), {\n name: EBrowserTools.UPLOAD_FILE,\n description: `Upload a file to a file input element.\nProvide the index of the file input and the URL of the file to upload.\nThe system will download the file and attach it to the input.\nAfter upload, you receive an updated element list.`,\n schema: BrowserUploadFileSchema,\n })\n );\n\n // browser_keypress - For keyboard shortcuts and special keys\n tools.push(\n tool(createToolFunction('keypress'), {\n name: EBrowserTools.KEYPRESS,\n description: `Press keyboard key(s) on the page.\nSingle keys: \"Enter\", \"Escape\", \"Tab\", \"ArrowDown\", \"ArrowUp\", \"Backspace\", \"Delete\"\nKey combos: \"Control+A\" (select all), \"Control+C\" (copy), \"Shift+Enter\" (newline)\nUse this for form submission, closing modals, navigating dropdowns.\nAfter keypress, you receive an updated element list.`,\n schema: BrowserKeypressSchema,\n })\n );\n\n return tools;\n}\n"],"names":["z","tools","tool"],"mappings":";;;;;AAIA;;;AAGG;AACU,MAAA,aAAa,GAAG;AAC3B,IAAA,KAAK,EAAE,eAAe;AACtB,IAAA,IAAI,EAAE,cAAc;AACpB,IAAA,QAAQ,EAAE,kBAAkB;AAC5B,IAAA,MAAM,EAAE,gBAAgB;AACxB,IAAA,OAAO,EAAE,iBAAiB;AAC1B,IAAA,KAAK,EAAE,eAAe;AACtB,IAAA,IAAI,EAAE,cAAc;AACpB,IAAA,IAAI,EAAE,cAAc;AACpB,IAAA,UAAU,EAAE,oBAAoB;AAChC,IAAA,cAAc,EAAE,wBAAwB;;AAExC,IAAA,aAAa,EAAE,uBAAuB;AACtC,IAAA,WAAW,EAAE,qBAAqB;AAClC,IAAA,QAAQ,EAAE,kBAAkB;;AA4D9B;;;;;AAKG;AACG,SAAU,oBAAoB,CAAC,GAEpC,EAAA;AACC,IAAA,IAAI,CAAC,GAAG,EAAE,OAAO,EAAE;AACjB,QAAA,OAAO,KAAK;;IAGd,MAAM,gBAAgB,GAAG,GAAG,CAAC,OAAO,CAAC,4BAA4B,CAAC;IAClE,MAAM,cAAc,GAAG,GAAG,CAAC,OAAO,CAAC,0BAA0B,CAAC;AAE9D,IAAA,OAAO,gBAAgB,KAAK,MAAM,IAAI,cAAc,KAAK,MAAM;AACjE;AAEA;AACA,MAAM,kBAAkB,GAAGA,KAAC,CAAC,MAAM,CAAC;AAClC,IAAA,KAAK,EAAEA;AACJ,SAAA,MAAM;SACN,QAAQ,CACP,0FAA0F,CAC3F;AACJ,CAAA,CAAC;AAEF,MAAM,iBAAiB,GAAGA,KAAC,CAAC,MAAM,CAAC;AACjC,IAAA,KAAK,EAAEA;AACJ,SAAA,MAAM;SACN,QAAQ,CAAC,oDAAoD,CAAC;IACjE,IAAI,EAAEA,KAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,mCAAmC,CAAC;AAC9D,IAAA,UAAU,EAAEA;AACT,SAAA,OAAO;AACP,SAAA,QAAQ;SACR,QAAQ,CAAC,+DAA+D,CAAC;AAC7E,CAAA,CAAC;AAEF,MAAM,qBAAqB,GAAGA,KAAC,CAAC,MAAM,CAAC;AACrC,IAAA,GAAG,EAAEA;AACF,SAAA,MAAM;SACN,QAAQ,CAAC,qDAAqD,CAAC;AACnE,CAAA,CAAC;AAEF,MAAM,mBAAmB,GAAGA,KAAC,CAAC,MAAM,CAAC;AACnC,IAAA,SAAS,EAAEA;SACR,IAAI,CAAC,CAAC,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,OAAO,CAAC;SACpC,QAAQ,CAAC,qBAAqB,CAAC;AAClC,IAAA,MAAM,EAAEA;AACL,SAAA,MAAM;AACN,SAAA,QAAQ;SACR,QAAQ,CAAC,iDAAiD,CAAC;AAC/D,CAAA,CAAC;AAEF,MAAM,oBAAoB,GAAGA,KAAC,CAAC,MAAM,CAAC;AACpC,IAAA,KAAK,EAAEA;AACJ,SAAA,MAAM;AACN,SAAA,QAAQ;SACR,QAAQ,CAAC,qDAAqD,CAAC;AACnE,CAAA,CAAC;AAEF,MAAM,kBAAkB,GAAGA,KAAC,CAAC,MAAM,CAAC;IAClC,KAAK,EAAEA,KAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,+CAA+C,CAAC;AAC5E,CAAA,CAAC;AAEF,MAAM,iBAAiB,GAAGA,KAAC,CAAC,MAAM,CAAC;AACjC,IAAA,QAAQ,EAAEA;AACP,SAAA,MAAM;AACN,SAAA,QAAQ;SACR,QAAQ,CAAC,sCAAsC,CAAC;AACpD,CAAA,CAAC;AAEF,MAAM,iBAAiB,GAAGA,KAAC,CAAC,MAAM,CAAC,EAAE,CAAC;AAEtC,MAAM,uBAAuB,GAAGA,KAAC,CAAC,MAAM,CAAC,EAAE,CAAC;AAE5C,MAAM,yBAAyB,GAAGA,KAAC,CAAC,MAAM,CAAC,EAAE,CAAC;AAE9C;AACA,MAAM,yBAAyB,GAAGA,KAAC,CAAC,MAAM,CAAC;AACzC,IAAA,KAAK,EAAEA;AACJ,SAAA,MAAM;SACN,QAAQ,CAAC,iDAAiD,CAAC;AAC9D,IAAA,KAAK,EAAEA;AACJ,SAAA,MAAM;AACN,SAAA,QAAQ;SACR,QAAQ,CAAC,gJAAgJ,CAAC;AAC9J,CAAA,CAAC;AAEF,MAAM,uBAAuB,GAAGA,KAAC,CAAC,MAAM,CAAC;AACvC,IAAA,KAAK,EAAEA;AACJ,SAAA,MAAM;SACN,QAAQ,CAAC,4CAA4C,CAAC;AACzD,IAAA,OAAO,EAAEA;AACN,SAAA,MAAM;SACN,QAAQ,CAAC,oEAAoE,CAAC;AAClF,CAAA,CAAC;AAEF,MAAM,qBAAqB,GAAGA,KAAC,CAAC,MAAM,CAAC;AACrC,IAAA,IAAI,EAAEA;AACH,SAAA,MAAM;SACN,QAAQ,CAAC,uGAAuG,CAAC;AACrH,CAAA,CAAC;AAyBF;;AAEG;AACH,SAAS,kBAAkB,CACzB,MAA2B,EAC3B,MAAc,EACd,UAAoC,EAAA;IAEpC,IAAI,CAAC,MAAM,CAAC,OAAO,IAAI,MAAM,CAAC,KAAK,EAAE;AACnC,QAAA,MAAM,SAAS,GAAG,MAAM,CAAC,SAAS,GAAG,CAAI,CAAA,EAAA,MAAM,CAAC,SAAS,CAAA,EAAA,CAAI,GAAG,EAAE;QAClE,OAAO,CAAA,EAAG,SAAS,CAAW,QAAA,EAAA,MAAM,aAAa,MAAM,CAAC,KAAK,CAAA,CAAE;;IAGjE,MAAM,KAAK,GAAa,EAAE;;AAG1B,IAAA,IAAI,MAAM,CAAC,YAAY,EAAE;AACvB,QAAA,MAAM,CAAC,GAAG,MAAM,CAAC,YAAY;AAC7B,QAAA,IAAI,CAAC,CAAC,YAAY,IAAI,CAAC,CAAC,QAAQ,IAAI,CAAC,CAAC,UAAU,EAAE;YAChD,KAAK,CAAC,IAAI,CAAC,CAAA,EAAA,EAAK,CAAC,CAAC,iBAAiB,CAAE,CAAA,CAAC;;AACjC,aAAA,IAAI,CAAC,CAAC,CAAC,iBAAiB,EAAE;AAC/B,YAAA,KAAK,CAAC,IAAI,CAAC,CAAA,oBAAA,CAAsB,CAAC;;;;IAKtC,IAAI,MAAM,CAAC,GAAG;QAAE,KAAK,CAAC,IAAI,CAAC,CAAA,KAAA,EAAQ,MAAM,CAAC,GAAG,CAAE,CAAA,CAAC;IAChD,IAAI,MAAM,CAAC,KAAK;QAAE,KAAK,CAAC,IAAI,CAAC,CAAA,OAAA,EAAU,MAAM,CAAC,KAAK,CAAE,CAAA,CAAC;IACtD,IAAI,MAAM,CAAC,WAAW;QAAE,KAAK,CAAC,IAAI,CAAC,CAAA,aAAA,EAAgB,MAAM,CAAC,WAAW,CAAE,CAAA,CAAC;AAExE,IAAA,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC;AACzB;AAEA;;;;;;;;;;;;;AAaG;AACG,SAAU,kBAAkB,CAChC,OAAmC,EAAA;AAEnC,IAAA,MAAM,EAAE,aAAa,EAAE,GAAG,OAAO,IAAI,EAAE;IACvC,MAAMC,OAAK,GAA4B,EAAE;AAEzC;;;AAGG;AACH,IAAA,MAAM,kBAAkB,GAAG,CAAC,MAAc,KAAI;AAC5C,QAAA,OAAO,OACL,IAA6B,EAC7B,MAAuC,KACpB;AACnB,YAAA,MAAM,UAAU,GACd,MAAM,EAAE,QAAQ,EAAE,EAAE;gBACpB,CAAQ,KAAA,EAAA,IAAI,CAAC,GAAG,EAAE,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE;;AAG7D,YAAA,MAAM,MAAM,GAAwB;AAClC,gBAAA,wBAAwB,EAAE,IAAI;gBAC9B,MAAM;gBACN,IAAI;gBACJ,UAAU;aACX;;YAGD,IAAI,CAAC,aAAa,EAAE;AAClB,gBAAA,OAAO,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC;;;;AAK/B,YAAA,IAAI;gBACF,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC,MAAM,EAAE,IAAI,EAAE,UAAU,CAAC;gBAC5D,OAAO,kBAAkB,CAAC,MAAM,EAAE,MAAM,EAAE,IAAI,CAAC;;YAC/C,OAAO,KAAK,EAAE;AACd,gBAAA,MAAM,YAAY,GAChB,KAAK,YAAY,KAAK,GAAG,KAAK,CAAC,OAAO,GAAG,MAAM,CAAC,KAAK,CAAC;gBACxD,OAAO,CAAA,qCAAA,EAAwC,MAAM,CAAA,UAAA,EAAa,YAAY;;6EAET;;AAEzE,SAAC;AACH,KAAC;;IAGDA,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,OAAO,CAAC,EAAE;QAChC,IAAI,EAAE,aAAa,CAAC,KAAK;AACzB,QAAA,WAAW,EAAE,CAAA;AACqE,uFAAA,CAAA;AAClF,QAAA,MAAM,EAAE,kBAAkB;AAC3B,KAAA,CAAC,CACH;;IAGDD,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,MAAM,CAAC,EAAE;QAC/B,IAAI,EAAE,aAAa,CAAC,IAAI;AACxB,QAAA,WAAW,EAAE,CAAA;AACsD,wEAAA,CAAA;AACnE,QAAA,MAAM,EAAE,iBAAiB;AAC1B,KAAA,CAAC,CACH;;IAGDD,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,UAAU,CAAC,EAAE;QACnC,IAAI,EAAE,aAAa,CAAC,QAAQ;AAC5B,QAAA,WAAW,EAAE,CAAwE,sEAAA,CAAA;AACrF,QAAA,MAAM,EAAE,qBAAqB;AAC9B,KAAA,CAAC,CACH;;IAGDD,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,QAAQ,CAAC,EAAE;QACjC,IAAI,EAAE,aAAa,CAAC,MAAM;AAC1B,QAAA,WAAW,EAAE,CAA4E,0EAAA,CAAA;AACzF,QAAA,MAAM,EAAE,mBAAmB;AAC5B,KAAA,CAAC,CACH;;IAGDD,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,SAAS,CAAC,EAAE;QAClC,IAAI,EAAE,aAAa,CAAC,OAAO;AAC3B,QAAA,WAAW,EAAE,CAAA;AACwB,0CAAA,CAAA;AACrC,QAAA,MAAM,EAAE,oBAAoB;AAC7B,KAAA,CAAC,CACH;;IAGDD,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,OAAO,CAAC,EAAE;QAChC,IAAI,EAAE,aAAa,CAAC,KAAK;AACzB,QAAA,WAAW,EAAE,CAAA;AACmE,qFAAA,CAAA;AAChF,QAAA,MAAM,EAAE,kBAAkB;AAC3B,KAAA,CAAC,CACH;;IAGDD,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,MAAM,CAAC,EAAE;QAC/B,IAAI,EAAE,aAAa,CAAC,IAAI;AACxB,QAAA,WAAW,EAAE,CAAA;;AAEiC,mDAAA,CAAA;AAC9C,QAAA,MAAM,EAAE,iBAAiB;AAC1B,KAAA,CAAC,CACH;;IAGDD,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,MAAM,CAAC,EAAE;QAC/B,IAAI,EAAE,aAAa,CAAC,IAAI;AACxB,QAAA,WAAW,EAAE,CAAA;AAC6C,+DAAA,CAAA;AAC1D,QAAA,MAAM,EAAE,iBAAiB;AAC1B,KAAA,CAAC,CACH;;IAGDD,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,YAAY,CAAC,EAAE;QACrC,IAAI,EAAE,aAAa,CAAC,UAAU;AAC9B,QAAA,WAAW,EAAE,CAAoE,kEAAA,CAAA;AACjF,QAAA,MAAM,EAAE,uBAAuB;AAChC,KAAA,CAAC,CACH;;IAGDD,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,gBAAgB,CAAC,EAAE;QACzC,IAAI,EAAE,aAAa,CAAC,cAAc;AAClC,QAAA,WAAW,EAAE,CAAgF,8EAAA,CAAA;AAC7F,QAAA,MAAM,EAAE,yBAAyB;AAClC,KAAA,CAAC,CACH;;IAGDD,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,eAAe,CAAC,EAAE;QACxC,IAAI,EAAE,aAAa,CAAC,aAAa;AACjC,QAAA,WAAW,EAAE,CAAsF,oFAAA,CAAA;AACnG,QAAA,MAAM,EAAE,yBAAyB;AAClC,KAAA,CAAC,CACH;;IAGDD,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,aAAa,CAAC,EAAE;QACtC,IAAI,EAAE,aAAa,CAAC,WAAW;AAC/B,QAAA,WAAW,EAAE,CAAA;;;AAGgC,kDAAA,CAAA;AAC7C,QAAA,MAAM,EAAE,uBAAuB;AAChC,KAAA,CAAC,CACH;;IAGDD,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,UAAU,CAAC,EAAE;QACnC,IAAI,EAAE,aAAa,CAAC,QAAQ;AAC5B,QAAA,WAAW,EAAE,CAAA;;;;AAIkC,oDAAA,CAAA;AAC/C,QAAA,MAAM,EAAE,qBAAqB;AAC9B,KAAA,CAAC,CACH;AAED,IAAA,OAAOD,OAAK;AACd;;;;;;"}
|
|
1
|
+
{"version":3,"file":"BrowserTools.cjs","sources":["../../../src/tools/BrowserTools.ts"],"sourcesContent":["import { z } from 'zod';\nimport { tool, DynamicStructuredTool } from '@langchain/core/tools';\nimport type * as _t from '@/types';\n\n/**\n * Browser tool names - keep in sync with ranger-browser extension\n * These tools execute locally in the browser extension, NOT on the server\n */\nexport const EBrowserTools = {\n CLICK: 'browser_click',\n TYPE: 'browser_type',\n NAVIGATE: 'browser_navigate',\n SCROLL: 'browser_scroll',\n EXTRACT: 'browser_extract',\n HOVER: 'browser_hover',\n WAIT: 'browser_wait',\n BACK: 'browser_back',\n SCREENSHOT: 'browser_screenshot',\n GET_PAGE_STATE: 'browser_get_page_state',\n // Skyvern-inspired additions for robust form handling\n SELECT_OPTION: 'browser_select_option',\n UPLOAD_FILE: 'browser_upload_file',\n KEYPRESS: 'browser_keypress',\n} as const;\n\nexport type BrowserToolName =\n (typeof EBrowserTools)[keyof typeof EBrowserTools];\n\n/**\n * Callback function type for waiting on browser action results\n * This allows the server (Ranger) to provide a callback that waits for the extension\n * to POST results back to the server before returning to the LLM.\n *\n * @param action - The browser action (click, type, navigate, etc.)\n * @param args - Arguments for the action\n * @param toolCallId - Unique ID for this tool call (from config.toolCall.id)\n * @returns Promise that resolves with the actual browser result (page state, etc.)