Heat Domes

From a selected PDF to a finished visual summary

This demonstration starts with a scientific paper selected on a Curio idea space. The goal was not merely to summarize the document in a chat window. The agent was asked to build the finished result in Curio, using native figures that could be rearranged, edited, and developed further.

The request was written in ordinary language:

I'm working on a Curio project about heat domes, and I have a PDF figure selected.

Help me plan a visual summary around it. I'd like:

1. A two-paragraph summary for a college freshman.
2. A Curio list containing the 10 main points.
3. Three generated images about heat domes, each 350 points wide and appropriately captioned.

Arrange everything neatly beside and below the PDF. Don't change the project yet—first give me a short plan that I can refine.

That final sentence establishes an important boundary: plan first, then act only after approval.

What the agent had to do

The request sounds simple, but completing it required the agent to coordinate several different capabilities exposed by the Curio CLI skills:

1. Use curio get to identify the selected PDF figure and obtain its underlying asset.
2. Read and summarize the scientific paper at the requested level.
3. Extract the ten most important points and format them as Markdown that Curio could import as a native list.
4. Generate three relevant illustrations with concise captions.
5. Use curio import to place the summary, list, and images into the project.
6. Use curio get and curio set to inspect and adjust bounds, captions, text colors, and background colors.
7. Arrange the new figures beside and below the source PDF.
8. Use curio save to persist the approved result.

The video accelerates the agent's thinking and image-generation pauses, but otherwise shows the complete workflow. The resulting summary and list remain editable Curio content; the generated images are normal image figures with captions.

Why skills matter

An AI agent does not inherently know Curio's object model or command syntax. The installed Curio skills provide concise instructions and references for tasks such as finding the selected figure, importing a collection, setting figure bounds, and saving a project.

For Codex, ChatGPT, and most other supported agents, install the agent skills from Terminal:

curio skills install agents

Claude uses its own installation target:

curio skills install claude

Then relaunch the agent so it discovers the new skills. You can inspect the installed material yourself:

curio skills reveal

The important pattern is that the agent translates an outcome-oriented request into a sequence of inspectable CLI operations. You can refine the plan without needing to dictate every command.

A reusable workflow

The subject does not have to be heat domes. The same pattern works for research papers, reports, course material, meeting packets, or any source document that would benefit from a visual working summary:

1. Select the source figure in Curio.
2. State the intended audience and desired outputs.
3. Specify which outputs should become native Curio structures.
4. Request a plan before any project changes.
5. Refine layout, tone, number of points, and visual style.
6. Approve the plan and watch the agent work.
7. Inspect the result in Curio and continue editing normally.

Useful constraints might include:

• "Use only information supported by the selected source."
• "Attach a source note to each factual claim."
• "Create a mind map instead of a list."
• "Do not generate images; use labeled placeholders."
• "Place everything in a new idea space."
• "Do not save until I review the result."

Learn more

Read the original forum post and discussion, watch the 4K video on YouTube, or explore the Curio CLI working guide.