# Vibes Algorithmic Art — first release validation

September 28, 2026. [Use the skill](https://thevibesarepeople.com/skills/vibes-algorithmic-art/), [read its instructions](https://thevibesarepeople.com/skills/vibes-algorithmic-art/SKILL.md), or [try the result of an independent agent run](https://thevibesarepeople.com/skills/vibes-algorithmic-art/examples/calm-currents.html).

## What is delivered

An original creation skill, a six-file downloadable ZIP, a dated seven-method catalog snapshot with each credit’s review status, a visual-language reference, taxonomy foundations and a completed offline Canvas starter. The separate Descramble contribution skill continues to handle database proposals. The package does not contain Anthropic’s template or copied skill text.

Ordinary visual requests can start a draft without an interview. Established method names come first; a new personal or process label remains an explicitly proposed name. People, roles, institutions and actual mechanisms stay distinguishable. The skill does not automatically rename catalog records or make public submissions.

The starter implements an interpolated seeded direction grid with finite paths. It is not Perlin noise or curl noise. Source links identify explanatory context, not sole invention of a visual appearance. More elaborate or animated requests require the agent to adapt the implementation and test it.

## Tests actually run

- Skill frontmatter, naming and scaffold validation passed.
- Starter: 13 browser checks passed, including each control’s pixel effect, repeatable regeneration, PNG signature and dimensions, JSON restoration, 20 invalid-input cases, offline HTML reopening/editing/re-export and mobile widths 320/390/768px.
- Independent forward task: create sparse fine trails in broad calm currents for a class, adjustable density/scale, revisitable state, image and working-file sharing, and a method credit.
- That agent delivered a real offline HTML artwork without a clarification round. It adapted the original starter’s directions around a shared heading to better meet the visual request and changed the algorithm identifier accordingly. It verified Tyler Hobbs’s explanatory essay, kept method credit separate from code authorship, and passed 15 browser checks.
- The public example differs from that tested file only in removing a pre-publication unavailable-link note and expanding its recipe disclosure. Its drawing and state/export code are unchanged.
- Vocabulary explorer: six looping illustrations, effective sliders, adjustable duration, pause/restart, reduced-motion behavior and mobile fit passed browser checks. Illustrative exercises remain clearly separate from complete implementations of every linked method.

Executable starter and vocabulary checks are in `tests/art-studio-browser.mjs` and `tests/vocabulary-loop-browser.mjs`. The package is built deterministically with `scripts/package-art-skill.py`; its public manifest records version, contents and SHA-256.

## Baseline and honest comparison

A separate agent used the original Anthropic skill for the same classroom request. It also delivered a working artwork and passed checks for controls, state preservation, PNG/HTML export and mobile fit. Its piece used network-loaded p5.js and fonts; our forward result ran without network requests.

This is one task in each condition, not a statistical benchmark or a blinded aesthetic comparison. It does not prove that the new skill always makes better art. Its concrete additions are the linked research corpus, visual terminology, explicit proposed-name policy, tested offline starter and a route into the taxonomy. The baseline success also shows why an incomplete source template should not be treated as proof that its generated outputs fail.

Verification used installed Chrome. Pixel identity was tested within that browser, not across browser engines. No paid model API calls were made for these development tests; visiting users employ their own AI account. The site’s capped contribution reviewer is unchanged.