ahd
// artificial human design
v0.1.0
·
2026-04-24
a framework for designing against ai‑generated slop
you already know the shape of the problem. another landing page with a centred hero, a pill badge, two buttons, three equal feature cards, and a purple-to-blue gradient on the word that shall not be gradiented. ahd is a brief compiler, a linter, and an eval harness that treats design taste as something you can specify, test, and diff.
pipeline
brief.yml ──▶ [compiler] ──▶ prompt ──▶ [model n]
│ │
▼ ▼
token graph ─────────▶ [28-rule linter]
│
▼
[vision critic] ──▶ score
│
▼
diff.md
run the eval
one-liner. ten samples, one brief, the specs you trust.
bash$ ahd eval-live swiss-editorial --brief briefs/landing.yml --models <specs> --n 10
what ships
in the repo, today, offline.
- brief compiler
- 28-rule linter
- eval harness
- vision-critic scaffold
- mcp server
- editor plugins
- eight tokens
what is gated
external resources required.
- live frontier-model calls — needs api keys
- live vision critique — needs multimodal key + screenshot pipeline
- standalone npm packages for the editor plugins
- additional tokens