Visualize

/visualize

Turn taste.md, taste.vision, and the current request into an inspectable comprehension artifact. This skill is for understanding and alignment, not implementation.

Contract

• Read taste.md and taste.vision first.
• If either file is missing, stop and route to /tastebootstrap.
• Read $ARGUMENTS, active SPEC.md when relevant, and obvious public repo

anchors. Do not read .env, .env.*, .claude/settings.local.json, private customer artifacts, credentials, audit logs, or secret files.

• Create one ignored run folder:

mkdir -p .taste/visualizations
RUN_DIR=".taste/visualizations/$(date +%Y%m%d-%H%M%S)-$(printf '%s' "$ARGUMENTS" | tr '[:upper:]' '[:lower:]' | sed -E 's/[^a-z0-9]+/-/g; s/^-|-$//g' | cut -c1-48)"
mkdir -p "$RUN_DIR"

• Always write $RUN_DIR/visualization.md.
• Do not implement, edit product code, promote assets, commit, push, deploy, or

mutate SPEC.md.

• Keep the package no-secret: do not include credentials, private customer

artifacts, or raw environment values.

• When the chosen mode is HTML, write the HTML as a companion artifact inside

RUN_DIR and keep visualization.md as the canonical human contract. The HTML must reference the canonical source files it renders or explains.

Understanding Card

Begin visualization.md with:

## Understanding Card

- Intended user:
- Product promise:
- Core experience:
- Non-goals:
- Visual or operational tone:
- Likely misunderstanding risks:
- Taste anchors:
- Vision anchors:

Surface Classification

Choose exactly one primary surface:

• frontend-web
• mobile-app
• backend-api
• data-dashboard
• game
• agent-runtime
• infra/devtool
• docs/brand

Record the chosen surface and a one-sentence reason.

Artifact Choice

• frontend-web, mobile-app, product UI: produce a production-quality

UI/product mockup image when the current runtime exposes image generation. Use the no-image fallback when image generation is unavailable: write a precise image prompt plus an HTML/SVG/Mermaid schematic or UX storyboard.

• data-dashboard: produce a dashboard mockup image when available; otherwise

a chart layout, metric hierarchy, and workflow schematic.

• game: produce gameplay screen/key-art prompt when image generation is

available; otherwise a gameplay loop diagram, screen-state storyboard, or interaction map.

• backend-api, infra/devtool, agent-runtime: do not force a fake UI image.

Produce an operator/user journey map, architecture schematic, state-flow, API experience narrative, Mermaid diagram, or HTML/SVG artifact.

• docs/brand: produce a brand/page/content system visualization or narrative

structure.

For HTML mode:

• prefer static, self-contained HTML with inline CSS
• avoid remote scripts, remote stylesheets, remote fonts, analytics, and external

asset fetches unless the user explicitly asks for a shareable hosted artifact

• include `, , `, and visible links or

labels for canonical files such as visualization.md, draft-SPEC.md, or approval.json

• do not put requirements, approvals, source ledgers, or verification evidence

only in HTML

• if the HTML is interactive, include a copy/export affordance that emits

Markdown, JSON, or prompt text the operator can paste back into the harness

If an image is generated, save or reference it inside RUN_DIR and record the actual path. If only a prompt, diagram, or markdown artifact is produced, say so plainly. Never claim an image was generated when it was not.

Required Sections

visualization.md must contain:

# Visualization: [short title]

## Understanding Card
## Source Inputs
## Surface Classification
## Chosen Artifact Mode
## Artifact Paths
## Image Prompt Or Diagram Source
## Assumptions
## Mismatch Risks
## Review Rubric
## Outcome

The review rubric must check:

• taste alignment
• vision alignment
• user/operator experience clarity
• component or workflow fidelity
• text/label correctness when visual text is involved
• privacy and no-secret posture
• whether the artifact is useful for implementation decisions

Output

Return a concise summary:

## Visualization Complete

- Surface: [classification]
- Mode: [image / prompt-only / markdown / Mermaid / HTML / SVG]
- Run Folder: [.taste/visualizations/...]
- Primary Artifact: [.taste/visualizations/.../visualization.md]
- Generated Image: [path / none]
- Implementation Performed: no
- Open Questions: [none or focused list]

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: waitdeadai
• Source: waitdeadai/minmaxing
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.