It doesn’t generate images.
It won’t fake good taste.

An aesthetic operating system that refuses. Direction before the work, independent acceptance before delivery. It won’t praise what it hasn’t seen.

Golden spiral composition analysis over a Bada Shanren ink bird; the red spiral converges on the bird's perch, the cyan centroid sits on a golden-section point
spiral · bird centroid 0.022 from golden point

It actually refuses.

Not a mock. Real output from aos.py verdict. Missing any piece of evidence, the interface never emits the word “pass.”

⛔ Refusing to issue “aesthetic pass”: - Aesthetic proposition (write one with propose) - Taste brief missing purpose, audience, signature, decisions 0/3, non-goals <2 - Artifact evidence 0 (need ≥2, different breakpoints or states) - Latest artifact has no quality critique: measured structure ≠ judged quality - Aesthetic reason under 20 chars (now 19) No verdict issued. You may only say “technical pass, aesthetic pending.”

Three red lines. Not suggestions.

Speed, obedience, user pressure, and sunk cost are not exceptions. Saying “use Aesthetics-OS” while violating any red line still counts as failure.

I

No inspection, no pass

“Technical pass, aesthetic pending.” Without seeing the real artifact, it will not claim refined, premium, or final. Technical acceptance cannot replace aesthetic acceptance.

II

Two rejections, no reskinning

“Stop. New fonts, colors, or effects are not a new direction.” Build new evidence and a checkpoint with real conceptual distance first. A near-identical fingerprint gets caught by numbers.

III

Vague tech words, no templates

“‘AI feel’ does not imply dark + cyan gradient + glowing grid.” Derive form from the task, trust, information density, and real content.

Give a direction. Get a qualified artifact.

Real case, “Lone Fisherman on a Cold River.” Proposition: the solitude of being pushed far away by heaven and earth. The judge looks at real pixels and refuses to rubber-stamp three times, until round four when the boat truly dissolves into a misty ink shadow. It checks the proposition’s intent, not the overall impression.

init→design harmony→generate→measure→judge quality→revise→pass
Four-round iteration of Lone Fisherman on a Cold River: R1 to R3 all revise, the boat shrinks and softens from dark and jumping forward, R4 passes as it dissolves into a misty ink shadow

Euler’s formula. Not mysticism.

e^(iθ) = cosθ + i·sinθ has several strictly verifiable structures, migrated into designable, measurable aesthetic signals. Hue is phase, structure lives in Fourier phase, material naturalness has orientation entropy. Each carries an honest boundary.

palette / hue

Hue phase

The hue circle is the unit circle. Complementary = +π, triadic = ±2π/3. Phase rotation derives a harmonious palette, written into the prompt.

Hue phase vector plot of A Thousand Li of Rivers and Mountains, concentrated in green-yellow-green and cyan-blue, resultant at 105 degrees
spectrum / pc

Fourier phase

Swap two images’ phase and the composition follows the phase donor (correlation 0.626 vs amplitude 0.218). Keep only phase and structure survives; keep only amplitude and it collapses.

Fourier phase versus amplitude swap experiment; structure follows the phase donor
pc

Phase congruency

An illumination-invariant structural feature map. The cold-river PC concentrates on the boat; A Thousand Li spreads along the brushwork.

Phase congruency heatmaps: sparse strong features of the cold river versus dense texture of A Thousand Li
gabor

Material plasticity

Low orientation entropy + high anisotropy = smooth gradient plastic; high entropy = natural brushwork.

aurora 0.696 · plastic
Qianli 0.995 · natural
Orientation entropynormalized 0-1, higher = more diverse
spiral

Golden spiral

PC weight centroid vs φ golden points. Off-center subjects align; centered is judged legal. A reference, not a verdict.

Golden spiral over the Bada ink bird, centroid aligned to a golden-section point
It raises the floor.
It does not create beauty.

A critique is a vision model’s judgment, not objective truth. The taste anchor is a calibration reference, not a definition of beauty. It guarantees the unqualified cannot pass as qualified, and forces iteration until qualified. The source of beauty is still model capability and your taste. Treating “Euler empowers everything” as a slogan is exactly the vague-word templating it intercepts.

A cinematic dusk landscape generated under a Euler split-complementary palette, warm amber light against cool cyan-blue shadow
palette · split-complementary guided, R=0.354 balance

Drop it into your agent.

Place it in the skills directory. State lives in the designed project and persists across sessions. Run the self-checks and regressions after install.

# clone into your agent skills directory git clone https://github.com/Greatbeing/Aesthetics-OS.git <skills-dir>/Aesthetics-OS # self-check structure and encoding, run regressions python scripts/lint_skill.py . python scripts/verify_utf8.py . python scripts/test_aos.py # 35 assertions, refusal paths python scripts/test_euler_lens.py # 18 assertions, Euler lenses # start a project's aesthetic loop python scripts/aos.py init --project my-project python scripts/aos.py propose "for [audience] doing [task] convey [emotion] via [form]"