slides-grab lets agents fix slide layouts

This is a smart fix for the worst part of AI-generated decks: the “almost right” slide that takes five prompts to nudge into shape. By turning a visual selection into machine-readable context, it gives agents the exact target instead of forcing users to describe layout bugs in prose.

• –The selection-plus-screenshot handoff is the real product, because it reduces ambiguity and keeps edits pointed at the right element.
• –It fits the current HTML/CSS slide generation stack better than traditional PowerPoint workflows, where AI edits are much harder to localize.
• –The built-in editor covers small manual tweaks, so users can reserve agent calls for the annoying structural fixes.
• –Experimental PPTX and Figma export show the ecosystem is still messy, but the workflow is good enough to matter now.