information-hierarchy — скилл для Claude Code: структура больших

---
name: information-hierarchy
description: |
  Load for large human-facing outputs: reports, docs, multi-section
  explanations, artifacts. Choose the right form for each beat (prose, diagram,
  table, mockup) and put the answer first, depth behind it. If the medium allows, use progressive disclosure.
---

# Information Hierarchy

The reader should get the answer in the first thing they see, and find depth
only when they reach for it. Everything else here follows from that: each beat
of the piece takes the form that carries it best, what matters most comes
first, and the medium stays as simple as the content allows.

## Lead with the answer/conclusion

Readers scan. Open with the finding itself, and the piece works in thirty
seconds. The same logic repeats at every scale: a section's first sentence is
its answer, a slide's title is its takeaway ("Cache misses double p99" rather
than "Performance"), a UI label carries its whole meaning alone.

Behind the answer sits depth (reasoning, evidence, edge cases), and behind
that, sources. Keep depth that most readers will skip, and *move* it: a later
section, a linked page, a footnote. Sources belong at the end.

## Pick the form that fits the idea

Prose is the default; break from it when another form genuinely carries the
idea better. An example grounds an abstract claim faster than more
explanation. A table holds enumerable facts that prose would bury. A chart
shows a trend that a sentence can only assert. A mockup shows a layout the
reader would otherwise have to imagine.

**Diagram-first for pipelines and relationships.** When the content is a
pipeline, a dependency graph, a state machine, or any structure where the
*relationship between parts* is the hard idea, put the diagram early — a
one-sentence takeaway, then the diagram, then detailed prose. The diagram
leads, the prose still explains; prose alone buries the shape. Keep inline
diagrams to a handful of nodes and give larger structures an explorable view.
Validate diagrams (see `/md-validation`).

Ask what the beat is trying to do, then give it the form that does that: an
explanation gets a paragraph, a comparison gets a table. Each form teaches
best inside its own job: a diagram standing in for an explanation teaches
less than the paragraph would.

## Match hierarchy to the medium

For any medium, ask one question: can it hide content until the reader asks
for it? Chat, email, and plain documents show everything at once, so
hierarchy there means density and order: every early sentence pays for
itself, and depth lives behind a link. Media that can hide (HTML, apps,
anything with expand or navigate) may fold depth into collapsed sections,
popovers, or child pages. Reserve each interaction for depth a reader will
actually reach for.

Structure helps the scanner in every medium: headers that carry the argument
on their own, asides for context off the main thread, a table of contents
when the piece is long enough to need a map. These organize the surface;
putting the answer first remains the job of the writing itself.

Choose the simplest medium that presents every beat well: a tight reply
beats an HTML site whenever both would teach the same thing. When HTML is
earned, build it phone-first and use `/structured-artifact`.

## Sources

- [How Users Read on the Web (NN/g)](https://www.nngroup.com/articles/how-users-read-on-the-web/):
  scanning, why the lede must come first.
- [Progressive Disclosure (NN/g)](https://www.nngroup.com/articles/progressive-disclosure/):
  defer secondary content to secondary screens.