auteur: a Claude Code skill that gates a landing page like a film shoot
The Claude Code skill that directs a website like a film. Commit-sheet, generated assets, build, and an executable anti-slop linter that gates every ship.
At a glance
- What is it?
- auteur turns a website build into a pipeline of locked art direction, locally generated assets, and an executable anti-slop linter. It is worth adopting if you ship agent-built pages and want the taste decisions checked by code, not by vibes.
- Who is it for?
- Adopt auteur if you already build pages with an agent and want art direction fixed before code plus a linter that fails the build on slop; skip it if you need a conventional CMS theme or you cannot run Node and Playwright locally. Before trusting it, open the live gallery, clone the repo, and run npm run slopscan against one of your own pages to see whether its rules match your taste.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 44 days ago.
- What is it written in?
- Mainly JavaScript, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 19, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
What auteur actually solves, and for whom
Most agent-generated landing pages fail in the same way. The model writes a hero, a gradient, a three-card feature grid, and calls it done. Nothing was decided before the first line of markup, so nothing holds the page together. auteur's answer is to move the decisions earlier. The README states that art direction is locked before a line of code, assets are generated by local CLIs, and every ship is gated by an executable anti-slop linter plus a real motion check.
The audience is narrow and specific. You need to be working inside an agent that reads a SKILL.md file, and you need to care about how the page looks. The README lists Claude Code, Codex, Cursor, OpenCode, Gemini CLI, Windsurf, Cline, Goose, Copilot, Hermes, Kiro, Roo and OpenHands, and points at skills.sh for a longer list. If you write your CSS by hand and never let a model near the layout, auteur has nothing to offer you. If you hand a brief to an agent and then spend an hour undoing its default taste, this is the tool aimed at that hour.
The pipeline: commit-sheet, assets, build, gate
The README describes a four-stage flow: commit-sheet, assets, build, gate. The commit-sheet is the first stage, and its name is the point. Before generation starts, the direction is written down and frozen, which is what stops the model from re-deciding the palette halfway through the page.
Assets come next, and the README is explicit that they are produced by local CLIs rather than fetched from a hosted service. The repository ships a scripts/ directory with eight entry points wired into package.json: slopscan, motionqa, shoot, refscout, moodboard, source, systemscan and chromadiff. Their names map onto the stages. refscout and moodboard sit on the direction side, source handles sourcing, shoot captures the built page, and slopscan, motionqa and systemscan are the gates.
The gate stage is the part the project treats as its identity. The README's own heading for it is "The gates. This is the point." The CI badge is labelled anti-slop gate, and the ten live sites are each described as passing the skill's own linter at 0 fails / 0 warns. That is a self-imposed bar, not an external benchmark, but it is a checkable one: the linter is a script in the repository, and the sites are live.
Installing auteur and running your first gated build
The README gives an install section headed "Install (30 seconds)" and states the skill is roughly 1MB with no dependencies, no API keys and no build step. The fastest path is the skills CLI, which the README says detects what you have installed and writes to each agent's skills folder.
npx skills add agiwhitelist/auteurIf you are in Claude Code specifically, the README offers a plugin route that installs and updates in place.
/plugin marketplace add agiwhitelist/auteur
/plugin install auteur@auteurFor any other agent that reads a SKILL.md, the README falls back to a shallow clone into the skills directory.
git clone --depth 1 https://github.com/agiwhitelist/auteur ~/.claude/skills/auteurWith the skill in place, the README's own example prompt is short: ask for a cinematic landing with auteur. The agent then runs the pipeline and returns the site. To drive the gates yourself, the package exposes them as npm scripts, and the README's claim of 0 fails / 0 warns is the output you should expect to see from a passing page.
npm run slopscan
npm run motionqaThe Playwright devDependency at ^1.62.1 is what the motion and screenshot scripts need, so a first run will pull a browser binary.
Where auteur breaks down
The gates are the selling point and also the sharpest limitation. A linter can only encode the taste of the people who wrote it. The README describes the tenth live site, HORO, as a portability test in which Kimi K3 was handed SKILL.md and a brief, saw none of the other nine sites, and produced something lit, serif and green in a category whose reflex is black and gold. That is a good story about the skill travelling between models. It is also a warning: if your brand is black and gold, the same rules that flagged the category reflex will now be working against you.