\n */\nexport type BrowserToolCallback = (\n action: string,\n args: Record<string, unknown>,\n toolCallId: string\n) => Promise<BrowserActionResult>;\n\n/**\n * Result returned from browser action execution\n */\nexport interface BrowserActionResult {\n success: boolean;\n url?: string;\n title?: string;\n elementList?: string; // Text-based element list\n error?: string;\n screenshot?: string; // Base64 screenshot (if requested)\n}\n\n/**\n * Check if browser capability is available based on request headers or context\n * The browser extension sets these headers when connected:\n * - X-Ranger-Browser-Extension: true\n * - X-Ranger-Browser-Capable: true\n */\nexport function hasBrowserCapability(req?: {\n headers?: Record<string, string | string[] | undefined>;\n}): boolean {\n if (!req?.headers) {\n return false;\n }\n\n const browserExtension = req.headers['x-ranger-browser-extension'];\n const browserCapable = req.headers['x-ranger-browser-capable'];\n\n return browserExtension === 'true' || browserCapable === 'true';\n}\n\n// Tool schemas\nconst BrowserClickSchema = z.object({\n index: z\n .number()\n .describe(\n 'The index number [0], [1], etc. of the element to click from the page state element list'\n ),\n});\n\nconst BrowserTypeSchema = z.object({\n index: z\n .number()\n .describe('The index number of the input element to type into'),\n text: z.string().describe('The text to type into the element'),\n pressEnter: z\n .boolean()\n .optional()\n .describe('Whether to press Enter after typing (useful for search forms)'),\n});\n\nconst BrowserNavigateSchema = z.object({\n url: z\n .string()\n .describe('The full URL to navigate to (must include https://)'),\n});\n\nconst BrowserScrollSchema = z.object({\n direction: z\n .enum(['up', 'down', 'left', 'right'])\n .describe('Direction to scroll'),\n amount: z\n .number()\n .optional()\n .describe('Pixels to scroll (default: one viewport height)'),\n});\n\nconst BrowserExtractSchema = z.object({\n query: z\n .string()\n .optional()\n .describe('Optional: specific content to extract from the page'),\n});\n\nconst BrowserHoverSchema = z.object({\n index: z.number().describe('The index number of the element to hover over'),\n});\n\nconst BrowserWaitSchema = z.object({\n duration: z\n .number()\n .optional()\n .describe('Milliseconds to wait (default: 1000)'),\n});\n\nconst BrowserBackSchema = z.object({});\n\nconst BrowserScreenshotSchema = z.object({});\n\nconst BrowserGetPageStateSchema = z.object({});\n\n// Skyvern-inspired schemas for robust form handling\nconst BrowserSelectOptionSchema = z.object({\n index: z\n .number()\n .describe('The index number of the select/dropdown element'),\n value: z\n .string()\n .optional()\n .describe('The value or label of the option to select. For native <select>, use the option text. For custom dropdowns, this is the option label to click.'),\n});\n\nconst BrowserUploadFileSchema = z.object({\n index: z\n .number()\n .describe('The index number of the file input element'),\n fileUrl: z\n .string()\n .describe('URL of the file to upload (the system will download and upload it)'),\n});\n\nconst BrowserKeypressSchema = z.object({\n keys: z\n .string()\n .describe('Key(s) to press. Single key: \"Enter\", \"Escape\", \"Tab\", \"ArrowDown\". Combo: \"Control+A\", \"Shift+Enter\"'),\n});\n\n/**\n * Browser tool response interface\n * This is what the extension returns after executing the action\n */\nexport interface BrowserToolResponse {\n requiresBrowserExecution: true;\n action: string;\n args: Record<string, unknown>;\n toolCallId?: string; // Added to help extension correlate with callback\n}\n\n/**\n * Options for creating browser tools\n */\nexport interface CreateBrowserToolsOptions {\n /**\n * Optional callback that waits for browser action results.\n * When provided, tools will await this callback to get actual results from the extension.\n * When not provided, tools return markers immediately (for non-server contexts).\n */\n waitForResult?: BrowserToolCallback;\n}\n\n/**\n * Format browser action result for LLM consumption\n */\nfunction formatResultForLLM(\n result: BrowserActionResult,\n action: string\n): string {\n if (!result.success && result.error) {\n return `Browser action \"${action}\" failed: ${result.error}`;\n }\n\n const parts: string[] = [];\n\n if (result.url != null && result.url !== '') {\n parts.push(`**Current URL:** ${result.url}`);\n }\n if (result.title != null && result.title !== '') {\n parts.push(`**Page Title:** ${result.title}`);\n }\n if (result.elementList != null && result.elementList !== '') {\n parts.push(`\\n**Interactive Elements:**\\n${result.elementList}`);\n }\n if (result.screenshot != null && result.screenshot !== '') {\n parts.push('\\n[Screenshot captured and displayed to user]');\n }\n\n if (parts.length === 0) {\n return `Browser action \"${action}\" completed successfully.`;\n }\n\n return parts.join('\\n');\n}\n\n/**\n * Create browser tools with optional callback for waiting on results\n *\n * When waitForResult callback is provided:\n * 1. Tool returns marker that triggers extension\n * 2. Tool then awaits callback to get actual results\n * 3. Returns real page state to LLM\n *\n * When no callback:\n * 1. Tool returns marker only (for non-server contexts)\n *\n * NOTE: These tools use TEXT-BASED element lists, NOT screenshots\n * Screenshots would be 100K+ tokens each - element lists are ~100 tokens\n */\nexport function createBrowserTools(\n options?: CreateBrowserToolsOptions\n): DynamicStructuredTool[] {\n const { waitForResult } = options || {};\n const tools: DynamicStructuredTool[] = [];\n\n /**\n * Helper to create tool function that optionally waits for results\n * The toolCallId is extracted from the RunnableConfig passed by LangChain\n */\n const createToolFunction = (action: string) => {\n return async (\n args: Record<string, unknown>,\n config?: { toolCall?: { id?: string } }\n ): Promise<string> => {\n const toolCallId =\n config?.toolCall?.id ??\n `tool_${Date.now()}_${Math.random().toString(36).slice(2)}`;\n\n // Create marker for extension\n const marker: BrowserToolResponse = {\n requiresBrowserExecution: true,\n action,\n args,\n toolCallId,\n };\n\n // If no callback, return marker immediately (extension handles via SSE interception)\n if (!waitForResult) {\n return JSON.stringify(marker);\n }\n\n // With callback: wait for actual results from extension\n // The marker is still returned initially via SSE, but we wait for the callback\n try {\n const result = await waitForResult(action, args, toolCallId);\n return formatResultForLLM(result, action);\n } catch (error) {\n const errorMessage =\n error instanceof Error ? error.message : String(error);\n return `Browser action \"${action}\" failed: ${errorMessage}`;\n }\n };\n };\n\n // browser_click\n tools.push(\n tool(createToolFunction('click'), {\n name: EBrowserTools.CLICK,\n description: `Click an element on the current web page by its index number.\nThe element list shows clickable items like: [0]<button>Submit</button> [1]<a href=\"/home\">Home</a>\nUse the index number in brackets to click that element.\nAfter clicking, you receive an updated element list showing the new page state.`,\n schema: BrowserClickSchema,\n })\n );\n\n // browser_type\n tools.push(\n tool(createToolFunction('type'), {\n name: EBrowserTools.TYPE,\n description: `Type text into an input element on the page.\nFind the input element in the list by its index (e.g., [5]<input placeholder=\"Search\">).\nSet pressEnter: true to submit forms after typing.\nAfter typing, you receive an updated element list.`,\n schema: BrowserTypeSchema,\n })\n );\n\n // browser_navigate\n tools.push(\n tool(createToolFunction('navigate'), {\n name: EBrowserTools.NAVIGATE,\n description: `Navigate to a URL. Always include the full URL with https://.\nAfter navigation, you receive the new page's element list.`,\n schema: BrowserNavigateSchema,\n })\n );\n\n // browser_scroll\n tools.push(\n tool(createToolFunction('scroll'), {\n name: EBrowserTools.SCROLL,\n description: `Scroll the page to reveal more content.\nUse 'down' to scroll down, 'up' to scroll up.\nAfter scrolling, you receive an updated element list with newly visible elements.`,\n schema: BrowserScrollSchema,\n })\n );\n\n // browser_extract\n tools.push(\n tool(createToolFunction('extract'), {\n name: EBrowserTools.EXTRACT,\n description: `Extract content from the current page.\nReturns page URL, title, and element list.`,\n schema: BrowserExtractSchema,\n })\n );\n\n // browser_hover\n tools.push(\n tool(createToolFunction('hover'), {\n name: EBrowserTools.HOVER,\n description: `Hover over an element to reveal tooltips, dropdowns, or other hover-triggered content.\nAfter hovering, you receive an updated element list with any newly revealed elements.`,\n schema: BrowserHoverSchema,\n })\n );\n\n // browser_wait\n tools.push(\n tool(createToolFunction('wait'), {\n name: EBrowserTools.WAIT,\n description: `Wait for a specified duration for page content to load.\nUse this after actions that trigger async content loading.\nAfter waiting, you receive an updated element list.`,\n schema: BrowserWaitSchema,\n })\n );\n\n // browser_back\n tools.push(\n tool(createToolFunction('back'), {\n name: EBrowserTools.BACK,\n description: `Go back to the previous page in browser history.\nAfter going back, you receive the previous page's element list.`,\n schema: BrowserBackSchema,\n })\n );\n\n // browser_screenshot\n tools.push(\n tool(createToolFunction('screenshot'), {\n name: EBrowserTools.SCREENSHOT,\n description: `Capture a screenshot of the current page.\nReturns the page state with a note that screenshot was displayed to the user.\nUse browser_get_page_state to get the element list for automation.`,\n schema: BrowserScreenshotSchema,\n })\n );\n\n // browser_get_page_state\n tools.push(\n tool(createToolFunction('get_page_state'), {\n name: EBrowserTools.GET_PAGE_STATE,\n description: `Get the current page state including URL, title, and all interactive elements.\nUse this at the start of a task to see what elements are available.\nReturns a text list of elements with their index numbers for interaction.`,\n schema: BrowserGetPageStateSchema,\n })\n );\n\n // browser_select_option - Skyvern-inspired for robust dropdown handling\n tools.push(\n tool(createToolFunction('select_option'), {\n name: EBrowserTools.SELECT_OPTION,\n description: `Select an option from a dropdown or select element.\nFor native <select> elements: finds and selects the option by value/label.\nFor custom dropdowns: clicks to open, then clicks the matching option.\nUse this instead of click for dropdowns - it handles both native and custom selects.\nAfter selection, you receive an updated element list.`,\n schema: BrowserSelectOptionSchema,\n })\n );\n\n // browser_upload_file - Skyvern-inspired for file input handling\n tools.push(\n tool(createToolFunction('upload_file'), {\n name: EBrowserTools.UPLOAD_FILE,\n description: `Upload a file to a file input element.\nProvide the index of the file input and the URL of the file to upload.\nThe system will download the file and attach it to the input.\nAfter upload, you receive an updated element list.