?ixdZddlZddlZddlZddlmZmZmZddlm Z ejdZ GddZ y)u Genesis Superior Browser (GSB) — Vision Bridge (Layer 2) ======================================================= Integrates Gemini Flash Vision to reason about page state and identify coordinates. N)DictListOptional)Pagezgenesis_v2.core.browser.visionc8eZdZd dZdededefdZdedefdZy) VisionBridgeNc||_yN)client)self gemini_clients F/mnt/e/genesis-system/genesis_v2/core/browser_engines/vision_bridge.py__init__zVisionBridge.__init__s # pagegoalreturncK|jddd{}d|d}|jstjddd d S d |d g}|jj ||d g}d dl}|j d|j|j}|r|jd} n |j} tj| S7#tj$r*tjd|jddd cYSwxYw#t$r/} tjd| dt| icYd} ~ Sd} ~ wwxYww)z Takes a screenshot, sends it to Gemini Flash, and returns the coordinates of the element to interact with. pngF)type full_pageNzF Analyze this screenshot of a webpage. Your goal is: a Return a JSON object with: 1. "reasoning": Why you choosing this action. 2. "action": One of ["click", "type", "scroll", "wait"]. 3. "coordinates": {"x": Number, "y": Number} of the target element. 4. "selector": (Optional) The CSS selector if you can identify it reliably. 5. "input_text": (Optional) If action is 'type'. The screen resolution is 1920x1080. ONLY return JSON. zDNo Gemini client provided to VisionBridge. Visual reasoning limited.waitzNo vision brain connected.)action reasoningz image/png) mime_typedatarz```json\n(.*?)\n```z6Failed to parse JSON from vision model. Raw response: z+Failed to parse response from vision model.zVision analysis failed: error) screenshotr loggerwarninggenerate_contentresearchtextDOTALLgroupjsonloadsJSONDecodeErrorr Exceptionstr) r rrscreenshot_bytesprompt image_partsresponser#match json_textes r analyze_pagezVisionBridge.analyze_pagesZ "&eu!MMf    {{ NNa b$3OP P %"-,K {{33V[^4LMH II4hmmRYYOE!KKN $MM  fzz),,SNT'' f UV^VcVcUdef"(7dee f % LL3A37 8SV$ $ %sdEC0E A1D=CE:DDEDD E $EE EE  E vision_resultcpK|jd}|jd}|dk(rF|rD|jj|d|dd{tj d||S|dk(ry|rw|jj|d|dd{|j j |jd d d{tj d ||S|d k(r9|jjd dd{tj d|S777b7!w)z3Executes the action determined by the vision brain.r coordinatesclickxyNz"Vision-grounded click executed at r input_textz#Vision-grounded typing executed at scrollriz Vision-grounded scroll executed.)getmouser8r infokeyboardrwheel)r rr5rcoordss rexecute_vision_movez VisionBridge.execute_vision_moveIs)""8,""=1 W **""6#;s < < < KKrOs=  ''%   ; <JJr