Digestif

What you are looking at

This page is the actual viewer that digestif build generated from the blog post. It is one self-contained HTML file with no server behind it. Everything you see was produced by the deterministic renderer from a single graph.json, which in turn was written by an agent following the digestif skill. Every node cites the passages it came from, and the validator checked that citations exist, that all 78 passages are covered, and that each quote matches the source text character for character.

• Each box is one atomic idea, a claim, an evidence item, an example, a nuance, and so on.
• Gray curves are the tree. Colored curves are typed relations across branches, green supports, red contradicts, blue qualifies, gray restates.
• Click a node to select it. The sidebar shows its type, its cited excerpts, and every relation.
• The sidebar's Source toggle shows the same citations highlighted inside the full source text. Click a highlighted sentence to jump to a node that cites it.
• The aligned reading view shows the original post as written, with the main points beside the passages they come from. Open any point with + to add its children in the next column.

Controls

• scroll moves, shift+scroll sideways, ctrl+scroll zooms, ctrl+= and ctrl+- zoom from the keyboard.
• arrows navigate in top to bottom order, left folds or goes to the parent, right unfolds and descends, space folds the selected branch.
• ctrl+1..9 show that many layers, ctrl+0 shows all, ctrl+alt+1..9 focus the selected branch.
• Drag the sidebar's left edge to resize it, or drag the canvas to pan freely.

loading...

Reproduce it

pipx install git+https://github.com/k1monfared/digestif then digestif build your-text.md. The extraction runs through an agent CLI you already have installed (Claude Code, Codex, or opencode), everything after that is plain Python, and the run folder gets this same viewer, the outline, the validated graph, and the full audit log.