`,\n schema: BrowserUploadFileSchema,\n })\n );\n\n // browser_keypress - For keyboard shortcuts and special keys\n tools.push(\n tool(createToolFunction('keypress'), {\n name: EBrowserTools.KEYPRESS,\n description: `Press keyboard key(s) on the page.\nSingle keys: \"Enter\", \"Escape\", \"Tab\", \"ArrowDown\", \"ArrowUp\", \"Backspace\", \"Delete\"\nKey combos: \"Control+A\" (select all), \"Control+C\" (copy), \"Shift+Enter\" (newline)\nUse this for form submission, closing modals, navigating dropdowns.\nAfter keypress, you receive an updated element list.`,\n schema: BrowserKeypressSchema,\n })\n );\n\n return tools;\n}\n"],"names":["z","tools","tool"],"mappings":";;;;;AAIA;;;AAGG;AACU,MAAA,aAAa,GAAG;AAC3B,IAAA,KAAK,EAAE,eAAe;AACtB,IAAA,IAAI,EAAE,cAAc;AACpB,IAAA,QAAQ,EAAE,kBAAkB;AAC5B,IAAA,MAAM,EAAE,gBAAgB;AACxB,IAAA,OAAO,EAAE,iBAAiB;AAC1B,IAAA,KAAK,EAAE,eAAe;AACtB,IAAA,IAAI,EAAE,cAAc;AACpB,IAAA,IAAI,EAAE,cAAc;AACpB,IAAA,UAAU,EAAE,oBAAoB;AAChC,IAAA,cAAc,EAAE,wBAAwB;;AAExC,IAAA,aAAa,EAAE,uBAAuB;AACtC,IAAA,WAAW,EAAE,qBAAqB;AAClC,IAAA,QAAQ,EAAE,kBAAkB;;AAkC9B;;;;;AAKG;AACG,SAAU,oBAAoB,CAAC,GAEpC,EAAA;AACC,IAAA,IAAI,CAAC,GAAG,EAAE,OAAO,EAAE;AACjB,QAAA,OAAO,KAAK;;IAGd,MAAM,gBAAgB,GAAG,GAAG,CAAC,OAAO,CAAC,4BAA4B,CAAC;IAClE,MAAM,cAAc,GAAG,GAAG,CAAC,OAAO,CAAC,0BAA0B,CAAC;AAE9D,IAAA,OAAO,gBAAgB,KAAK,MAAM,IAAI,cAAc,KAAK,MAAM;AACjE;AAEA;AACA,MAAM,kBAAkB,GAAGA,KAAC,CAAC,MAAM,CAAC;AAClC,IAAA,KAAK,EAAEA;AACJ,SAAA,MAAM;SACN,QAAQ,CACP,0FAA0F,CAC3F;AACJ,CAAA,CAAC;AAEF,MAAM,iBAAiB,GAAGA,KAAC,CAAC,MAAM,CAAC;AACjC,IAAA,KAAK,EAAEA;AACJ,SAAA,MAAM;SACN,QAAQ,CAAC,oDAAoD,CAAC;IACjE,IAAI,EAAEA,KAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,mCAAmC,CAAC;AAC9D,IAAA,UAAU,EAAEA;AACT,SAAA,OAAO;AACP,SAAA,QAAQ;SACR,QAAQ,CAAC,+DAA+D,CAAC;AAC7E,CAAA,CAAC;AAEF,MAAM,qBAAqB,GAAGA,KAAC,CAAC,MAAM,CAAC;AACrC,IAAA,GAAG,EAAEA;AACF,SAAA,MAAM;SACN,QAAQ,CAAC,qDAAqD,CAAC;AACnE,CAAA,CAAC;AAEF,MAAM,mBAAmB,GAAGA,KAAC,CAAC,MAAM,CAAC;AACnC,IAAA,SAAS,EAAEA;SACR,IAAI,CAAC,CAAC,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,OAAO,CAAC;SACpC,QAAQ,CAAC,qBAAqB,CAAC;AAClC,IAAA,MAAM,EAAEA;AACL,SAAA,MAAM;AACN,SAAA,QAAQ;SACR,QAAQ,CAAC,iDAAiD,CAAC;AAC/D,CAAA,CAAC;AAEF,MAAM,oBAAoB,GAAGA,KAAC,CAAC,MAAM,CAAC;AACpC,IAAA,KAAK,EAAEA;AACJ,SAAA,MAAM;AACN,SAAA,QAAQ;SACR,QAAQ,CAAC,qDAAqD,CAAC;AACnE,CAAA,CAAC;AAEF,MAAM,kBAAkB,GAAGA,KAAC,CAAC,MAAM,CAAC;IAClC,KAAK,EAAEA,KAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,+CAA+C,CAAC;AAC5E,CAAA,CAAC;AAEF,MAAM,iBAAiB,GAAGA,KAAC,CAAC,MAAM,CAAC;AACjC,IAAA,QAAQ,EAAEA;AACP,SAAA,MAAM;AACN,SAAA,QAAQ;SACR,QAAQ,CAAC,sCAAsC,CAAC;AACpD,CAAA,CAAC;AAEF,MAAM,iBAAiB,GAAGA,KAAC,CAAC,MAAM,CAAC,EAAE,CAAC;AAEtC,MAAM,uBAAuB,GAAGA,KAAC,CAAC,MAAM,CAAC,EAAE,CAAC;AAE5C,MAAM,yBAAyB,GAAGA,KAAC,CAAC,MAAM,CAAC,EAAE,CAAC;AAE9C;AACA,MAAM,yBAAyB,GAAGA,KAAC,CAAC,MAAM,CAAC;AACzC,IAAA,KAAK,EAAEA;AACJ,SAAA,MAAM;SACN,QAAQ,CAAC,iDAAiD,CAAC;AAC9D,IAAA,KAAK,EAAEA;AACJ,SAAA,MAAM;AACN,SAAA,QAAQ;SACR,QAAQ,CAAC,gJAAgJ,CAAC;AAC9J,CAAA,CAAC;AAEF,MAAM,uBAAuB,GAAGA,KAAC,CAAC,MAAM,CAAC;AACvC,IAAA,KAAK,EAAEA;AACJ,SAAA,MAAM;SACN,QAAQ,CAAC,4CAA4C,CAAC;AACzD,IAAA,OAAO,EAAEA;AACN,SAAA,MAAM;SACN,QAAQ,CAAC,oEAAoE,CAAC;AAClF,CAAA,CAAC;AAEF,MAAM,qBAAqB,GAAGA,KAAC,CAAC,MAAM,CAAC;AACrC,IAAA,IAAI,EAAEA;AACH,SAAA,MAAM;SACN,QAAQ,CAAC,uGAAuG,CAAC;AACrH,CAAA,CAAC;AAyBF;;AAEG;AACH,SAAS,kBAAkB,CACzB,MAA2B,EAC3B,MAAc,EAAA;IAEd,IAAI,CAAC,MAAM,CAAC,OAAO,IAAI,MAAM,CAAC,KAAK,EAAE;AACnC,QAAA,OAAO,mBAAmB,MAAM,CAAA,UAAA,EAAa,MAAM,CAAC,KAAK,EAAE;;IAG7D,MAAM,KAAK,GAAa,EAAE;AAE1B,IAAA,IAAI,MAAM,CAAC,GAAG,IAAI,IAAI,IAAI,MAAM,CAAC,GAAG,KAAK,EAAE,EAAE;QAC3C,KAAK,CAAC,IAAI,CAAC,CAAA,iBAAA,EAAoB,MAAM,CAAC,GAAG,CAAE,CAAA,CAAC;;AAE9C,IAAA,IAAI,MAAM,CAAC,KAAK,IAAI,IAAI,IAAI,MAAM,CAAC,KAAK,KAAK,EAAE,EAAE;QAC/C,KAAK,CAAC,IAAI,CAAC,CAAA,gBAAA,EAAmB,MAAM,CAAC,KAAK,CAAE,CAAA,CAAC;;AAE/C,IAAA,IAAI,MAAM,CAAC,WAAW,IAAI,IAAI,IAAI,MAAM,CAAC,WAAW,KAAK,EAAE,EAAE;QAC3D,KAAK,CAAC,IAAI,CAAC,CAAA,6BAAA,EAAgC,MAAM,CAAC,WAAW,CAAE,CAAA,CAAC;;AAElE,IAAA,IAAI,MAAM,CAAC,UAAU,IAAI,IAAI,IAAI,MAAM,CAAC,UAAU,KAAK,EAAE,EAAE;AACzD,QAAA,KAAK,CAAC,IAAI,CAAC,+CAA+C,CAAC;;AAG7D,IAAA,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE;QACtB,OAAO,CAAA,gBAAA,EAAmB,MAAM,CAAA,yBAAA,CAA2B;;AAG7D,IAAA,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC;AACzB;AAEA;;;;;;;;;;;;;AAaG;AACG,SAAU,kBAAkB,CAChC,OAAmC,EAAA;AAEnC,IAAA,MAAM,EAAE,aAAa,EAAE,GAAG,OAAO,IAAI,EAAE;IACvC,MAAMC,OAAK,GAA4B,EAAE;AAEzC;;;AAGG;AACH,IAAA,MAAM,kBAAkB,GAAG,CAAC,MAAc,KAAI;AAC5C,QAAA,OAAO,OACL,IAA6B,EAC7B,MAAuC,KACpB;AACnB,YAAA,MAAM,UAAU,GACd,MAAM,EAAE,QAAQ,EAAE,EAAE;gBACpB,CAAQ,KAAA,EAAA,IAAI,CAAC,GAAG,EAAE,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE;;AAG7D,YAAA,MAAM,MAAM,GAAwB;AAClC,gBAAA,wBAAwB,EAAE,IAAI;gBAC9B,MAAM;gBACN,IAAI;gBACJ,UAAU;aACX;;YAGD,IAAI,CAAC,aAAa,EAAE;AAClB,gBAAA,OAAO,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC;;;;AAK/B,YAAA,IAAI;gBACF,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC,MAAM,EAAE,IAAI,EAAE,UAAU,CAAC;AAC5D,gBAAA,OAAO,kBAAkB,CAAC,MAAM,EAAE,MAAM,CAAC;;YACzC,OAAO,KAAK,EAAE;AACd,gBAAA,MAAM,YAAY,GAChB,KAAK,YAAY,KAAK,GAAG,KAAK,CAAC,OAAO,GAAG,MAAM,CAAC,KAAK,CAAC;AACxD,gBAAA,OAAO,CAAmB,gBAAA,EAAA,MAAM,CAAa,UAAA,EAAA,YAAY,EAAE;;AAE/D,SAAC;AACH,KAAC;;IAGDA,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,OAAO,CAAC,EAAE;QAChC,IAAI,EAAE,aAAa,CAAC,KAAK;AACzB,QAAA,WAAW,EAAE,CAAA;;;AAG6D,+EAAA,CAAA;AAC1E,QAAA,MAAM,EAAE,kBAAkB;AAC3B,KAAA,CAAC,CACH;;IAGDD,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,MAAM,CAAC,EAAE;QAC/B,IAAI,EAAE,aAAa,CAAC,IAAI;AACxB,QAAA,WAAW,EAAE,CAAA;;;AAGgC,kDAAA,CAAA;AAC7C,QAAA,MAAM,EAAE,iBAAiB;AAC1B,KAAA,CAAC,CACH;;IAGDD,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,UAAU,CAAC,EAAE;QACnC,IAAI,EAAE,aAAa,CAAC,QAAQ;AAC5B,QAAA,WAAW,EAAE,CAAA;AACwC,0DAAA,CAAA;AACrD,QAAA,MAAM,EAAE,qBAAqB;AAC9B,KAAA,CAAC,CACH;;IAGDD,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,QAAQ,CAAC,EAAE;QACjC,IAAI,EAAE,aAAa,CAAC,MAAM;AAC1B,QAAA,WAAW,EAAE,CAAA;;AAE+D,iFAAA,CAAA;AAC5E,QAAA,MAAM,EAAE,mBAAmB;AAC5B,KAAA,CAAC,CACH;;IAGDD,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,SAAS,CAAC,EAAE;QAClC,IAAI,EAAE,aAAa,CAAC,OAAO;AAC3B,QAAA,WAAW,EAAE,CAAA;AACwB,0CAAA,CAAA;AACrC,QAAA,MAAM,EAAE,oBAAoB;AAC7B,KAAA,CAAC,CACH;;IAGDD,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,OAAO,CAAC,EAAE;QAChC,IAAI,EAAE,aAAa,CAAC,KAAK;AACzB,QAAA,WAAW,EAAE,CAAA;AACmE,qFAAA,CAAA;AAChF,QAAA,MAAM,EAAE,kBAAkB;AAC3B,KAAA,CAAC,CACH;;IAGDD,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,MAAM,CAAC,EAAE;QAC/B,IAAI,EAAE,aAAa,CAAC,IAAI;AACxB,QAAA,WAAW,EAAE,CAAA;;AAEiC,mDAAA,CAAA;AAC9C,QAAA,MAAM,EAAE,iBAAiB;AAC1B,KAAA,CAAC,CACH;;IAGDD,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,MAAM,CAAC,EAAE;QAC/B,IAAI,EAAE,aAAa,CAAC,IAAI;AACxB,QAAA,WAAW,EAAE,CAAA;AAC6C,+DAAA,CAAA;AAC1D,QAAA,MAAM,EAAE,iBAAiB;AAC1B,KAAA,CAAC,CACH;;IAGDD,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,YAAY,CAAC,EAAE;QACrC,IAAI,EAAE,aAAa,CAAC,UAAU;AAC9B,QAAA,WAAW,EAAE,CAAA;;AAEgD,kEAAA,CAAA;AAC7D,QAAA,MAAM,EAAE,uBAAuB;AAChC,KAAA,CAAC,CACH;;IAGDD,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,gBAAgB,CAAC,EAAE;QACzC,IAAI,EAAE,aAAa,CAAC,cAAc;AAClC,QAAA,WAAW,EAAE,CAAA;;AAEuD,yEAAA,CAAA;AACpE,QAAA,MAAM,EAAE,yBAAyB;AAClC,KAAA,CAAC,CACH;;IAGDD,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,eAAe,CAAC,EAAE;QACxC,IAAI,EAAE,aAAa,CAAC,aAAa;AACjC,QAAA,WAAW,EAAE,CAAA;;;;AAImC,qDAAA,CAAA;AAChD,QAAA,MAAM,EAAE,yBAAyB;AAClC,KAAA,CAAC,CACH;;IAGDD,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,aAAa,CAAC,EAAE;QACtC,IAAI,EAAE,aAAa,CAAC,WAAW;AAC/B,QAAA,WAAW,EAAE,CAAA;;;AAGgC,kDAAA,CAAA;AAC7C,QAAA,MAAM,EAAE,uBAAuB;AAChC,KAAA,CAAC,CACH;;IAGDD,OAAK,CAAC,IAAI,CACRC,UAAI,CAAC,kBAAkB,CAAC,UAAU,CAAC,EAAE;QACnC,IAAI,EAAE,aAAa,CAAC,QAAQ;AAC5B,QAAA,WAAW,EAAE,CAAA;;;;AAIkC,oDAAA,CAAA;AAC/C,QAAA,MAAM,EAAE,qBAAqB;AAC9B,KAAA,CAAC,CACH;AAED,IAAA,OAAOD,OAAK;AACd;;;;;;"}
|
|
@@ -109,29 +109,26 @@ const BrowserKeypressSchema = z.object({
|
|
|
109
109
|
/**
|
|
110
110
|
* Format browser action result for LLM consumption
|
|
111
111
|
*/
|
|
112
|
-
function formatResultForLLM(result, action
|
|
112
|
+
function formatResultForLLM(result, action) {
|
|
113
113
|
if (!result.success && result.error) {
|
|
114
|
-
|
|
115
|
-
return `${errorType}Action "${action}" failed: ${result.error}`;
|
|
114
|
+
return `Browser action "${action}" failed: ${result.error}`;
|
|
116
115
|
}
|
|
117
116
|
const parts = [];
|
|
118
|
-
|
|
119
|
-
|
|
120
|
-
|
|
121
|
-
|
|
122
|
-
|
|
123
|
-
|
|
124
|
-
|
|
125
|
-
|
|
126
|
-
|
|
117
|
+
if (result.url != null && result.url !== '') {
|
|
118
|
+
parts.push(`**Current URL:** ${result.url}`);
|
|
119
|
+
}
|
|
120
|
+
if (result.title != null && result.title !== '') {
|
|
121
|
+
parts.push(`**Page Title:** ${result.title}`);
|
|
122
|
+
}
|
|
123
|
+
if (result.elementList != null && result.elementList !== '') {
|
|
124
|
+
parts.push(`\n**Interactive Elements:**\n${result.elementList}`);
|
|
125
|
+
}
|
|
126
|
+
if (result.screenshot != null && result.screenshot !== '') {
|
|
127
|
+
parts.push('\n[Screenshot captured and displayed to user]');
|
|
128
|
+
}
|
|
129
|
+
if (parts.length === 0) {
|
|
130
|
+
return `Browser action "${action}" completed successfully.`;
|
|
127
131
|
}
|
|
128
|
-
// Page state
|
|
129
|
-
if (result.url)
|
|
130
|
-
parts.push(`URL: ${result.url}`);
|
|
131
|
-
if (result.title)
|
|
132
|
-
parts.push(`Title: ${result.title}`);
|
|
133
|
-
if (result.elementList)
|
|
134
|
-
parts.push(`\nElements:\n${result.elementList}`);
|
|
135
132
|
return parts.join('\n');
|
|
136
133
|
}
|
|
137
134
|
/**
|
|
@@ -174,40 +171,45 @@ function createBrowserTools(options) {
|
|
|
174
171
|
// The marker is still returned initially via SSE, but we wait for the callback
|
|
175
172
|
try {
|
|
176
173
|
const result = await waitForResult(action, args, toolCallId);
|
|
177
|
-
return formatResultForLLM(result, action
|
|
174
|
+
return formatResultForLLM(result, action);
|
|
178
175
|
}
|
|
179
176
|
catch (error) {
|
|
180
177
|
const errorMessage = error instanceof Error ? error.message : String(error);
|
|
181
|
-
return
|
|
182
|
-
|
|
183
|
-
**REQUIRED**: Analyze why the action failed and try an alternative approach.`;
|
|
178
|
+
return `Browser action "${action}" failed: ${errorMessage}`;
|
|
184
179
|
}
|
|
185
180
|
};
|
|
186
181
|
};
|
|
187
182
|
// browser_click
|
|
188
183
|
tools.push(tool(createToolFunction('click'), {
|
|
189
184
|
name: EBrowserTools.CLICK,
|
|
190
|
-
description: `Click an element
|
|
191
|
-
|
|
185
|
+
description: `Click an element on the current web page by its index number.
