Point, click, and nag your AI about slides

What it does slides-grab is a workflow and browser editor for building slide decks as plain HTML and CSS files. An agent—Claude Code or Codex—generates the markup, and you refine it by dragging a bounding box over whatever looks wrong and asking the agent to rewrite it. The editor also handles direct text tweaks for the small stuff. Once done, the deck exports to PDF or to experimental PPTX and Figma formats.

The interesting bit The README frames HTML/CSS as the “programming language (which is not) that outperformed by AI agents”—meaning agents handle it better than most real code. Because slides are just web pages, the editor can render 95 bundled design styles as CSS, validate assets with Playwright, and even generate images through god-tibo-imagen, a backend that reuses your local Codex ChatGPT login instead of an API key.

Key highlights

• Browser-based editor with bbox selection for agent-driven rewrites.
• 95 bundled design styles drawn from community packs and original sets.
• PDF export in capture (raster) or print (searchable text) modes.
• Experimental PPTX and Figma import flows, explicitly labeled unstable.
• Asset validation via Playwright to catch missing local images before export.
• Square 1:1 “card news” mode for Instagram-style output.
• Image generation via god-tibo-imagen reusing local Codex ChatGPT credentials, plus optional OpenAI/Google fallbacks.

Caveats

• PPTX and Figma exports are experimental/unstable; expect layout shifts and manual cleanup.
• The default image generator, god-tibo-imagen, relies on an unsupported private Codex backend that may break without notice.
• Remote http(s):// image URLs are disallowed in saved slides; everything must be local or data URLs.

Verdict Developers who want AI-generated slides but need a visual feedback loop instead of prompt-only iteration will find this useful. If you need reliable PPTX round-tripping or are wary of experimental backend hacks for image generation, look elsewhere.