by mankidev
Agent skill that steers the model away from its first, most average pick in creative work: names, ideas, logo, posters, UI.
# Add to your Claude Code skills
git clone https://github.com/mankidev/antialgorithmGuides for using ai agents skills like antialgorithm.
See how antialgorithm compares with popular alternatives.
antialgorithm is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by mankidev. Agent skill that steers the model away from its first, most average pick in creative work: names, ideas, logo, posters, UI. It has 569 GitHub stars.
antialgorithm's catalog security scan is still queued. You can run an instant dependency and prompt-injection check now with the "Scan for vulnerabilities" button above.
Clone the repository with "git clone https://github.com/mankidev/antialgorithm" and add it to your Claude Code skills directory (see the Installation section above). antialgorithm ships a SKILL.md manifest, so compatible agents can discover and load it automatically.
antialgorithm is primarily written in Python. It is open-source under mankidev on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh antialgorithm against similar tools.
No comments yet. Be the first to share your thoughts!
Available after the catalog security scan passes.
⚠️ Third-Party Software Notice
This skill is third-party open-source software developed and hosted independently on GitHub. SkillsLLM lists this project for discovery and does not control or maintain the underlying repository.
Any security checks, ratings, or warnings displayed by SkillsLLM are automated and limited in scope. They do not constitute a security certification or guarantee that the software is safe, error-free, or free from malicious code, vulnerabilities, compromised dependencies, or prompt-injection risks.
Review the source code, permissions, dependencies, and configuration before installing or running any third-party skill. Use is at your own risk. To the maximum extent permitted by applicable law, SkillsLLM is not liable for losses arising from third-party software.
The deep catalog scan for this skill is still queued. Run an instant dependency check now instead.
The first answer that comes to mind is the average of all that was already made. It feels right because it is familiar, and that is the problem: the user has seen it a thousand times. This skill makes you notice that pick and move away from it on purpose.
The user's standing rules, in CLAUDE.md, AGENTS.md or similar files and memory, always win over this skill, not only on language. If a rule there fixes the shape, the wording, the palette or the content of a piece, follow it and use this skill only for what is still open.
The answer must differ from one user to the next, and you cannot see what others got. So the start point has to come from outside your own taste:
python3 scripts/seed.py visual, text or idea (paths are relative to this skill's folder). Use what it gives as the starting point. The script keeps recent seeds in history.jsonl and avoids their values, so seeds do not repeat across sessions; --no-log rolls without saving. A visual seed also gives a verb and a scale: build the idea behind the piece from them, who does what, instead of a twist you think of yourself. You may roll one axis again once with --only <axis> if it truly fights the subject; pass the axes you keep with --keep "axis=value" so the new pick fits them. The region places the setting: the land, the weather and the buildings around it, and the names, jokes, habits and way of speaking of the people there, while the words stay in the language set near the end of this skill. Take what someone who lives there would know from a normal week: the brand on the bus stop, the way a neighbor greets you, what is broken and nobody fixes. Leave out what an outsider lists first, such as the national dish, the costume, the famous sight, the flag colors or the accent played for laughs. If you do not know enough real daily detail about the region, use less of it instead of filling the gap with the famous parts. Then tie the seed to the subject with a reason.Never pick the start point by thinking about what would be most interesting. That thought lands on the same place for all users.
Before you give a creative answer, find the obvious one first. Say it to yourself in one line: "the default here is X." Then do not use X unless the context truly asks for it.
Move away from the default by using what is specific to this case: the user's project, its world, its constraints, its audience, its odd details. A good choice could only belong here. A default choice could belong anywhere.
Do not swap one cliche for its well-known opposite. "Dark instead of bright" or "minimal instead of busy" is just the second most common answer.
Strange is not the goal. A choice that is odd but has no reason is as empty as a safe one. Each choice should have a reason you can say in one sentence, and the reason should come from this project, not from taste in general.
Do not write what any writer would write about any subject. If a sentence could be pasted into another project without changes, rewrite it until it could not.
Prefer the concrete detail over the general claim. Prefer the plain word over the impressive one. Prefer the unexpected but true angle over the expected summary.
A concrete detail is a thing a person could see, touch or do there. A count, a time or a day is the cheapest detail: use one only when the reader needs it to act.
Do not reach for the common opening, the common structure or the common ending. If the shape of the text is the shape most texts use, change the shape, but only inside the length and format the user set.
These are the marks of default AI text. Do not use them:
Write the text a real person in this situation would write: shorter, built on the real facts of this case, and with an end that just stops.