|
|
186
|
+
The element list shows clickable items like: [0]<button>Submit</button> [1]<a href="/home">Home</a>
|
|
187
|
+
Use the index number in brackets to click that element.
|
|
188
|
+
After clicking, you receive an updated element list showing the new page state.`,
|
|
192
189
|
schema: BrowserClickSchema,
|
|
193
190
|
}));
|
|
194
191
|
// browser_type
|
|
195
192
|
tools.push(tool(createToolFunction('type'), {
|
|
196
193
|
name: EBrowserTools.TYPE,
|
|
197
|
-
description: `Type text into an input element
|
|
198
|
-
|
|
194
|
+
description: `Type text into an input element on the page.
|
|
195
|
+
Find the input element in the list by its index (e.g., [5]<input placeholder="Search">).
|
|
196
|
+
Set pressEnter: true to submit forms after typing.
|
|
197
|
+
After typing, you receive an updated element list.`,
|
|
199
198
|
schema: BrowserTypeSchema,
|
|
200
199
|
}));
|
|
201
200
|
// browser_navigate
|
|
202
201
|
tools.push(tool(createToolFunction('navigate'), {
|
|
203
202
|
name: EBrowserTools.NAVIGATE,
|
|
204
|
-
description: `Navigate to a URL
|
|
203
|
+
description: `Navigate to a URL. Always include the full URL with https://.
|
|
204
|
+
After navigation, you receive the new page's element list.`,
|
|
205
205
|
schema: BrowserNavigateSchema,
|
|
206
206
|
}));
|
|
207
207
|
// browser_scroll
|
|
208
208
|
tools.push(tool(createToolFunction('scroll'), {
|
|
209
209
|
name: EBrowserTools.SCROLL,
|
|
210
|
-
description: `Scroll the page
|
|
210
|
+
description: `Scroll the page to reveal more content.
|
|
211
|
+
Use 'down' to scroll down, 'up' to scroll up.
|
|
212
|
+
After scrolling, you receive an updated element list with newly visible elements.`,
|
|
211
213
|
schema: BrowserScrollSchema,
|
|
212
214
|
}));
|
|
213
215
|
// browser_extract
|
|
@@ -242,19 +244,27 @@ After going back, you receive the previous page's element list.`,
|
|
|
242
244
|
// browser_screenshot
|
|
243
245
|
tools.push(tool(createToolFunction('screenshot'), {
|
|
244
246
|
name: EBrowserTools.SCREENSHOT,
|
|
245
|
-
description: `Capture a screenshot
|
|
247
|
+
description: `Capture a screenshot of the current page.
|
|
248
|
+
Returns the page state with a note that screenshot was displayed to the user.
|
|
249
|
+
Use browser_get_page_state to get the element list for automation.`,
|
|
246
250
|
schema: BrowserScreenshotSchema,
|
|
247
251
|
}));
|
|
248
252
|
// browser_get_page_state
|
|
249
253
|
tools.push(tool(createToolFunction('get_page_state'), {
|
|
250
254
|
name: EBrowserTools.GET_PAGE_STATE,
|
|
251
|
-
description: `Get current page URL, title, and interactive elements
|
|
255
|
+
description: `Get the current page state including URL, title, and all interactive elements.
|
|
256
|
+
Use this at the start of a task to see what elements are available.
|
|
257
|
+
Returns a text list of elements with their index numbers for interaction.`,
|
|
252
258
|
schema: BrowserGetPageStateSchema,
|
|
253
259
|
}));
|
|
254
|
-
// browser_select_option
|
|
260
|
+
// browser_select_option - Skyvern-inspired for robust dropdown handling
|
|
255
261
|
tools.push(tool(createToolFunction('select_option'), {
|
|
256
262
|
name: EBrowserTools.SELECT_OPTION,
|
|
257
|
-
description: `Select an option from a dropdown
|
|
263
|
+
description: `Select an option from a dropdown or select element.
|
|
264
|
+
For native <select> elements: finds and selects the option by value/label.
|
|
265
|
+
For custom dropdowns: clicks to open, then clicks the matching option.
|
|
266
|
+
Use this instead of click for dropdowns - it handles both native and custom selects.
|
|
267
|
+
After selection, you receive an updated element list.`,
|
|
258
268
|
schema: BrowserSelectOptionSchema,
|
|
259
269
|
}));
|
|
260
270
|
// browser_upload_file - Skyvern-inspired for file input handling
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"BrowserTools.mjs","sources":["../../../src/tools/BrowserTools.ts"],"sourcesContent":["import { z } from 'zod';\nimport { tool, DynamicStructuredTool } from '@langchain/core/tools';\nimport type * as _t from '@/types';\n\n/**\n * Browser tool names - keep in sync with ranger-browser extension\n * These tools execute locally in the browser extension, NOT on the server\n */\nexport const EBrowserTools = {\n CLICK: 'browser_click',\n TYPE: 'browser_type',\n NAVIGATE: 'browser_navigate',\n SCROLL: 'browser_scroll',\n EXTRACT: 'browser_extract',\n HOVER: 'browser_hover',\n WAIT: 'browser_wait',\n BACK: 'browser_back',\n SCREENSHOT: 'browser_screenshot',\n GET_PAGE_STATE: 'browser_get_page_state',\n // Skyvern-inspired additions for robust form handling\n SELECT_OPTION: 'browser_select_option',\n UPLOAD_FILE: 'browser_upload_file',\n KEYPRESS: 'browser_keypress',\n} as const;\n\nexport type BrowserToolName =\n (typeof EBrowserTools)[keyof typeof EBrowserTools];\n\n/**\n * Callback function type for waiting on browser action results\n * This allows the server (Ranger) to provide a callback that waits for the extension\n * to POST results back to the server before returning to the LLM.\n *\n * @param action - The browser action (click, type, navigate, etc.)\n * @param args - Arguments for the action\n * @param toolCallId - Unique ID for this tool call (from config.toolCall.id)\n * @returns Promise that resolves with the actual browser result (page state, etc.)\n */\nexport type BrowserToolCallback = (\n action: string,\n args: Record<string, unknown>,\n toolCallId: string\n) => Promise<BrowserActionResult>;\n\n/**\n * Action status - matches Skyvern's ActionStatus\n */\nexport type ActionStatus = 'pending' | 'completed' | 'failed' | 'skipped';\n\n/**\n * Verification status - matches Skyvern's VerificationStatus\n * - complete: Goal achieved\n * - terminate: Goal cannot be achieved, stop\n * - continue: Goal not yet achieved, keep going\n */\nexport type VerificationStatus = 'complete' | 'terminate' | 'continue';\n\n/**\n * Result returned from browser action execution\n */\nexport interface BrowserActionResult {\n success: boolean;\n status?: ActionStatus;\n url?: string;\n title?: string;\n elementList?: string;\n error?: string;\n errorType?: string; // Typed error classification\n screenshot?: string;\n elementDescription?: string;\n // State verification (before/after comparison)\n verification?: {\n urlChanged: boolean;\n titleChanged: boolean;\n elementCountDelta: number;\n significantChange: boolean;\n dialogClosed: boolean;\n formGone: boolean;\n changeDescription: string;\n };\n}\n\n/**\n * Check if browser capability is available based on request headers or context\n * The browser extension sets these headers when connected:\n * - X-Ranger-Browser-Extension: true\n * - X-Ranger-Browser-Capable: true\n */\nexport function hasBrowserCapability(req?: {\n headers?: Record<string, string | string[] | undefined>;\n}): boolean {\n if (!req?.headers) {\n return false;\n }\n\n const browserExtension = req.headers['x-ranger-browser-extension'];\n const browserCapable = req.headers['x-ranger-browser-capable'];\n\n return browserExtension === 'true' || browserCapable === 'true';\n}\n\n// Tool schemas\nconst BrowserClickSchema = z.object({\n index: z\n .number()\n .describe(\n 'The index number [0], [1], etc. of the element to click from the page state element list'\n ),\n});\n\nconst BrowserTypeSchema = z.object({\n index: z\n .number()\n .describe('The index number of the input element to type into'),\n text: z.string().describe('The text to type into the element'),\n pressEnter: z\n .boolean()\n .optional()\n .describe('Whether to press Enter after typing (useful for search forms)'),\n});\n\nconst BrowserNavigateSchema = z.object({\n url: z\n .string()\n .describe('The full URL to navigate to (must include https://)'),\n});\n\nconst BrowserScrollSchema = z.object({\n direction: z\n .enum(['up', 'down', 'left', 'right'])\n .describe('Direction to scroll'),\n amount: z\n .number()\n .optional()\n .describe('Pixels to scroll (default: one viewport height)'),\n});\n\nconst BrowserExtractSchema = z.object({\n query: z\n .string()\n .optional()\n .describe('Optional: specific content to extract from the page'),\n});\n\nconst BrowserHoverSchema = z.object({\n index: z.number().describe('The index number of the element to hover over'),\n});\n\nconst BrowserWaitSchema = z.object({\n duration: z\n .number()\n .optional()\n .describe('Milliseconds to wait (default: 1000)'),\n});\n\nconst BrowserBackSchema = z.object({});\n\nconst BrowserScreenshotSchema = z.object({});\n\nconst BrowserGetPageStateSchema = z.object({});\n\n// Skyvern-inspired schemas for robust form handling\nconst BrowserSelectOptionSchema = z.object({\n index: z\n .number()\n .describe('The index number of the select/dropdown element'),\n value: z\n .string()\n .optional()\n .describe('The value or label of the option to select. For native <select>, use the option text. For custom dropdowns, this is the option label to click.'),\n});\n\nconst BrowserUploadFileSchema = z.object({\n index: z\n .number()\n .describe('The index number of the file input element'),\n fileUrl: z\n .string()\n .describe('URL of the file to upload (the system will download and upload it)'),\n});\n\nconst BrowserKeypressSchema = z.object({\n keys: z\n .string()\n .describe('Key(s) to press. Single key: \"Enter\", \"Escape\", \"Tab\", \"ArrowDown\". Combo: \"Control+A\", \"Shift+Enter\"'),\n});\n\n/**\n * Browser tool response interface\n * This is what the extension returns after executing the action\n */\nexport interface BrowserToolResponse {\n requiresBrowserExecution: true;\n action: string;\n args: Record<string, unknown>;\n toolCallId?: string; // Added to help extension correlate with callback\n}\n\n/**\n * Options for creating browser tools\n */\nexport interface CreateBrowserToolsOptions {\n /**\n * Optional callback that waits for browser action results.\n * When provided, tools will await this callback to get actual results from the extension.\n * When not provided, tools return markers immediately (for non-server contexts).\n */\n waitForResult?: BrowserToolCallback;\n}\n\n/**\n * Format browser action result for LLM consumption\n */\nfunction formatResultForLLM(\n result: BrowserActionResult,\n action: string,\n actionArgs?: Record<string, unknown>\n): string {\n if (!result.success && result.error) {\n const errorType = result.errorType ? `[${result.errorType}] ` : '';\n return `${errorType}Action \"${action}\" failed: ${result.error}`;\n }\n\n const parts: string[] = [];\n\n // Verification result (Skyvern pattern)\n if (result.verification) {\n const v = result.verification;\n if (v.dialogClosed || v.formGone || v.urlChanged) {\n parts.push(`✓ ${v.changeDescription}`);\n } else if (!v.significantChange) {\n parts.push(`⚠ No change detected`);\n }\n }\n\n // Page state\n if (result.url) parts.push(`URL: ${result.url}`);\n if (result.title) parts.push(`Title: ${result.title}`);\n if (result.elementList) parts.push(`\\nElements:\\n${result.elementList}`);\n\n return parts.join('\\n');\n}\n\n/**\n * Create browser tools with optional callback for waiting on results\n *\n * When waitForResult callback is provided:\n * 1. Tool returns marker that triggers extension\n * 2. Tool then awaits callback to get actual results\n * 3. Returns real page state to LLM\n *\n * When no callback:\n * 1. Tool returns marker only (for non-server contexts)\n *\n * NOTE: These tools use TEXT-BASED element lists, NOT screenshots\n * Screenshots would be 100K+ tokens each - element lists are ~100 tokens\n */\nexport function createBrowserTools(\n options?: CreateBrowserToolsOptions\n): DynamicStructuredTool[] {\n const { waitForResult } = options || {};\n const tools: DynamicStructuredTool[] = [];\n\n /**\n * Helper to create tool function that optionally waits for results\n * The toolCallId is extracted from the RunnableConfig passed by LangChain\n */\n const createToolFunction = (action: string) => {\n return async (\n args: Record<string, unknown>,\n config?: { toolCall?: { id?: string } }\n ): Promise<string> => {\n const toolCallId =\n config?.toolCall?.id ??