| name | comprehensive-visual-extraction-v5 |
| description | 2026 SOTA V5 skill for Pixel-Semantic Resonance: exhaustive, hallucination-free extraction of 3D WebGL, AI dashboards, and adaptive layouts into executable agentic blueprints. |
| category | visual-extraction |
| version | 1.0.0 |
Comprehensive Visual Extraction (CVE) Skill - V5 Paradigm (2026 Edition)
Overview
This skill equips the autonomous agent with the V5 State-of-the-Art (SOTA) methodology for Pixel-Semantic Resonance. Moving far beyond the legacy V4 protocol's static 2D coordinate mapping, V5 treats visual perfection and semantic/structural correctness as a single, unified hybrid dimension.
The objective is absolute agentic reconstruction: the output must serve as a flawless, executable blueprint allowing a secondary "blind" agent to generate browser-native code (HTML/CSS/JS/WebGL) that achieves both pixel-perfect visual rendering and flawless functional equivalence across all devices. V5 natively handles the complexities of AI-native data streams, 3D WebGL raycasting, and cross-device adaptive fluid layouts.
Core Capabilities & The V5 Paradigm Shift
- Pixel-Semantic Resonance: You do not merely "see" pixels; you perceive the functional intent behind them. A shadow is not just a hex code; it is a CSS
box-shadow indicating elevation and interactive state. A 3D sphere is not just a circle; it is a WebGL mesh with material properties and raycast event listeners.
- Unified Spatial Reasoning (Layer Abstraction): You must classify the rendering pipeline. Distinguish between
DOM/UI, WebGL/3D Scene, and Canvas/SVG layers. Utilize Normalized Device Coordinates (NDC) and estimate 3D Camera Parameters (perspective, field of view) to map 2D screen clicks to 3D spatial intersections (Raycasting).
- State & Dynamics Modeling: Interfaces are living systems. You must extract Actionable Event Listeners (
onClick, onHover, onDrag), infer State Transition Logic (e.g., loading states, modal toggles), and identify Data Binding Semantics (how UI components map to underlying JSON/API data streams).
- Architectural Awareness (Cross-Device Adaptation): Reject absolute pixel dependency. You must infer the underlying CSS Grid/Flexbox topology, fluid units (
vw, vh, rem, %), and conditional Media Query breakpoints to ensure the blind agent's recreation is natively responsive.
The Extraction Protocol
When invoked, you must strictly execute this multi-layered, resonance-driven protocol.
Phase 1: Global Topology & Rendering Pipeline Analysis
- Define the Viewport & NDC: Map the image space conceptually from
[0, 0] to [100, 100] (NDC percentage scale), but immediately establish the assumed base viewport (e.g., 1440x900) to calculate fluid unit ratios.
- Rendering Layer Classification: Scan the image and segment it into rendering contexts:
[DOM]: Standard HTML/CSS elements.[WebGL/3D]: GPU-rendered scenes, depth buffers, lighting.[Overlay/HUD]: Native DOM elements floating above a 3D canvas.
- State Machine Hypothesis: Provide a 3-sentence summary of the application's core functional purpose, current interactive state (e.g., "Dashboard in 'Dark Mode' with active data-stream polling"), and primary user flow.
Phase 2: The Resonant GCoT Deep Extraction Process
Generate the output using a heavily structured format. For every distinct component, execute the V5 GCoT Loop.
The V5 GCoT Loop (Mandatory for every element)
- Anchor & Layer: State the NDC bounding box
[x1, y1, x2, y2], the estimated Z-Index/Depth, and the [Rendering Layer].
- Identify & Abstract: Classify the element semantically (e.g., "React Data Grid", "Three.js Interactive Mesh", "ARIA Navigation Landmark").
- Re-engage (Zoom & Raycast): Explicitly state: (Zooming into RoI / Casting Ray from Camera). If 3D, estimate the camera angle and the mesh normal at the point of interaction.
- Extract Pixel-Semantic Attributes:
- Visual (Pixel): Exact Hex/RGB, Material properties (metallic/roughness for 3D), Gradient angles/stops, Drop-shadow physics.
- Structural (CSS/Box): Flexbox/Grid alignment, padding/gap estimations, fluid unit translations (e.g.,
width: 35% instead of 504px).
- Functional (State/Events): Inferred event listeners, ARIA roles, data-binding keys (e.g.,
v-for="item in apiResponse").
- Spatial & Logical Relationship: Describe geometric alignment and logical hierarchy (e.g., "DOM child of
[Parent Anchor], acts as a raycast trigger for [WebGL Anchor]").
Categorical Breakdown (Applying the V5 GCoT Loop)
You must apply the loop comprehensively across these four dimensions:
- Adaptive Structural Layout: Map the macroscopic CSS Grid/Flexbox architecture. Extract Global Max-Width Wrappers, container queries, and explicit
@media breakpoint logic based on element wrapping behavior.
- 3D & WebGL Scene Graph: Reconstruct the 3D world. Define Cameras, Light Sources (directional/ambient), Mesh Geometries, Textures, and Raycast interaction zones.
- Dynamic Data & AI Dashboards: Extract chart topologies. Map visual data points to semantic values. Infer the API payload structure required to populate the widgets.
- Total Text & Semantic Hierarchy: Extract all text hierarchically. Reconstruct tables into perfect Markdown. Ensure logical DOM order for screen readers, independent of visual layout.
Phase 3: The Agentic Reconstruction Blueprint (Validation)
- Topology & Physics Check: Verify 3D depth/occlusion logic. Ensure DOM nesting rules are mathematically sound (child coordinates must logically fit within parent fluid containers).
- Responsive Fluidity Check: Confirm that absolute visual measurements have been successfully translated into scalable CSS logic (
%, vw, fr units).
- The Blind Agent Execution Test: Mentally simulate a blind agent reading your output. Could it write a functional
index.html, styles.css, app.js, and scene.jsx that passes dual-criteria validation (Pixel-perfect rendering + Functional equivalence)? If not, iterate and refine the blueprint.
Output Format Specification
The final output must be formatted in clean, highly structured Markdown.
- Executive Blueprint Summary: A high-level architectural overview (Tech stack inference, State management pattern, 3D Engine requirements).
- CSS & Responsive Architecture: Code blocks defining the inferred CSS Grid/Flexbox layouts, CSS Variables (Design Tokens), and Media Queries.
- The V5 Scene Graph (JSON): Use code blocks to output a unified JSON array representing the DOM tree and 3D Scene Graph, including bounding boxes, semantic tags, and event listeners.
- Resonant GCoT Element Inventory: Standard Markdown headers for each major component, strictly preceded by its spatial anchor
[x1, y1, x2, y2] | [Layer].
- Behavioral & State Rules: A declarative list mapping user interactions to application responses (e.g.,
ON_CLICK [Submit Button] -> TRIGGER API_POST -> MUTATE STATE [Loading]).
End of Skill Definition