The local-CLI asset generation is a second constraint. Nothing in the README suggests a hosted renderer or a cloud asset service, so the quality of what you get depends on what is installed on your machine. The tier list makes the ceiling visible: FLUX, STATIC, SWARM and DRIFT are real-time WebGL, ATLAS and ABYSS are video-scrub and depend on reference/scroll-flight.md, and TRUE NOON is described as 145KB with zero raster assets. Those are three different production tiers with three different cost profiles, and the README does not present a fallback for a machine that cannot run the heavy ones.
Finally, this is the wrong tool for a content site. auteur is built around a directed landing page or a small multi-route product surface. PROOF is described as five routes gated for design-system drift, which is the upper end of what the gates are shaped for. If you are shipping a documentation portal or a blog, the commit-sheet and motion checks are overhead you will pay for and never use.
How it differs from a component library or a page builder
The nearest familiar alternative is a component library or a template kit: you pick a hero, a pricing table and a footer, and the visual decisions arrive pre-made. auteur inverts that. There is no component catalogue in the repository listing. The top level holds SKILL.md, DESIGN.md, PRODUCT.md, a reference/ directory of recipes and a templates/ directory, and the README points at reference/scroll-flight.md as the recipe behind the video-scrub tier. Direction is authored per project rather than selected from a shelf.
The second difference is verification. A component library gives you a rendered result and no opinion about whether the page is any good. auteur ships the opinion as a script. slopscan, motionqa and systemscan are runnable, and the CI workflow runs the anti-slop gate on the project itself. That is a meaningfully different posture from a design system, where the guardrails live in a style guide that nobody executes.
The cost of that posture is that you inherit someone else's aesthetic judgements as build failures. A component library never fails your build because it dislikes your type scale.
Maintenance, licence and what upgrading costs
The repository is not archived, and the last push was on 2026-08-06. Three releases landed within roughly a day of each other in early August 2026: 1.2.0, 1.3.0 and 1.3.1, the last titled "a state is not a variant". The version in package.json is 1.3.1.
That release pattern cuts both ways. Rapid iteration on a skill file means the rules you tuned your page against can move, and because the gates are executable, a rule change can turn a passing page into a failing one. The plugin route updates in place, which is convenient and also means you may not notice the shift. If you pin anything, pin the clone and read the diff on SKILL.md and the scripts before you pull.
The licence is MIT, declared in both LICENSE and package.json. MIT is permissive: it allows commercial use and modification, and it requires that the copyright notice and permission notice be preserved. That is the extent of what the repository states, and it is not legal advice. The sites under assets/ and the live gallery are separate works from the skill itself, and the README describes HALE's geometry and HDRI as CC0, which is a different licence from the MIT covering the code.
What to check before you commit to it
Open the live gallery first. Ten sites are linked, each built by the skill, and the README is unusually direct about this being the proof rather than the pitch. If none of the ten look like something you would ship, the linter's rules are not your rules and no amount of installation will fix that.
Then read SKILL.md. It is the product. The install is a copy of a markdown file plus a scripts directory, and the README says as much when it calls the skill a SKILL.md plus reference recipes and a few runnable scripts. Everything the agent will do comes from that file, so it is the thing to evaluate, not the badge count.
Finally, run the scanner on a page you already like. If slopscan flags it, you have learned where the two taste systems disagree, and that is the real adoption question. If it passes clean, auteur will mostly stay out of your way.
Editorial conclusion
Adopt auteur if you already build pages with an agent and want art direction fixed before code plus a linter that fails the build on slop; skip it if you need a conventional CMS theme or you cannot run Node and Playwright locally. Before trusting it, open the live gallery, clone the repo, and run npm run slopscan against one of your own pages to see whether its rules match your taste. The SKILL.md file is the whole product, so read it before you install it.
Frequently asked questions
What is auteur?
auteur is an Agent Skill for Claude Code and other agents that reads a SKILL.md file. It directs a website build through a commit-sheet, locally generated assets, a build, and an executable anti-slop linter that gates the ship.
What is auteur meaning?
In this project the word names a working method rather than a film credit: art direction is locked before code, assets are generated by local CLIs, and every ship passes an anti-slop linter plus a motion check.
What does being an auteur mean?
The README does not define the film term. It borrows the idea of a single controlling direction, which in auteur's case is a commit-sheet written and frozen before any generation begins.
What is auteur theory in film?
The README does not discuss film theory. It uses the word for locking a visual direction before code is written, the way a director fixes a look before shooting.
What is auteur director?
The README does not cover film directors. The closest analogue in the repository is the commit-sheet stage, which fixes direction before the agent writes any markup.
Community notes