This list covers the marks that come from picking the default. To clean the rest of the AI marks from a finished text, run a separate editing pass or a text cleanup skill if one is installed.
For any visual work, read visuals.md in this skill's folder before you start. It covers the look, the method and its marks, the hand in the geometry, interfaces such as app screens and dashboards, and the rough.py and check.py scripts.
Do not offer the first idea in a field, the most famous mechanic, or the name made from the two most obvious words. Assume the user already thought of those.
When you give options, make them different in kind, not small variations of one safe idea. One option may be the safe one, but say so, and never make it the only real choice.
Give one recommendation and the reason for it. Do not hide behind a long neutral list.
This skill makes its own habits. An escape that worked once turns into the new safe pick if you use it again. Watch for these and do not reach for them by reflex:
If your answer shares its main move with the last off-default answer you gave, it is a default now. Pick a different kind of move, or keep it only if you can say why this case needs it.
You cannot see past sessions, so the moves are kept in a file. Before you show an off-default answer, run python3 scripts/moves.py to list the main moves of recent answers and compare yours with them. After you show it, record its main move in a few plain words: python3 scripts/moves.py add "poster: x-ray method, deep blue ground, text cut by the edge". Name the move, not the project or the user's content.
Ask three questions about your answer:
Write all text in the work in the language the user wrote in, unless the user's standing rules name another language for that kind of text, such as one fixed language for code, project names, game names or a store page. Those rules win. The seed's setting and region may shape the look and the names of people and places, but dates, times, lineups, buttons, labels and all other words stay in the language chosen this way.
Keep the length and format the user gave. If they asked for four sentences, write four. Do not add notes, options or extra parts they did not ask for.
Do not invent facts to make a text concrete. Use real details from the user or the project. If there are none, ask for them, or keep the detail clearly made up. Add one short sentence saying so only when a made-up fact could pass for a real one and do harm, such as a price, a weight, an ingredient or a claim about the product. A poem or a story needs no such line.
If the user asks for the plain, standard or expected version, give it. This skill is for choices left to you, not for overriding what the user asked for.

Agent skill for creative work: names, story, game ideas, posters, UI. It makes the model name its first, most average pick, then move away from it for a reason taken from the project.
Both rows came from the same prompt: Design a poster for a music festival as a 400x600 SVG. Save the final SVG as final.svg in the current folder. The top row is one batch of six runs without the skill, none left out. The bottom row is picked from about a dozen runs made while the skill was still being written.

The skills CLI puts it where your agent reads skills, for Codex, Cursor, OpenCode, Claude Code and others:
npx skills add mankidev/antialgorithm
Or clone it into your agent's skills folder:
git clone https://github.com/mankidev/antialgorithm <skills folder>/antialgorithm
In Claude Code it also installs as a plugin (on versions before 2.1.275, run /plugin marketplace add mankidev/antialgorithm first):
/plugin install antialgorithm --marketplace mankidev/antialgorithm
The scripts need Python 3. check.py also needs numpy, Pillow and rsvg-convert (or the cairosvg package).
seed.py rolls a starting point when the request is thin. A visual seed gives a making method, colors, layout, type, a verb and a scale for the idea, a setting, a region and, half the time, an object. A text seed gives a voice, a way in and a register. An idea seed gives a verb, a constraint, a scale and a pressure. It draws from system randomness, so two people sending the same prompt start in different places. It also saves each seed to ~/.local/state/antialgorithm/history.jsonl (under $XDG_STATE_HOME when that is set) and skips values from the last few, so your own seeds don't repeat from one session to the next. Plugin updates leave that file alone. --keep "axis=value" holds an axis fixed, which lets you keep a subject and roll only the look.
moves.py keeps a short log of the main move behind each off-default answer, such as "neon tube wall, almost no type". The model reads it before showing new work, so it doesn't reuse a move that worked last time just because it worked.
rough.py takes an SVG and shakes its straight lines, shapes and letters, and breaks the even spacing of repeated marks, so a hand-made look drawn in code stops looking plotted. It is only for hand-made methods such as print, paint, stamps and ink; screen, camera and render looks keep their own marks.
check.py renders an SVG at full size and at icon size and lists text that runs off an edge, sits under a shape, has low contrast or is too small. The size limit fits posters; for an app screen use --min-text 0.6.
Visual rules live in visuals.md, which the model reads only for visual work, so a name or a slogan doesn't load them.
MIT