# Core Collage

> Find your own micro-aesthetic - your "-core", the way people say balletcore or kidcore - from a list of things you love, pinned to a mood board, then style one real thing you own in it. The language reading, the clash check, a local palette and the mood board run free in the browser; two metered lanes name the core and art-direct the thing. https://core-collage.skillsafe.ai/

## What it does

- Free, in the browser (corekit.js, boardview.js): every loved line is read against word lists for twelve 2026 visual languages (Naive Design, Type Collage, Blueprint Design, Trinket Design, Punk Grunge, Future Medieval, Distorted Portrait, Surveillance Design, Grainy Blur, Signal Graphics, Frutiger Aero, Kid Core) with their pair-well / avoid table; the page shows the leading languages, a clash warning when the two leaders fight, its own guess (cohesive / split / scattered), a local palette (your colour words first) with an ink-on-paper contrast check, and a pinned SVG mood board (ransom-note title, palette strip, type specimen, one tile per line dressed in its language) exportable as PNG, SVG and CSS variables. For the styling lane it guesses keep / tweak / drop per owned item from colour distance to the palette and language words.
- Paid lane `core` (verdict cohesive | split | scattered): a core name ending in "core", a tagline, the visual languages it blends with weights, five colours with roles (paper, ink, accent, pop, shadow), type direction described generically, one motif per line quoting the line, textures, we are / we are not, languages to keep out, the tension (split: both groups and a bridge; scattered: the seed lines), and the first thing to style.
- Paid lane `apply` (verdict go | tweak | rethink): one thing (a shelf, an outfit, a table, a desk, a room, a phone screen, gift wrap, an invitation) art-directed in the core: a premise, a keep / tweak / drop call on every item the person listed, the look (composition, light, where each palette colour goes, texture), ordered steps, up to three things worth finding or making (no products or prices), three photo shots, what to avoid.
- The page re-checks every reply: motifs must quote the person's lines and cover them all, items must be the ones they listed, colours named must be in the palette, a blend of two fighting languages must be called split, verdicts must match the calls, no internal field names in prose.
- Exports: the board as PNG / SVG (or straight to the device share sheet where the browser supports it), the palette as CSS custom properties, Markdown style notes, JSON. Saved boards live in the person's SkillSafe account (collection `boards`); boards saved as a guest move into the account on sign-in. After an edit the page names the lines added or gone since the last reply.

## Contract

- Run input: `task` (`core` | `apply`), `loves` (numbered lines); `core` adds `nots`, `colors`, `page_board` (JSON), `page_verdict`; `apply` adds `core` (JSON), `thing`, `occasion`, `owned` (numbered lines), `page_items` (JSON), `page_verdict`. See https://core-collage.skillsafe.ai/api.html
- Reply: one JSON object per task.

## Sources

- Built on the SkillSafe skills @guia-matthieu/design-trends-2026 (twelve visual languages of 2026 graphic design, with pair-well and avoid rules, for mood boards and briefs) and @rampstackco/art-direction (direct visual work across story, look, execution and standards; evaluate creative for fit).
- The page's language reading is a word-list guess and its palettes are starting points, not a style rule.