STONE

SkillDev tools

The one-page tier: distill a whole system onto a single dense legible page whose star is the relationships

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the STONE skill

What this skill tells your AI

The instructions your AI receives, as published by simhacker/moollm in skills/stone/SKILL.md and read by ahel’s review.

"If you only read one page, read this one."

Read the seed first

This file is not the authority. examples/STONE.md is.

That file is three things at once: a STONE (its subject is Stone Librande), the tribute page it needs to be to earn the name, and the prototype every other STONE clones. The protocol below was read off that instance rather than imposed on it, which is the prototype way — instance precedes class. If this file and the seed ever disagree, the seed wins and this file is stale.

So the shortest possible use of this skill is: open the seed, copy it, swap the subject. Everything below is commentary on what you will find there.

Why the tier exists

Stone Librande's question is the whole argument:

💬 "Why create a document with more than one page if most people only read the first page anyway?" — Stone Librande

And its companion, which is the part people skip:

💬 "Diagrams make relationships visible. Words make them invisible." — Stone Librande

MOOLLM already had tiers above and below this one. A GLANCE answers is this relevant; a CARD answers what can it do; a README answers why was it built and is read by almost nobody. The gap was the page you would keep if you could keep only one — dense enough to be the real thing, short enough to actually get read.

GLANCECARDDESCRIPTIONSTONEREADME → directory

Lower tiers tease; the README sprawls; the STONE is the keeper.

The four properties

A file is a STONE when it has all four. Three out of four is a nice document with the wrong name.

PropertyThe test
One pageYou stopped. Denser than a DESCRIPTION, shorter than a README.
Relationships are the starThere is an actual arrows block, not prose that mentions connections.
Made from understandingYou could not have drawn those arrows without knowing the system.
The read-one-page viewIf every other tier were deleted, this one would still serve.

One page means stopping

The hard property, because every other one is additive and this one is subtractive. The page is finished when the next true sentence would push it to two pages and you leave that sentence out. A two-page STONE is a README with a haircut.

What makes this affordable rather than lossy is that the other tiers exist. You are not throwing the detail away, you are putting it where it belongs and linking to it. A STONE that tries to be complete has misunderstood which document it is.

The arrows are structural, not decorative

"X influenced Y" inside a paragraph is invisible; the reader has to assemble the graph in their head and mostly will not. A relationships block hands them the graph. From the seed:

Will Wright ──"make the complex graspable, don't oversimplify"──▶┐
Chaim Gingold ──reverse-diagram lineage─────────────────────────▶┤
                                              🪨 Stone Librande ──┤── one-page designs
                                                                  ├──▶ Don Hopkins — one-page UIs & control panels for Micropolis
                                                                  └──▶ STONE — this very skill / resolution tier (named for him)

Label the edges. An unlabeled arrow asserts that a relationship exists while concealing what it is, which is the prose failure in ASCII form.

Earned, which is why it cannot be faked

You cannot draw a correct arrows block for a system you have not understood. This is the property that makes the tier trustworthy: a STONE is evidence of comprehension, not merely a claim of it. It is also why "summarize this to one page" is the wrong instruction to give yourself — summarizing what you do not understand produces a legible lie, which is worse than sprawl because it reads as authoritative.

Making one

STONE [subject]
  1. CLONE examples/STONE.md beside the subject
  2. set parents: [ moollm://…/skills/stone/examples/STONE.md ]
  3. ARROWS  — draw the relationships block first, before the prose
  4. write the page around the arrows
  5. AUDIT   — check the four properties; cut until one page is true

Draw the arrows before writing the prose. Doing it in that order surfaces what you do not yet understand while it is still cheap to go find out. Doing it in the other order produces prose that gets diagrammed, which is decoration.

Front matter carries is_a, parents, and — when the subject is a real person — representation: with a consent level per representation-ethics. The seed demonstrates all of it.

The prototype clause

The seed governs by example, not by decree. Improve the seed and every clone inherits the improvement; the rules in this file are a description of what the seed does, so improving the seed makes this file wrong rather than making the clone wrong. That is the intended direction of authority.

Practically: if you find a better way to lay out a STONE, change examples/STONE.md and update this file to match. Do not add a rule here that the seed does not demonstrate.

Ornament

Density is not permission to decorate. A STONE runs against no-decorative-comments like everything else — no rules of repeated dashes, no visual filler. But the tier is also the one place where earned ornament pays: emoji as rungs, an arrows block that is pleasant to look at, a quote placed where it lands. Clear the ground first; then what you add is legible against it.

Methods

MethodSignatureDoes
STONESTONE [subject]Full pipeline: clone, arrows, page, audit
CLONECLONE [seed] FOR [subject]Copy the seed and reparent it
AUDITAUDIT [file]Report which of the four properties fail
ARROWSARROWS [subject]The relationships block alone
LADDERLADDER [file]Recommend a resolution tier for a document

LADDER is the one that gets used most and it usually answers not a STONE. Most documents want a CARD. The tier is for systems whose relationships are the interesting part.

Part of MOOLLM

A skill in MOOLLM — see skills/ for the ecosystem. The tier sits inside the semantic pyramid described in the repo root rules; its nearest neighbour with a skill of its own is card, one rung down. GLANCE.yml is a file convention rather than a skill.

Signals

GitHub stars
52
Forks
5
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
stone
Source
github.com/simhacker/moollm