TasteGraph Skill

Local-first personal taste graph for aesthetic, creative, design, and cultural preference.

Build a structured, versioned TasteGraph, refine it from images and feedback,
export system prompts for image models, and consult it before you design or generate.

What is a TasteGraph?

A TasteGraph is a living multi-dimensional preference graph:

• Core identity roles (who the taste serves - not legal identity)
• Emotional / aesthetic registers ("planets")
• Hard loves with evidence
• Hard antis (quiet dissatisfaction generators)
• Motion / tech preference stacks
• Density-by-surface and color temperature rules
• Process values (how taste becomes finished work)
• Quick tags, collaboration protocol, and a single north star sentence
• Optional affinity graph (nodes + edges)

It is not a vibe paragraph, a fixed brand kit, or a cloud profile.

Privacy first

• Files stay on your disk unless you choose otherwise
• Schema and skill forbid PII in graph content
• Examples are synthetic / generic
• Evidence log entries are anonymized summaries

Install (Grok / Agent Skills)

Option A - copy the skill folder

Copy skill/tastegraph/ into your agent skills directory, for example:

~/.grok/skills/tastegraph/

Restart or reload skills so the agent sees the new description.

Option B - clone this repo

git clone https://github.com/Pitchfork-and-Torch/tastegraph-skill.git
# then copy or symlink skill/tastegraph into your skills path

Python helpers (optional but recommended)

cd tastegraph-skill
python -m pip install -r requirements.txt
python skill/tastegraph/scripts/validate_tastegraph.py examples/sample-tastegraph.json

Quick start (with an agent)

Say things like:

• "Build my TasteGraph from these images"
• "Update taste with this feedback"
• "Export a TasteGraph system prompt for image generation"
• "Run a GOOD/BAD taste test from my graph"
• "Consult the TasteGraph before redesigning this landing page"

The skill will scaffold tastegraph/tastegraph.json, validate it, and keep a changelog.

Quick start (CLI only)

# validate
python skill/tastegraph/scripts/validate_tastegraph.py examples/sample-tastegraph.json

# export markdown dashboard
python skill/tastegraph/scripts/export_tastegraph.py examples/sample-tastegraph.json \
  --format md --out my-TasteGraph.md

# export system prompt for Imagine / other models
python skill/tastegraph/scripts/export_tastegraph.py examples/sample-tastegraph.json \
  --format system-prompt --out taste-system-prompt.txt

# local HTML dashboard (no network)
python skill/tastegraph/scripts/render_dashboard_html.py examples/sample-tastegraph.json \
  --out tastegraph-dashboard.html

Brand

Asset	Use
brand/icon-512.png	Primary mark (README, docs)
brand/icon-1024.png	High-res / app icon
brand/logo.png	Small 128px
brand/social-preview.png	GitHub / link unfurl 1280x640

Mark language: gold constellation graph on charcoal - one hero node, living edges, zero effect soup.

Repository layout

tastegraph-skill/
  README.md
  LICENSE
  CONTRIBUTING.md
  requirements.txt
  brand/                     # icon + social preview
  schemas/tastegraph.schema.json
  examples/
    sample-tastegraph.json
    sample-TasteGraph.md
    sample-dashboard.html
  skill/tastegraph/
    SKILL.md
    assets/                  # skill-local copies of the mark
    references/
    scripts/

Schema highlights

Canonical schema: schemas/tastegraph.schema.json

Required top-level fields:

• metadata (schema_version, graph_version, created, updated, domains)
• hard_loves
• hard_antis
• north_star

Optional: identity, registers, motion stack, density, color, process values, tags, collaboration protocol, evidence log, affinity graph, changelog.

Collaboration protocol (default)

1. Classify register / surface
2. Load matching nodes and antis
3. Distance check against sibling work
4. Prove once and log outcome

Aesthetic defaults for skill UIs

When this project generates UI or docs chrome:

• Object-as-hero clarity
• Intentional density
• Zero effect soup (no glass + beam + aurora + confetti stacks)

License

MIT - see LICENSE. Attribution required.

Support

Use GitHub Issues on this repository. Do not send private personal data in issue reports.