mankidev/antialgorithm
Developer ToolsA Claude Code skill that pushes creative output past the model's average first pick for names, copy, ideas and UI.
Overview
antialgorithm is a Claude Code skill aimed at a common failure in model-generated creative work: the model defaults to its most average, first-choice answer. The skill makes the model state that first pick explicitly, then deliberately move away from it using a reason drawn from the project itself. It covers names, copy, story, game ideas, posters and UI. The repository includes a side-by-side poster example produced from one identical prompt, with and without the skill visual comparison.
Install
Install by cloning the skill into the Claude Code skills directory:
git clone https://github.com/mankidev/antialgorithm ~/.claude/skills/antialgorithm
The bundled scripts require Python 3. check.py additionally needs numpy, Pillow and rsvg-convert (or the cairosvg package).
Scripts
seed.pyproduces a random starting point for thin prompts: a visual seed (making method, colors, layout, type, a verb, scale, setting, region, and sometimes an object), a text seed (voice, way in, register), or an idea seed (verb, constraint, scale, pressure). It draws from system randomness, so the same prompt yields different starts for different people.rough.pytakes an SVG and perturbs straight lines, shapes and letters, and breaks even spacing, so code-drawn work in hand-made styles (print, paint, stamps, ink) stops looking plotted. It should not be applied to screen, camera or render looks.check.pyrenders an SVG at full size and icon size and reports text that runs off an edge, sits under a shape, has low contrast or is too small.
Constraints
Python 3 is required for all scripts. The SVG checking workflow depends on numpy, Pillow and either rsvg-convert or cairosvg. The rough.py effect is scoped to hand-made visual methods only. The project is MIT licensed.
HotRepo's take
On the HotRepo new-project board for Python, the repository recorded 6 Stars and 0 forks gained over a measured one-hour window on 2026-11-17, reaching 109 Stars total. The available data is a single short snapshot, so it supports only that day's observation, not a trend or quality judgment. What stands out is the project's narrow scope: it addresses a specific model behavior (averaging toward the obvious choice) rather than offering a general prompt pack, which makes its value easy to judge against a concrete output.
Written by AI from the project's README, updated . The project's own docs are the source of truth.
You might also like
- phuryn/product-engineeringAgent skills for non-coders to take a product from idea to paying users with coding agents.
- jakubkrehel/skillsAgent skills that give coding assistants concrete rules for reviewing and building UI, typography, color, accessibility and interface copy.
- humanlayer/skillsA collection of installable Claude Code and Codex skills for research, implementation, visual PRs, and agent-loop design.
- t3dotgg/tstackAgent skills that turn a code change into a reviewed, babysat, merged pull request for Claude Code and Codex users.
- danyuchn/claude-work-modeClaude Code plugin that switches session work modes for autonomy, delegation, and decision wrap-up.