\n `tool_${Date.now()}_${Math.random().toString(36).slice(2)}`;\n\n // Create marker for extension\n const marker: BrowserToolResponse = {\n requiresBrowserExecution: true,\n action,\n args,\n toolCallId,\n };\n\n // If no callback, return marker immediately (extension handles via SSE interception)\n if (!waitForResult) {\n return JSON.stringify(marker);\n }\n\n // With callback: wait for actual results from extension\n // The marker is still returned initially via SSE, but we wait for the callback\n try {\n const result = await waitForResult(action, args, toolCallId);\n return formatResultForLLM(result, action, args);\n } catch (error) {\n const errorMessage =\n error instanceof Error ? error.message : String(error);\n return `❌ **ACTION FAILED**: Browser action \"${action}\" failed: ${errorMessage}\n\n**REQUIRED**: Analyze why the action failed and try an alternative approach.`;\n }\n };\n };\n\n // browser_click\n tools.push(\n tool(createToolFunction('click'), {\n name: EBrowserTools.CLICK,\n description: `Click an element by its index number from the element list (e.g., [0], [1]).\nReturns state verification showing if UI changed (dialog closed, form submitted, etc.).`,\n schema: BrowserClickSchema,\n })\n );\n\n // browser_type\n tools.push(\n tool(createToolFunction('type'), {\n name: EBrowserTools.TYPE,\n description: `Type text into an input element by index. Set pressEnter: true to submit.\nReturns state verification showing if form submitted or errors appeared.`,\n schema: BrowserTypeSchema,\n })\n );\n\n // browser_navigate\n tools.push(\n tool(createToolFunction('navigate'), {\n name: EBrowserTools.NAVIGATE,\n description: `Navigate to a URL (include https://). Returns new page's element list.`,\n schema: BrowserNavigateSchema,\n })\n );\n\n // browser_scroll\n tools.push(\n tool(createToolFunction('scroll'), {\n name: EBrowserTools.SCROLL,\n description: `Scroll the page. Returns updated element list with newly visible elements.`,\n schema: BrowserScrollSchema,\n })\n );\n\n // browser_extract\n tools.push(\n tool(createToolFunction('extract'), {\n name: EBrowserTools.EXTRACT,\n description: `Extract content from the current page.\nReturns page URL, title, and element list.`,\n schema: BrowserExtractSchema,\n })\n );\n\n // browser_hover\n tools.push(\n tool(createToolFunction('hover'), {\n name: EBrowserTools.HOVER,\n description: `Hover over an element to reveal tooltips, dropdowns, or other hover-triggered content.\nAfter hovering, you receive an updated element list with any newly revealed elements.`,\n schema: BrowserHoverSchema,\n })\n );\n\n // browser_wait\n tools.push(\n tool(createToolFunction('wait'), {\n name: EBrowserTools.WAIT,\n description: `Wait for a specified duration for page content to load.\nUse this after actions that trigger async content loading.\nAfter waiting, you receive an updated element list.`,\n schema: BrowserWaitSchema,\n })\n );\n\n // browser_back\n tools.push(\n tool(createToolFunction('back'), {\n name: EBrowserTools.BACK,\n description: `Go back to the previous page in browser history.\nAfter going back, you receive the previous page's element list.`,\n schema: BrowserBackSchema,\n })\n );\n\n // browser_screenshot\n tools.push(\n tool(createToolFunction('screenshot'), {\n name: EBrowserTools.SCREENSHOT,\n description: `Capture a screenshot. Use browser_get_page_state for element list.`,\n schema: BrowserScreenshotSchema,\n })\n );\n\n // browser_get_page_state\n tools.push(\n tool(createToolFunction('get_page_state'), {\n name: EBrowserTools.GET_PAGE_STATE,\n description: `Get current page URL, title, and interactive elements list with index numbers.`,\n schema: BrowserGetPageStateSchema,\n })\n );\n\n // browser_select_option\n tools.push(\n tool(createToolFunction('select_option'), {\n name: EBrowserTools.SELECT_OPTION,\n description: `Select an option from a dropdown. Handles both native <select> and custom dropdowns.`,\n schema: BrowserSelectOptionSchema,\n })\n );\n\n // browser_upload_file - Skyvern-inspired for file input handling\n tools.push(\n tool(createToolFunction('upload_file'), {\n name: EBrowserTools.UPLOAD_FILE,\n description: `Upload a file to a file input element.\nProvide the index of the file input and the URL of the file to upload.\nThe system will download the file and attach it to the input.\nAfter upload, you receive an updated element list.`,\n schema: BrowserUploadFileSchema,\n })\n );\n\n // browser_keypress - For keyboard shortcuts and special keys\n tools.push(\n tool(createToolFunction('keypress'), {\n name: EBrowserTools.KEYPRESS,\n description: `Press keyboard key(s) on the page.\nSingle keys: \"Enter\", \"Escape\", \"Tab\", \"ArrowDown\", \"ArrowUp\", \"Backspace\", \"Delete\"\nKey combos: \"Control+A\" (select all), \"Control+C\" (copy), \"Shift+Enter\" (newline)\nUse this for form submission, closing modals, navigating dropdowns.\nAfter keypress, you receive an updated element list.`,\n schema: BrowserKeypressSchema,\n })\n );\n\n return tools;\n}\n"],"names":[],"mappings":";;;AAIA;;;AAGG;AACU,MAAA,aAAa,GAAG;AAC3B,IAAA,KAAK,EAAE,eAAe;AACtB,IAAA,IAAI,EAAE,cAAc;AACpB,IAAA,QAAQ,EAAE,kBAAkB;AAC5B,IAAA,MAAM,EAAE,gBAAgB;AACxB,IAAA,OAAO,EAAE,iBAAiB;AAC1B,IAAA,KAAK,EAAE,eAAe;AACtB,IAAA,IAAI,EAAE,cAAc;AACpB,IAAA,IAAI,EAAE,cAAc;AACpB,IAAA,UAAU,EAAE,oBAAoB;AAChC,IAAA,cAAc,EAAE,wBAAwB;;AAExC,IAAA,aAAa,EAAE,uBAAuB;AACtC,IAAA,WAAW,EAAE,qBAAqB;AAClC,IAAA,QAAQ,EAAE,kBAAkB;;AA4D9B;;;;;AAKG;AACG,SAAU,oBAAoB,CAAC,GAEpC,EAAA;AACC,IAAA,IAAI,CAAC,GAAG,EAAE,OAAO,EAAE;AACjB,QAAA,OAAO,KAAK;;IAGd,MAAM,gBAAgB,GAAG,GAAG,CAAC,OAAO,CAAC,4BAA4B,CAAC;IAClE,MAAM,cAAc,GAAG,GAAG,CAAC,OAAO,CAAC,0BAA0B,CAAC;AAE9D,IAAA,OAAO,gBAAgB,KAAK,MAAM,IAAI,cAAc,KAAK,MAAM;AACjE;AAEA;AACA,MAAM,kBAAkB,GAAG,CAAC,CAAC,MAAM,CAAC;AAClC,IAAA,KAAK,EAAE;AACJ,SAAA,MAAM;SACN,QAAQ,CACP,0FAA0F,CAC3F;AACJ,CAAA,CAAC;AAEF,MAAM,iBAAiB,GAAG,CAAC,CAAC,MAAM,CAAC;AACjC,IAAA,KAAK,EAAE;AACJ,SAAA,MAAM;SACN,QAAQ,CAAC,oDAAoD,CAAC;IACjE,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,mCAAmC,CAAC;AAC9D,IAAA,UAAU,EAAE;AACT,SAAA,OAAO;AACP,SAAA,QAAQ;SACR,QAAQ,CAAC,+DAA+D,CAAC;AAC7E,CAAA,CAAC;AAEF,MAAM,qBAAqB,GAAG,CAAC,CAAC,MAAM,CAAC;AACrC,IAAA,GAAG,EAAE;AACF,SAAA,MAAM;SACN,QAAQ,CAAC,qDAAqD,CAAC;AACnE,CAAA,CAAC;AAEF,MAAM,mBAAmB,GAAG,CAAC,CAAC,MAAM,CAAC;AACnC,IAAA,SAAS,EAAE;SACR,IAAI,CAAC,CAAC,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,OAAO,CAAC;SACpC,QAAQ,CAAC,qBAAqB,CAAC;AAClC,IAAA,MAAM,EAAE;AACL,SAAA,MAAM;AACN,SAAA,QAAQ;SACR,QAAQ,CAAC,iDAAiD,CAAC;AAC/D,CAAA,CAAC;AAEF,MAAM,oBAAoB,GAAG,CAAC,CAAC,MAAM,CAAC;AACpC,IAAA,KAAK,EAAE;AACJ,SAAA,MAAM;AACN,SAAA,QAAQ;SACR,QAAQ,CAAC,qDAAqD,CAAC;AACnE,CAAA,CAAC;AAEF,MAAM,kBAAkB,GAAG,CAAC,CAAC,MAAM,CAAC;IAClC,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,+CAA+C,CAAC;AAC5E,CAAA,CAAC;AAEF,MAAM,iBAAiB,GAAG,CAAC,CAAC,MAAM,CAAC;AACjC,IAAA,QAAQ,EAAE;AACP,SAAA,MAAM;AACN,SAAA,QAAQ;SACR,QAAQ,CAAC,sCAAsC,CAAC;AACpD,CAAA,CAAC;AAEF,MAAM,iBAAiB,GAAG,CAAC,CAAC,MAAM,CAAC,EAAE,CAAC;AAEtC,MAAM,uBAAuB,GAAG,CAAC,CAAC,MAAM,CAAC,EAAE,CAAC;AAE5C,MAAM,yBAAyB,GAAG,CAAC,CAAC,MAAM,CAAC,EAAE,CAAC;AAE9C;AACA,MAAM,yBAAyB,GAAG,CAAC,CAAC,MAAM,CAAC;AACzC,IAAA,KAAK,EAAE;AACJ,SAAA,MAAM;SACN,QAAQ,CAAC,iDAAiD,CAAC;AAC9D,IAAA,KAAK,EAAE;AACJ,SAAA,MAAM;AACN,SAAA,QAAQ;SACR,QAAQ,CAAC,gJAAgJ,CAAC;AAC9J,CAAA,CAAC;AAEF,MAAM,uBAAuB,GAAG,CAAC,CAAC,MAAM,CAAC;AACvC,IAAA,KAAK,EAAE;AACJ,SAAA,MAAM;SACN,QAAQ,CAAC,4CAA4C,CAAC;AACzD,IAAA,OAAO,EAAE;AACN,SAAA,MAAM;SACN,QAAQ,CAAC,oEAAoE,CAAC;AAClF,CAAA,CAAC;AAEF,MAAM,qBAAqB,GAAG,CAAC,CAAC,MAAM,CAAC;AACrC,IAAA,IAAI,EAAE;AACH,SAAA,MAAM;SACN,QAAQ,CAAC,uGAAuG,CAAC;AACrH,CAAA,CAAC;AAyBF;;AAEG;AACH,SAAS,kBAAkB,CACzB,MAA2B,EAC3B,MAAc,EACd,UAAoC,EAAA;IAEpC,IAAI,CAAC,MAAM,CAAC,OAAO,IAAI,MAAM,CAAC,KAAK,EAAE;AACnC,QAAA,MAAM,SAAS,GAAG,MAAM,CAAC,SAAS,GAAG,CAAI,CAAA,EAAA,MAAM,CAAC,SAAS,CAAA,EAAA,CAAI,GAAG,EAAE;QAClE,OAAO,CAAA,EAAG,SAAS,CAAW,QAAA,EAAA,MAAM,aAAa,MAAM,CAAC,KAAK,CAAA,CAAE;;IAGjE,MAAM,KAAK,GAAa,EAAE;;AAG1B,IAAA,IAAI,MAAM,CAAC,YAAY,EAAE;AACvB,QAAA,MAAM,CAAC,GAAG,MAAM,CAAC,YAAY;AAC7B,QAAA,IAAI,CAAC,CAAC,YAAY,IAAI,CAAC,CAAC,QAAQ,IAAI,CAAC,CAAC,UAAU,EAAE;YAChD,KAAK,CAAC,IAAI,CAAC,CAAA,EAAA,EAAK,CAAC,CAAC,iBAAiB,CAAE,CAAA,CAAC;;AACjC,aAAA,IAAI,CAAC,CAAC,CAAC,iBAAiB,EAAE;AAC/B,YAAA,KAAK,CAAC,IAAI,CAAC,CAAA,oBAAA,CAAsB,CAAC;;;;IAKtC,IAAI,MAAM,CAAC,GAAG;QAAE,KAAK,CAAC,IAAI,CAAC,CAAA,KAAA,EAAQ,MAAM,CAAC,GAAG,CAAE,CAAA,CAAC;IAChD,IAAI,MAAM,CAAC,KAAK;QAAE,KAAK,CAAC,IAAI,CAAC,CAAA,OAAA,EAAU,MAAM,CAAC,KAAK,CAAE,CAAA,CAAC;IACtD,IAAI,MAAM,CAAC,WAAW;QAAE,KAAK,CAAC,IAAI,CAAC,CAAA,aAAA,EAAgB,MAAM,CAAC,WAAW,CAAE,CAAA,CAAC;AAExE,IAAA,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC;AACzB;AAEA;;;;;;;;;;;;;AAaG;AACG,SAAU,kBAAkB,CAChC,OAAmC,EAAA;AAEnC,IAAA,MAAM,EAAE,aAAa,EAAE,GAAG,OAAO,IAAI,EAAE;IACvC,MAAM,KAAK,GAA4B,EAAE;AAEzC;;;AAGG;AACH,IAAA,MAAM,kBAAkB,GAAG,CAAC,MAAc,KAAI;AAC5C,QAAA,OAAO,OACL,IAA6B,EAC7B,MAAuC,KACpB;AACnB,YAAA,MAAM,UAAU,GACd,MAAM,EAAE,QAAQ,EAAE,EAAE;gBACpB,CAAQ,KAAA,EAAA,IAAI,CAAC,GAAG,EAAE,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE;;AAG7D,YAAA,MAAM,MAAM,GAAwB;AAClC,gBAAA,wBAAwB,EAAE,IAAI;gBAC9B,MAAM;gBACN,IAAI;gBACJ,UAAU;aACX;;YAGD,IAAI,CAAC,aAAa,EAAE;AAClB,gBAAA,OAAO,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC;;;;AAK/B,YAAA,IAAI;gBACF,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC,MAAM,EAAE,IAAI,EAAE,UAAU,CAAC;gBAC5D,OAAO,kBAAkB,CAAC,MAAM,EAAE,MAAM,EAAE,IAAI,CAAC;;YAC/C,OAAO,KAAK,EAAE;AACd,gBAAA,MAAM,YAAY,GAChB,KAAK,YAAY,KAAK,GAAG,KAAK,CAAC,OAAO,GAAG,MAAM,CAAC,KAAK,CAAC;gBACxD,OAAO,CAAA,qCAAA,EAAwC,MAAM,CAAA,UAAA,EAAa,YAAY;;6EAET;;AAEzE,SAAC;AACH,KAAC;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,OAAO,CAAC,EAAE;QAChC,IAAI,EAAE,aAAa,CAAC,KAAK;AACzB,QAAA,WAAW,EAAE,CAAA;AACqE,uFAAA,CAAA;AAClF,QAAA,MAAM,EAAE,kBAAkB;AAC3B,KAAA,CAAC,CACH;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,MAAM,CAAC,EAAE;QAC/B,IAAI,EAAE,aAAa,CAAC,IAAI;AACxB,QAAA,WAAW,EAAE,CAAA;AACsD,wEAAA,CAAA;AACnE,QAAA,MAAM,EAAE,iBAAiB;AAC1B,KAAA,CAAC,CACH;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,UAAU,CAAC,EAAE;QACnC,IAAI,EAAE,aAAa,CAAC,QAAQ;AAC5B,QAAA,WAAW,EAAE,CAAwE,sEAAA,CAAA;AACrF,QAAA,MAAM,EAAE,qBAAqB;AAC9B,KAAA,CAAC,CACH;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,QAAQ,CAAC,EAAE;QACjC,IAAI,EAAE,aAAa,CAAC,MAAM;AAC1B,QAAA,WAAW,EAAE,CAA4E,0EAAA,CAAA;AACzF,QAAA,MAAM,EAAE,mBAAmB;AAC5B,KAAA,CAAC,CACH;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,SAAS,CAAC,EAAE;QAClC,IAAI,EAAE,aAAa,CAAC,OAAO;AAC3B,QAAA,WAAW,EAAE,CAAA;AACwB,0CAAA,CAAA;AACrC,QAAA,MAAM,EAAE,oBAAoB;AAC7B,KAAA,CAAC,CACH;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,OAAO,CAAC,EAAE;QAChC,IAAI,EAAE,aAAa,CAAC,KAAK;AACzB,QAAA,WAAW,EAAE,CAAA;AACmE,qFAAA,CAAA;AAChF,QAAA,MAAM,EAAE,kBAAkB;AAC3B,KAAA,CAAC,CACH;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,MAAM,CAAC,EAAE;QAC/B,IAAI,EAAE,aAAa,CAAC,IAAI;AACxB,QAAA,WAAW,EAAE,CAAA;;AAEiC,mDAAA,CAAA;AAC9C,QAAA,MAAM,EAAE,iBAAiB;AAC1B,KAAA,CAAC,CACH;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,MAAM,CAAC,EAAE;QAC/B,IAAI,EAAE,aAAa,CAAC,IAAI;AACxB,QAAA,WAAW,EAAE,CAAA;AAC6C,+DAAA,CAAA;AAC1D,QAAA,MAAM,EAAE,iBAAiB;AAC1B,KAAA,CAAC,CACH;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,YAAY,CAAC,EAAE;QACrC,IAAI,EAAE,aAAa,CAAC,UAAU;AAC9B,QAAA,WAAW,EAAE,CAAoE,kEAAA,CAAA;AACjF,QAAA,MAAM,EAAE,uBAAuB;AAChC,KAAA,CAAC,CACH;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,gBAAgB,CAAC,EAAE;QACzC,IAAI,EAAE,aAAa,CAAC,cAAc;AAClC,QAAA,WAAW,EAAE,CAAgF,8EAAA,CAAA;AAC7F,QAAA,MAAM,EAAE,yBAAyB;AAClC,KAAA,CAAC,CACH;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,eAAe,CAAC,EAAE;QACxC,IAAI,EAAE,aAAa,CAAC,aAAa;AACjC,QAAA,WAAW,EAAE,CAAsF,oFAAA,CAAA;AACnG,QAAA,MAAM,EAAE,yBAAyB;AAClC,KAAA,CAAC,CACH;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,aAAa,CAAC,EAAE;QACtC,IAAI,EAAE,aAAa,CAAC,WAAW;AAC/B,QAAA,WAAW,EAAE,CAAA;;;AAGgC,kDAAA,CAAA;AAC7C,QAAA,MAAM,EAAE,uBAAuB;AAChC,KAAA,CAAC,CACH;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,UAAU,CAAC,EAAE;QACnC,IAAI,EAAE,aAAa,CAAC,QAAQ;AAC5B,QAAA,WAAW,EAAE,CAAA;;;;AAIkC,oDAAA,CAAA;AAC/C,QAAA,MAAM,EAAE,qBAAqB;AAC9B,KAAA,CAAC,CACH;AAED,IAAA,OAAO,KAAK;AACd;;;;"}
|
|
1
|
+
{"version":3,"file":"BrowserTools.mjs","sources":["../../../src/tools/BrowserTools.ts"],"sourcesContent":["import { z } from 'zod';\nimport { tool, DynamicStructuredTool } from '@langchain/core/tools';\nimport type * as _t from '@/types';\n\n/**\n * Browser tool names - keep in sync with ranger-browser extension\n * These tools execute locally in the browser extension, NOT on the server\n */\nexport const EBrowserTools = {\n CLICK: 'browser_click',\n TYPE: 'browser_type',\n NAVIGATE: 'browser_navigate',\n SCROLL: 'browser_scroll',\n EXTRACT: 'browser_extract',\n HOVER: 'browser_hover',\n WAIT: 'browser_wait',\n BACK: 'browser_back',\n SCREENSHOT: 'browser_screenshot',\n GET_PAGE_STATE: 'browser_get_page_state',\n // Skyvern-inspired additions for robust form handling\n SELECT_OPTION: 'browser_select_option',\n UPLOAD_FILE: 'browser_upload_file',\n KEYPRESS: 'browser_keypress',\n} as const;\n\nexport type BrowserToolName =\n (typeof EBrowserTools)[keyof typeof EBrowserTools];\n\n/**\n * Callback function type for waiting on browser action results\n * This allows the server (Ranger) to provide a callback that waits for the extension\n * to POST results back to the server before returning to the LLM.\n *\n * @param action - The browser action (click, type, navigate, etc.)\n * @param args - Arguments for the action\n * @param toolCallId - Unique ID for this tool call (from config.toolCall.id)\n * @returns Promise that resolves with the actual browser result (page state, etc.)\n */\nexport type BrowserToolCallback = (\n action: string,\n args: Record<string, unknown>,\n toolCallId: string\n) => Promise<BrowserActionResult>;\n\n/**\n * Result returned from browser action execution\n */\nexport interface BrowserActionResult {\n success: boolean;\n url?: string;\n title?: string;\n elementList?: string; // Text-based element list\n error?: string;\n screenshot?: string; // Base64 screenshot (if requested)\n}\n\n/**\n * Check if browser capability is available based on request headers or context\n * The browser extension sets these headers when connected:\n * - X-Ranger-Browser-Extension: true\n * - X-Ranger-Browser-Capable: true\n */\nexport function hasBrowserCapability(req?: {\n headers?: Record<string, string | string[] | undefined>;\n}): boolean {\n if (!req?.headers) {\n return false;\n }\n\n const browserExtension = req.headers['x-ranger-browser-extension'];\n const browserCapable = req.headers['x-ranger-browser-capable'];\n\n return browserExtension === 'true' || browserCapable === 'true';\n}\n\n// Tool schemas\nconst BrowserClickSchema = z.object({\n index: z\n .number()\n .describe(\n 'The index number [0], [1], etc. of the element to click from the page state element list'\n ),\n});\n\nconst BrowserTypeSchema = z.object({\n index: z\n .number()\n .describe('The index number of the input element to type into'),\n text: z.string().describe('The text to type into the element'),\n pressEnter: z\n .boolean()\n .optional()\n .describe('Whether to press Enter after typing (useful for search forms)'),\n});\n\nconst BrowserNavigateSchema = z.object({\n url: z\n .string()\n .describe('The full URL to navigate to (must include https://)'),\n});\n\nconst BrowserScrollSchema = z.object({\n direction: z\n .enum(['up', 'down', 'left', 'right'])\n .describe('Direction to scroll'),\n amount: z\n .number()\n .optional()\n .describe('Pixels to scroll (default: one viewport height)'),\n});\n\nconst BrowserExtractSchema = z.object({\n query: z\n .string()\n .optional()\n .describe('Optional: specific content to extract from the page'),\n});\n\nconst BrowserHoverSchema = z.object({\n index: z.number().describe('The index number of the element to hover over'),\n});\n\nconst BrowserWaitSchema = z.object({\n duration: z\n .number()\n .optional()\n .describe('Milliseconds to wait (default: 1000)'),\n});\n\nconst BrowserBackSchema = z.object({});\n\nconst BrowserScreenshotSchema = z.object({});\n\nconst BrowserGetPageStateSchema = z.object({});\n\n// Skyvern-inspired schemas for robust form handling\nconst BrowserSelectOptionSchema = z.object({\n index: z\n .number()\n .describe('The index number of the select/dropdown element'),\n value: z\n .string()\n .optional()\n .describe('The value or label of the option to select. For native <select>, use the option text. For custom dropdowns, this is the option label to click.'),\n});\n\nconst BrowserUploadFileSchema = z.object({\n index: z\n .number()\n .describe('The index number of the file input element'),\n fileUrl: z\n .string()\n .describe('URL of the file to upload (the system will download and upload it)'),\n});\n\nconst BrowserKeypressSchema = z.object({\n keys: z\n .string()\n .describe('Key(s) to press. Single key: \"Enter\", \"Escape\", \"Tab\", \"ArrowDown\". Combo: \"Control+A\", \"Shift+Enter\"'),\n});\n\n/**\n * Browser tool response interface\n * This is what the extension returns after executing the action\n */\nexport interface BrowserToolResponse {\n requiresBrowserExecution: true;\n action: string;\n args: Record<string, unknown>;\n toolCallId?: string; // Added to help extension correlate with callback\n}\n\n/**\n * Options for creating browser tools\n */\nexport interface CreateBrowserToolsOptions {\n /**\n * Optional callback that waits for browser action results.\n * When provided, tools will await this callback to get actual results from the extension.\n * When not provided, tools return markers immediately (for non-server contexts).\n */\n waitForResult?: BrowserToolCallback;\n}\n\n/**\n * Format browser action result for LLM consumption\n */\nfunction formatResultForLLM(\n result: BrowserActionResult,\n action: string\n): string {\n if (!result.success && result.error) {\n return `Browser action \"${action}\" failed: ${result.error}`;\n }\n\n const parts: string[] = [];\n\n if (result.url != null && result.url !== '') {\n parts.push(`**Current URL:** ${result.url}`);\n }\n if (result.title != null && result.title !== '') {\n parts.push(`**Page Title:** ${result.title}`);\n }\n if (result.elementList != null && result.elementList !== '') {\n parts.push(`\\n**Interactive Elements:**\\n${result.elementList}`);\n }\n if (result.screenshot != null && result.screenshot !== '') {\n parts.push('\\n[Screenshot captured and displayed to user]');\n }\n\n if (parts.length === 0) {\n return `Browser action \"${action}\" completed successfully.`;\n }\n\n return parts.join('\\n');\n}\n\n/**\n * Create browser tools with optional callback for waiting on results\n *\n * When waitForResult callback is provided:\n * 1. Tool returns marker that triggers extension\n * 2. Tool then awaits callback to get actual results\n * 3. Returns real page state to LLM\n *\n * When no callback:\n * 1. Tool returns marker only (for non-server contexts)\n *\n * NOTE: These tools use TEXT-BASED element lists, NOT screenshots\n * Screenshots would be 100K+ tokens each - element lists are ~100 tokens\n */\nexport function createBrowserTools(\n options?: CreateBrowserToolsOptions\n): DynamicStructuredTool[] {\n const { waitForResult } = options || {};\n const tools: DynamicStructuredTool[] = [];\n\n /**\n * Helper to create tool function that optionally waits for results\n * The toolCallId is extracted from the RunnableConfig passed by LangChain\n */\n const createToolFunction = (action: string) => {\n return async (\n args: Record<string, unknown>,\n config?: { toolCall?: { id?: string } }\n ): Promise<string> => {\n const toolCallId =\n config?.toolCall?.id ??\n `tool_${Date.now()}_${Math.random().toString(36).slice(2)}`;\n\n // Create marker for extension\n const marker: BrowserToolResponse = {\n requiresBrowserExecution: true,\n action,\n args,\n toolCallId,\n };\n\n // If no callback, return marker immediately (extension handles via SSE interception)\n if (!waitForResult) {\n return JSON.stringify(marker);\n }\n\n // With callback: wait for actual results from extension\n // The marker is still returned initially via SSE, but we wait for the callback\n try {\n const result = await waitForResult(action, args, toolCallId);\n return formatResultForLLM(result, action);\n } catch (error) {\n const errorMessage =\n error instanceof Error ? error.message : String(error);\n return `Browser action \"${action}\" failed: ${errorMessage}`;\n }\n };\n };\n\n // browser_click\n tools.push(\n tool(createToolFunction('click'), {\n name: EBrowserTools.CLICK,\n description: `Click an element on the current web page by its index number.\nThe element list shows clickable items like: [0]<button>Submit</button> [1]<a href=\"/home\">Home</a>\nUse the index number in brackets to click that element.\nAfter clicking, you receive an updated element list showing the new page state.`,\n schema: BrowserClickSchema,\n })\n );\n\n // browser_type\n tools.push(\n tool(createToolFunction('type'), {\n name: EBrowserTools.TYPE,\n description: `Type text into an input element on the page.\nFind the input element in the list by its index (e.g., [5]<input placeholder=\"Search\">).\nSet pressEnter: true to submit forms after typing.\nAfter typing, you receive an updated element list.`,\n schema: BrowserTypeSchema,\n })\n );\n\n // browser_navigate\n tools.push(\n tool(createToolFunction('navigate'), {\n name: EBrowserTools.NAVIGATE,\n description: `Navigate to a URL. Always include the full URL with https://.