Adaptive Selector Generation From Semantic Intent

via “intent-preserving semantic decomposition and restructuring”

[CVPR 2026] PromptEnhancer is a prompt-rewriting tool, refining prompts into clearer, structured versions for better image generation.

Unique: Explicitly models semantic decomposition and intent preservation as core capabilities, using chain-of-thought reasoning to make the transformation process interpretable. This differs from black-box prompt expansion that doesn't explicitly track semantic elements.

vs others: Provides more interpretable and intent-preserving prompt enhancement than generic text expansion, because it explicitly decomposes and validates semantic elements rather than treating the prompt as unstructured text.

** - Enable AI agents to get structured data from unstructured web with [AgentQL](https://www.agentql.com/).

Unique: Generates selectors from semantic intent rather than requiring agents to write or understand CSS — the system infers what elements match the intent and creates resilient selectors that tolerate minor DOM variations

vs others: More maintainable than hardcoded CSS selectors because it adapts to DOM changes automatically, and more accessible than XPath/CSS because agents express intent in natural language rather than selector syntax

via “adaptive-selector-generation”