\nAfter navigation, you receive the new page's element list.`,\n schema: BrowserNavigateSchema,\n })\n );\n\n // browser_scroll\n tools.push(\n tool(createToolFunction('scroll'), {\n name: EBrowserTools.SCROLL,\n description: `Scroll the page to reveal more content.\nUse 'down' to scroll down, 'up' to scroll up.\nAfter scrolling, you receive an updated element list with newly visible elements.`,\n schema: BrowserScrollSchema,\n })\n );\n\n // browser_extract\n tools.push(\n tool(createToolFunction('extract'), {\n name: EBrowserTools.EXTRACT,\n description: `Extract content from the current page.\nReturns page URL, title, and element list.`,\n schema: BrowserExtractSchema,\n })\n );\n\n // browser_hover\n tools.push(\n tool(createToolFunction('hover'), {\n name: EBrowserTools.HOVER,\n description: `Hover over an element to reveal tooltips, dropdowns, or other hover-triggered content.\nAfter hovering, you receive an updated element list with any newly revealed elements.`,\n schema: BrowserHoverSchema,\n })\n );\n\n // browser_wait\n tools.push(\n tool(createToolFunction('wait'), {\n name: EBrowserTools.WAIT,\n description: `Wait for a specified duration for page content to load.\nUse this after actions that trigger async content loading.\nAfter waiting, you receive an updated element list.`,\n schema: BrowserWaitSchema,\n })\n );\n\n // browser_back\n tools.push(\n tool(createToolFunction('back'), {\n name: EBrowserTools.BACK,\n description: `Go back to the previous page in browser history.\nAfter going back, you receive the previous page's element list.`,\n schema: BrowserBackSchema,\n })\n );\n\n // browser_screenshot\n tools.push(\n tool(createToolFunction('screenshot'), {\n name: EBrowserTools.SCREENSHOT,\n description: `Capture a screenshot of the current page.\nReturns the page state with a note that screenshot was displayed to the user.\nUse browser_get_page_state to get the element list for automation.`,\n schema: BrowserScreenshotSchema,\n })\n );\n\n // browser_get_page_state\n tools.push(\n tool(createToolFunction('get_page_state'), {\n name: EBrowserTools.GET_PAGE_STATE,\n description: `Get the current page state including URL, title, and all interactive elements.\nUse this at the start of a task to see what elements are available.\nReturns a text list of elements with their index numbers for interaction.`,\n schema: BrowserGetPageStateSchema,\n })\n );\n\n // browser_select_option - Skyvern-inspired for robust dropdown handling\n tools.push(\n tool(createToolFunction('select_option'), {\n name: EBrowserTools.SELECT_OPTION,\n description: `Select an option from a dropdown or select element.\nFor native <select> elements: finds and selects the option by value/label.\nFor custom dropdowns: clicks to open, then clicks the matching option.\nUse this instead of click for dropdowns - it handles both native and custom selects.\nAfter selection, you receive an updated element list.`,\n schema: BrowserSelectOptionSchema,\n })\n );\n\n // browser_upload_file - Skyvern-inspired for file input handling\n tools.push(\n tool(createToolFunction('upload_file'), {\n name: EBrowserTools.UPLOAD_FILE,\n description: `Upload a file to a file input element.\nProvide the index of the file input and the URL of the file to upload.\nThe system will download the file and attach it to the input.\nAfter upload, you receive an updated element list.`,\n schema: BrowserUploadFileSchema,\n })\n );\n\n // browser_keypress - For keyboard shortcuts and special keys\n tools.push(\n tool(createToolFunction('keypress'), {\n name: EBrowserTools.KEYPRESS,\n description: `Press keyboard key(s) on the page.\nSingle keys: \"Enter\", \"Escape\", \"Tab\", \"ArrowDown\", \"ArrowUp\", \"Backspace\", \"Delete\"\nKey combos: \"Control+A\" (select all), \"Control+C\" (copy), \"Shift+Enter\" (newline)\nUse this for form submission, closing modals, navigating dropdowns.\nAfter keypress, you receive an updated element list.`,\n schema: BrowserKeypressSchema,\n })\n );\n\n return tools;\n}\n"],"names":[],"mappings":";;;AAIA;;;AAGG;AACU,MAAA,aAAa,GAAG;AAC3B,IAAA,KAAK,EAAE,eAAe;AACtB,IAAA,IAAI,EAAE,cAAc;AACpB,IAAA,QAAQ,EAAE,kBAAkB;AAC5B,IAAA,MAAM,EAAE,gBAAgB;AACxB,IAAA,OAAO,EAAE,iBAAiB;AAC1B,IAAA,KAAK,EAAE,eAAe;AACtB,IAAA,IAAI,EAAE,cAAc;AACpB,IAAA,IAAI,EAAE,cAAc;AACpB,IAAA,UAAU,EAAE,oBAAoB;AAChC,IAAA,cAAc,EAAE,wBAAwB;;AAExC,IAAA,aAAa,EAAE,uBAAuB;AACtC,IAAA,WAAW,EAAE,qBAAqB;AAClC,IAAA,QAAQ,EAAE,kBAAkB;;AAkC9B;;;;;AAKG;AACG,SAAU,oBAAoB,CAAC,GAEpC,EAAA;AACC,IAAA,IAAI,CAAC,GAAG,EAAE,OAAO,EAAE;AACjB,QAAA,OAAO,KAAK;;IAGd,MAAM,gBAAgB,GAAG,GAAG,CAAC,OAAO,CAAC,4BAA4B,CAAC;IAClE,MAAM,cAAc,GAAG,GAAG,CAAC,OAAO,CAAC,0BAA0B,CAAC;AAE9D,IAAA,OAAO,gBAAgB,KAAK,MAAM,IAAI,cAAc,KAAK,MAAM;AACjE;AAEA;AACA,MAAM,kBAAkB,GAAG,CAAC,CAAC,MAAM,CAAC;AAClC,IAAA,KAAK,EAAE;AACJ,SAAA,MAAM;SACN,QAAQ,CACP,0FAA0F,CAC3F;AACJ,CAAA,CAAC;AAEF,MAAM,iBAAiB,GAAG,CAAC,CAAC,MAAM,CAAC;AACjC,IAAA,KAAK,EAAE;AACJ,SAAA,MAAM;SACN,QAAQ,CAAC,oDAAoD,CAAC;IACjE,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,mCAAmC,CAAC;AAC9D,IAAA,UAAU,EAAE;AACT,SAAA,OAAO;AACP,SAAA,QAAQ;SACR,QAAQ,CAAC,+DAA+D,CAAC;AAC7E,CAAA,CAAC;AAEF,MAAM,qBAAqB,GAAG,CAAC,CAAC,MAAM,CAAC;AACrC,IAAA,GAAG,EAAE;AACF,SAAA,MAAM;SACN,QAAQ,CAAC,qDAAqD,CAAC;AACnE,CAAA,CAAC;AAEF,MAAM,mBAAmB,GAAG,CAAC,CAAC,MAAM,CAAC;AACnC,IAAA,SAAS,EAAE;SACR,IAAI,CAAC,CAAC,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,OAAO,CAAC;SACpC,QAAQ,CAAC,qBAAqB,CAAC;AAClC,IAAA,MAAM,EAAE;AACL,SAAA,MAAM;AACN,SAAA,QAAQ;SACR,QAAQ,CAAC,iDAAiD,CAAC;AAC/D,CAAA,CAAC;AAEF,MAAM,oBAAoB,GAAG,CAAC,CAAC,MAAM,CAAC;AACpC,IAAA,KAAK,EAAE;AACJ,SAAA,MAAM;AACN,SAAA,QAAQ;SACR,QAAQ,CAAC,qDAAqD,CAAC;AACnE,CAAA,CAAC;AAEF,MAAM,kBAAkB,GAAG,CAAC,CAAC,MAAM,CAAC;IAClC,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,+CAA+C,CAAC;AAC5E,CAAA,CAAC;AAEF,MAAM,iBAAiB,GAAG,CAAC,CAAC,MAAM,CAAC;AACjC,IAAA,QAAQ,EAAE;AACP,SAAA,MAAM;AACN,SAAA,QAAQ;SACR,QAAQ,CAAC,sCAAsC,CAAC;AACpD,CAAA,CAAC;AAEF,MAAM,iBAAiB,GAAG,CAAC,CAAC,MAAM,CAAC,EAAE,CAAC;AAEtC,MAAM,uBAAuB,GAAG,CAAC,CAAC,MAAM,CAAC,EAAE,CAAC;AAE5C,MAAM,yBAAyB,GAAG,CAAC,CAAC,MAAM,CAAC,EAAE,CAAC;AAE9C;AACA,MAAM,yBAAyB,GAAG,CAAC,CAAC,MAAM,CAAC;AACzC,IAAA,KAAK,EAAE;AACJ,SAAA,MAAM;SACN,QAAQ,CAAC,iDAAiD,CAAC;AAC9D,IAAA,KAAK,EAAE;AACJ,SAAA,MAAM;AACN,SAAA,QAAQ;SACR,QAAQ,CAAC,gJAAgJ,CAAC;AAC9J,CAAA,CAAC;AAEF,MAAM,uBAAuB,GAAG,CAAC,CAAC,MAAM,CAAC;AACvC,IAAA,KAAK,EAAE;AACJ,SAAA,MAAM;SACN,QAAQ,CAAC,4CAA4C,CAAC;AACzD,IAAA,OAAO,EAAE;AACN,SAAA,MAAM;SACN,QAAQ,CAAC,oEAAoE,CAAC;AAClF,CAAA,CAAC;AAEF,MAAM,qBAAqB,GAAG,CAAC,CAAC,MAAM,CAAC;AACrC,IAAA,IAAI,EAAE;AACH,SAAA,MAAM;SACN,QAAQ,CAAC,uGAAuG,CAAC;AACrH,CAAA,CAAC;AAyBF;;AAEG;AACH,SAAS,kBAAkB,CACzB,MAA2B,EAC3B,MAAc,EAAA;IAEd,IAAI,CAAC,MAAM,CAAC,OAAO,IAAI,MAAM,CAAC,KAAK,EAAE;AACnC,QAAA,OAAO,mBAAmB,MAAM,CAAA,UAAA,EAAa,MAAM,CAAC,KAAK,EAAE;;IAG7D,MAAM,KAAK,GAAa,EAAE;AAE1B,IAAA,IAAI,MAAM,CAAC,GAAG,IAAI,IAAI,IAAI,MAAM,CAAC,GAAG,KAAK,EAAE,EAAE;QAC3C,KAAK,CAAC,IAAI,CAAC,CAAA,iBAAA,EAAoB,MAAM,CAAC,GAAG,CAAE,CAAA,CAAC;;AAE9C,IAAA,IAAI,MAAM,CAAC,KAAK,IAAI,IAAI,IAAI,MAAM,CAAC,KAAK,KAAK,EAAE,EAAE;QAC/C,KAAK,CAAC,IAAI,CAAC,CAAA,gBAAA,EAAmB,MAAM,CAAC,KAAK,CAAE,CAAA,CAAC;;AAE/C,IAAA,IAAI,MAAM,CAAC,WAAW,IAAI,IAAI,IAAI,MAAM,CAAC,WAAW,KAAK,EAAE,EAAE;QAC3D,KAAK,CAAC,IAAI,CAAC,CAAA,6BAAA,EAAgC,MAAM,CAAC,WAAW,CAAE,CAAA,CAAC;;AAElE,IAAA,IAAI,MAAM,CAAC,UAAU,IAAI,IAAI,IAAI,MAAM,CAAC,UAAU,KAAK,EAAE,EAAE;AACzD,QAAA,KAAK,CAAC,IAAI,CAAC,+CAA+C,CAAC;;AAG7D,IAAA,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE;QACtB,OAAO,CAAA,gBAAA,EAAmB,MAAM,CAAA,yBAAA,CAA2B;;AAG7D,IAAA,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC;AACzB;AAEA;;;;;;;;;;;;;AAaG;AACG,SAAU,kBAAkB,CAChC,OAAmC,EAAA;AAEnC,IAAA,MAAM,EAAE,aAAa,EAAE,GAAG,OAAO,IAAI,EAAE;IACvC,MAAM,KAAK,GAA4B,EAAE;AAEzC;;;AAGG;AACH,IAAA,MAAM,kBAAkB,GAAG,CAAC,MAAc,KAAI;AAC5C,QAAA,OAAO,OACL,IAA6B,EAC7B,MAAuC,KACpB;AACnB,YAAA,MAAM,UAAU,GACd,MAAM,EAAE,QAAQ,EAAE,EAAE;gBACpB,CAAQ,KAAA,EAAA,IAAI,CAAC,GAAG,EAAE,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE;;AAG7D,YAAA,MAAM,MAAM,GAAwB;AAClC,gBAAA,wBAAwB,EAAE,IAAI;gBAC9B,MAAM;gBACN,IAAI;gBACJ,UAAU;aACX;;YAGD,IAAI,CAAC,aAAa,EAAE;AAClB,gBAAA,OAAO,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC;;;;AAK/B,YAAA,IAAI;gBACF,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC,MAAM,EAAE,IAAI,EAAE,UAAU,CAAC;AAC5D,gBAAA,OAAO,kBAAkB,CAAC,MAAM,EAAE,MAAM,CAAC;;YACzC,OAAO,KAAK,EAAE;AACd,gBAAA,MAAM,YAAY,GAChB,KAAK,YAAY,KAAK,GAAG,KAAK,CAAC,OAAO,GAAG,MAAM,CAAC,KAAK,CAAC;AACxD,gBAAA,OAAO,CAAmB,gBAAA,EAAA,MAAM,CAAa,UAAA,EAAA,YAAY,EAAE;;AAE/D,SAAC;AACH,KAAC;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,OAAO,CAAC,EAAE;QAChC,IAAI,EAAE,aAAa,CAAC,KAAK;AACzB,QAAA,WAAW,EAAE,CAAA;;;AAG6D,+EAAA,CAAA;AAC1E,QAAA,MAAM,EAAE,kBAAkB;AAC3B,KAAA,CAAC,CACH;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,MAAM,CAAC,EAAE;QAC/B,IAAI,EAAE,aAAa,CAAC,IAAI;AACxB,QAAA,WAAW,EAAE,CAAA;;;AAGgC,kDAAA,CAAA;AAC7C,QAAA,MAAM,EAAE,iBAAiB;AAC1B,KAAA,CAAC,CACH;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,UAAU,CAAC,EAAE;QACnC,IAAI,EAAE,aAAa,CAAC,QAAQ;AAC5B,QAAA,WAAW,EAAE,CAAA;AACwC,0DAAA,CAAA;AACrD,QAAA,MAAM,EAAE,qBAAqB;AAC9B,KAAA,CAAC,CACH;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,QAAQ,CAAC,EAAE;QACjC,IAAI,EAAE,aAAa,CAAC,MAAM;AAC1B,QAAA,WAAW,EAAE,CAAA;;AAE+D,iFAAA,CAAA;AAC5E,QAAA,MAAM,EAAE,mBAAmB;AAC5B,KAAA,CAAC,CACH;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,SAAS,CAAC,EAAE;QAClC,IAAI,EAAE,aAAa,CAAC,OAAO;AAC3B,QAAA,WAAW,EAAE,CAAA;AACwB,0CAAA,CAAA;AACrC,QAAA,MAAM,EAAE,oBAAoB;AAC7B,KAAA,CAAC,CACH;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,OAAO,CAAC,EAAE;QAChC,IAAI,EAAE,aAAa,CAAC,KAAK;AACzB,QAAA,WAAW,EAAE,CAAA;AACmE,qFAAA,CAAA;AAChF,QAAA,MAAM,EAAE,kBAAkB;AAC3B,KAAA,CAAC,CACH;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,MAAM,CAAC,EAAE;QAC/B,IAAI,EAAE,aAAa,CAAC,IAAI;AACxB,QAAA,WAAW,EAAE,CAAA;;AAEiC,mDAAA,CAAA;AAC9C,QAAA,MAAM,EAAE,iBAAiB;AAC1B,KAAA,CAAC,CACH;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,MAAM,CAAC,EAAE;QAC/B,IAAI,EAAE,aAAa,CAAC,IAAI;AACxB,QAAA,WAAW,EAAE,CAAA;AAC6C,+DAAA,CAAA;AAC1D,QAAA,MAAM,EAAE,iBAAiB;AAC1B,KAAA,CAAC,CACH;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,YAAY,CAAC,EAAE;QACrC,IAAI,EAAE,aAAa,CAAC,UAAU;AAC9B,QAAA,WAAW,EAAE,CAAA;;AAEgD,kEAAA,CAAA;AAC7D,QAAA,MAAM,EAAE,uBAAuB;AAChC,KAAA,CAAC,CACH;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,gBAAgB,CAAC,EAAE;QACzC,IAAI,EAAE,aAAa,CAAC,cAAc;AAClC,QAAA,WAAW,EAAE,CAAA;;AAEuD,yEAAA,CAAA;AACpE,QAAA,MAAM,EAAE,yBAAyB;AAClC,KAAA,CAAC,CACH;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,eAAe,CAAC,EAAE;QACxC,IAAI,EAAE,aAAa,CAAC,aAAa;AACjC,QAAA,WAAW,EAAE,CAAA;;;;AAImC,qDAAA,CAAA;AAChD,QAAA,MAAM,EAAE,yBAAyB;AAClC,KAAA,CAAC,CACH;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,aAAa,CAAC,EAAE;QACtC,IAAI,EAAE,aAAa,CAAC,WAAW;AAC/B,QAAA,WAAW,EAAE,CAAA;;;AAGgC,kDAAA,CAAA;AAC7C,QAAA,MAAM,EAAE,uBAAuB;AAChC,KAAA,CAAC,CACH;;IAGD,KAAK,CAAC,IAAI,CACR,IAAI,CAAC,kBAAkB,CAAC,UAAU,CAAC,EAAE;QACnC,IAAI,EAAE,aAAa,CAAC,QAAQ;AAC5B,QAAA,WAAW,EAAE,CAAA;;;;AAIkC,oDAAA,CAAA;AAC/C,QAAA,MAAM,EAAE,qBAAqB;AAC9B,KAAA,CAAC,CACH;AAED,IAAA,OAAO,KAAK;AACd;;;;"}
|
|
@@ -30,39 +30,16 @@ export type BrowserToolName = (typeof EBrowserTools)[keyof typeof EBrowserTools]
|
|
|
30
30
|
* @returns Promise that resolves with the actual browser result (page state, etc.)
|
|
31
31
|
*/
|
|
32
32
|
export type BrowserToolCallback = (action: string, args: Record<string, unknown>, toolCallId: string) => Promise<BrowserActionResult>;
|
|
33
|
-
/**
|
|
34
|
-
* Action status - matches Skyvern's ActionStatus
|
|
35
|
-
*/
|
|
36
|
-
export type ActionStatus = 'pending' | 'completed' | 'failed' | 'skipped';
|
|
37
|
-
/**
|
|
38
|
-
* Verification status - matches Skyvern's VerificationStatus
|
|
39
|
-
* - complete: Goal achieved
|
|
40
|
-
* - terminate: Goal cannot be achieved, stop
|
|
41
|
-
* - continue: Goal not yet achieved, keep going
|
|
42
|
-
*/
|
|
43
|
-
export type VerificationStatus = 'complete' | 'terminate' | 'continue';
|
|
44
33
|
/**
|
|
45
34
|
* Result returned from browser action execution
|
|
46
35
|
*/
|
|
47
36
|
export interface BrowserActionResult {
|
|
48
37
|
success: boolean;
|
|
49
|
-
status?: ActionStatus;
|
|
50
38
|
url?: string;
|
|
51
39
|
title?: string;
|
|
52
40
|
elementList?: string;
|
|
53
41
|
error?: string;
|
|
54
|
-
errorType?: string;
|
|
55
42
|
screenshot?: string;
|
|
56
|
-
elementDescription?: string;
|
|
57
|
-
verification?: {
|
|
58
|
-
urlChanged: boolean;
|
|
59
|
-
titleChanged: boolean;
|
|
60
|
-
elementCountDelta: number;
|
|
61
|
-
significantChange: boolean;
|
|
62
|
-
dialogClosed: boolean;
|
|
63
|
-
formGone: boolean;
|
|
64
|
-
changeDescription: string;
|
|
65
|
-
};
|
|
66
43
|
}
|
|
67
44
|
/**
|
|
68
45
|
* Check if browser capability is available based on request headers or context
|
package/package.json
CHANGED
|
@@ -42,42 +42,16 @@ export type BrowserToolCallback = (
|
|
|
42
42
|
toolCallId: string
|
|
43
43
|
) => Promise<BrowserActionResult>;
|
|
44
44
|
|
|
45
|
-
/**
|
|
46
|
-
* Action status - matches Skyvern's ActionStatus
|
|
47
|
-
*/
|
|
48
|
-
export type ActionStatus = 'pending' | 'completed' | 'failed' | 'skipped';
|
|
49
|
-
|
|
50
|
-
/**
|
|
51
|
-
* Verification status - matches Skyvern's VerificationStatus
|
|
52
|
-
* - complete: Goal achieved
|
|
53
|
-
* - terminate: Goal cannot be achieved, stop
|
|
54
|
-
* - continue: Goal not yet achieved, keep going
|
|
55
|
-
*/
|
|
56
|
-
export type VerificationStatus = 'complete' | 'terminate' | 'continue';
|
|
57
|
-
|
|
58
45
|
/**
|
|
59
46
|
* Result returned from browser action execution
|
|
60
47
|
*/
|
|
61
48
|
export interface BrowserActionResult {
|
|
62
49
|
success: boolean;
|
|
63
|
-
status?: ActionStatus;
|
|
64
50
|
url?: string;
|
|
65
51
|
title?: string;
|
|
66
|
-
elementList?: string;
|
|
52
|
+
elementList?: string; // Text-based element list
|
|
67
53
|
error?: string;
|
|
68
|
-
|
|
69
|
-
screenshot?: string;
|
|
70
|
-
elementDescription?: string;
|
|
71
|
-
// State verification (before/after comparison)
|
|
72
|
-
verification?: {
|
|
73
|
-
urlChanged: boolean;
|
|
74
|
-
titleChanged: boolean;
|
|
75
|
-
elementCountDelta: number;
|
|
76
|
-
significantChange: boolean;
|
|
77
|
-
dialogClosed: boolean;
|
|
78
|
-
formGone: boolean;
|
|
79
|
-
changeDescription: string;
|
|
80
|
-
};
|
|
54
|
+
screenshot?: string; // Base64 screenshot (if requested)
|
|
81
55
|
}
|
|
82
56
|
|
|
83
57
|
/**
|
|
@@ -213,30 +187,30 @@ export interface CreateBrowserToolsOptions {
|
|
|
213
187
|
*/
|
|
214
188
|
function formatResultForLLM(
|
|
215
189
|
result: BrowserActionResult,
|
|
216
|
-
action: string
|
|
217
|
-
actionArgs?: Record<string, unknown>
|
|
190
|
+
action: string
|
|
218
191
|
): string {
|
|
219
192
|
if (!result.success && result.error) {
|
|
220
|
-
|
|
221
|
-
return `${errorType}Action "${action}" failed: ${result.error}`;
|
|
193
|
+
return `Browser action "${action}" failed: ${result.error}`;
|
|
222
194
|
}
|
|
223
195
|
|
|
224
196
|
const parts: string[] = [];
|
|
225
197
|
|
|
226
|
-
|
|
227
|
-
|
|
228
|
-
|
|
229
|
-
|
|
230
|
-
|
|
231
|
-
|
|
232
|
-
|
|
233
|
-
}
|
|
198
|
+
if (result.url != null && result.url !== '') {
|
|
199
|
+
parts.push(`**Current URL:** ${result.url}`);
|
|
200
|
+
}
|
|
201
|
+
if (result.title != null && result.title !== '') {
|
|
202
|
+
parts.push(`**Page Title:** ${result.title}`);
|
|
203
|
+
}
|
|
204
|
+
if (result.elementList != null && result.elementList !== '') {
|
|
205
|
+
parts.push(`\n**Interactive Elements:**\n${result.elementList}`);
|
|
206
|
+
}
|
|
207
|
+
if (result.screenshot != null && result.screenshot !== '') {
|
|
208
|
+
parts.push('\n[Screenshot captured and displayed to user]');
|
|
234
209
|
}
|
|
235
210
|
|
|
236
|
-
|
|
237
|
-
|
|
238
|
-
|
|
239
|
-
if (result.elementList) parts.push(`\nElements:\n${result.elementList}`);
|
|
211
|
+
if (parts.length === 0) {
|
|
212
|
+
return `Browser action "${action}" completed successfully.`;
|
|
213
|
+
}
|
|
240
214
|
|
|
241
215
|
return parts.join('\n');
|
|
242
216
|
}
|
|
@@ -291,13 +265,11 @@ export function createBrowserTools(
|
|
|
291
265
|
// The marker is still returned initially via SSE, but we wait for the callback
|
|
292
266
|
try {
|
|
293
267
|
const result = await waitForResult(action, args, toolCallId);
|
|
294
|
-
return formatResultForLLM(result, action
|
|
268
|
+
return formatResultForLLM(result, action);
|
|
295
269
|
} catch (error) {
|
|
296
270
|
const errorMessage =
|
|
297
271
|
error instanceof Error ? error.message : String(error);
|
|
298
|
-
return
|
|
299
|
-
|
|
300
|
-
**REQUIRED**: Analyze why the action failed and try an alternative approach.`;
|
|
272
|
+
return `Browser action "${action}" failed: ${errorMessage}`;
|
|
301
273
|
}
|
|
302
274
|
};
|
|
303
275
|
};
|
|
@@ -306,8 +278,10 @@ export function createBrowserTools(
|
|
|
306
278
|
tools.push(
|
|
307
279
|
tool(createToolFunction('click'), {
|
|
308
280
|
name: EBrowserTools.CLICK,
|
|
309
|
-
description: `Click an element
|
|
310
|
-
|
|
281
|
+
description: `Click an element on the current web page by its index number.
|
|
282
|
+
The element list shows clickable items like: [0]<button>Submit</button> [1]<a href="/home">Home</a>
|
|
283
|
+
Use the index number in brackets to click that element.
|
|
284
|
+
After clicking, you receive an updated element list showing the new page state.`,
|
|
311
285
|
schema: BrowserClickSchema,
|
|
312
286
|
})
|
|
313
287
|
);
|
|
@@ -316,8 +290,10 @@ Returns state verification showing if UI changed (dialog closed, form submitted,
|
|
|
316
290
|
tools.push(
|
|
317
291
|
tool(createToolFunction('type'), {
|
|
318
292
|
name: EBrowserTools.TYPE,
|
|
319
|
-
description: `Type text into an input element
|
|
320
|
-
|
|
293
|
+
description: `Type text into an input element on the page.
|
|
294
|
+
Find the input element in the list by its index (e.g., [5]<input placeholder="Search">).
|
|
295
|
+
Set pressEnter: true to submit forms after typing.
|
|
296
|
+
After typing, you receive an updated element list.`,
|
|
321
297
|
schema: BrowserTypeSchema,
|
|
322
298
|
})
|
|
323
299
|
);
|
|
@@ -326,7 +302,8 @@ Returns state verification showing if form submitted or errors appeared.`,
|
|
|
326
302
|
tools.push(
|
|
327
303
|
tool(createToolFunction('navigate'), {
|
|
328
304
|
name: EBrowserTools.NAVIGATE,
|
|
329
|
-
description: `Navigate to a URL
|
|
305
|
+
description: `Navigate to a URL. Always include the full URL with https://.
|
|
306
|
+
After navigation, you receive the new page's element list.`,
|
|
330
307
|
schema: BrowserNavigateSchema,
|
|
331
308
|
})
|
|
332
309
|
);
|
|
@@ -335,7 +312,9 @@ Returns state verification showing if form submitted or errors appeared.`,
|
|
|
335
312
|
tools.push(
|
|
336
313
|
tool(createToolFunction('scroll'), {
|
|
337
314
|
name: EBrowserTools.SCROLL,
|
|
338
|
-
description: `Scroll the page
|
|
315
|
+
description: `Scroll the page to reveal more content.
|
|
316
|
+
Use 'down' to scroll down, 'up' to scroll up.
|
|
317
|
+
After scrolling, you receive an updated element list with newly visible elements.`,
|
|
339
318
|
schema: BrowserScrollSchema,
|
|
340
319
|
})
|
|
341
320
|
);
|
|
@@ -385,7 +364,9 @@ After going back, you receive the previous page's element list.`,
|
|
|
385
364
|
tools.push(
|
|
386
365
|
tool(createToolFunction('screenshot'), {
|
|
387
366
|
name: EBrowserTools.SCREENSHOT,
|
|
388
|
-
description: `Capture a screenshot
|
|
367
|
+
description: `Capture a screenshot of the current page.
|
|
368
|
+
Returns the page state with a note that screenshot was displayed to the user.
|
|
369
|
+
Use browser_get_page_state to get the element list for automation.`,
|
|
389
370
|
schema: BrowserScreenshotSchema,
|
|
390
371
|
})
|
|
391
372
|
);
|
|
@@ -394,16 +375,22 @@ After going back, you receive the previous page's element list.`,
|
|
|
394
375
|
tools.push(
|
|
395
376
|
tool(createToolFunction('get_page_state'), {
|
|
396
377
|
name: EBrowserTools.GET_PAGE_STATE,
|
|
397
|
-
description: `Get current page URL, title, and interactive elements
|
|
378
|
+
description: `Get the current page state including URL, title, and all interactive elements.
|
|
379
|
+
Use this at the start of a task to see what elements are available.
|
|
380
|
+
Returns a text list of elements with their index numbers for interaction.`,
|
|
398
381
|
schema: BrowserGetPageStateSchema,
|
|
399
382
|
})
|
|
400
383
|
);
|
|
401
384
|
|
|
402
|
-
// browser_select_option
|
|
385
|
+
// browser_select_option - Skyvern-inspired for robust dropdown handling
|
|
403
386
|
tools.push(
|
|
404
387
|
tool(createToolFunction('select_option'), {
|
|
405
388
|
name: EBrowserTools.SELECT_OPTION,
|
|
406
|
-
description: `Select an option from a dropdown
|
|
389
|
+
description: `Select an option from a dropdown or select element.
|
|
390
|
+
For native <select> elements: finds and selects the option by value/label.
|
|
391
|
+
For custom dropdowns: clicks to open, then clicks the matching option.
|
|
392
|
+
Use this instead of click for dropdowns - it handles both native and custom selects.
|
|
393
|
+
After selection, you receive an updated element list.`,
|
|
407
394
|
schema: BrowserSelectOptionSchema,
|
|
408
395
|
})
|
|
409
396
|
);
|