Gridgeist skill
A skill for creating, redesigning and reviewing web interfaces: one design thesis and system instead of generic SaaS looks.
- 120 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The skill is instructions and references, and the agent edits the interface only with your permission.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/gridgeistDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The skill is instructions and references, and the agent edits the interface only with your permission.
This is third-party code. Review the repository files before installing.
What it does
Gridgeist walks an agent through three modes: create an interface, rework an existing one or only review it without edits. The agent first studies the audience, brand and the rendered interface, then states one thesis and builds a compact system contract: color, typography, grid, components, states and motion. When the direction is too broad, the skill offers options and waits for agreement, postponing edits. At the end the agent checks screen widths, focus, overflow and reduced motion and separates what it observed from assumptions. The references `design-language.md`, `system-contract.md` and `review-checklist.md` ship alongside.
Who it is for. Frontend developers and designers whose agent keeps producing look-alike SaaS landing pages and dashboards.
Good fit when
- A landing page, dashboard or documentation site is built from scratch and needs its own character.
- An existing interface is reworked without breaking behavior or brand.
- A draft is reviewed and gets prioritized findings with evidence.
- Several design skills conflict and one needs to own the direction.
Not a fit when
- You need a ready component library with code rather than a decision process.
- The task is tweaking one margin or color.
- The agent cannot open and render a page.
Example request
Use the Gridgeist skill to review this interface without editing it, give a one-line verdict and one redesign direction.Limitations
The skill defines a process and references but contains no ready interface code. It works only with agents that read `SKILL.md`, and discovery and invocation differ between products. Checks across widths need a browser or another way to render, otherwise the agent states the check was not done. Updates do not arrive on their own: run the command of your install channel. The last commit was in July 2026.
How to disable. Delete the `gridgeist` directory from the agent's skills folder or remove the plugin through `codex plugin`. Then start a new session.
FAQ
How do I install Gridgeist in Claude Code?
With `npx skills add ohmiler/gridgeist -g`. Add `-a claude-code` to pick the agent.
How do I update the skill?
With `npx skills update gridgeist -g -y`. For Codex, refresh the marketplace snapshot with `codex plugin marketplace upgrade gridgeist`.
Does the skill edit files on its own?
No, not until you allow it. Review mode leaves files untouched, and for a broad direction the skill asks for agreement first.
Which modes does the skill have?
Three: create an interface, redesign with edits and review without changing files.
How do I call the skill directly?
In agents with explicit skill invocation write `$gridgeist` in the prompt. Others can simply name Gridgeist in the text.
Discussion
No questions or reviews yet.
Related
SkillsA UI design skill: art direction per product, a shared DESIGN.md, tokens, and layout and accessibility checks in a browser.
A skill for interfaces without the generic AI look: typography, color, motion, component states and a critique of the result.
A UI, presentation and graphics design skill for coding agents: canvas drafts driven through the superdesign CLI
A Codex skill: three distinct visual concepts, screenshot review and cutting excess, so interfaces look less templated.
Meta-skill that reads a brand from a URL, name or screenshot and turns it into a full design system as a new skill
A design skill against templated AI interfaces: new pages, audits, redesigns and design extraction from a screenshot or URL
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Gridgeist (this card) | no | 120 | |
| Taste Skill | no | 114,100 | a set of skills with presets against generic interfaces; Gridgeist derives direction from brand and product and adds a system contract |
| Frontend Design (Ilm-Alan) | no | 167 | a general frontend design skill; Gridgeist separates create, redesign and review modes and requires checks across screens |
README in short
Version 1.2.0 treats visual decisions as one system and comes with a before and after comparison on a fictional project.
README badge
Original SKILL.md
--- name: gridgeist description: Use when creating, redesigning, or reviewing web interfaces that need product-specific structure, clear hierarchy, responsive composition, accessible interaction, or relief from generic AI-generated SaaS aesthetics. Adapt grid, typography, imagery, motion, and product evidence to the established brand—including technical, editorial, image-led, warm, playful, and utilitarian directions—across React, Next.js, Tailwind CSS, HTML/CSS, landing pages, dashboards, documentation, portfolios, and interactive tools. --- # Gridgeist ## Overview Craft distinctive, product-native interfaces from audience, content, behavior, and brand intent. Use grid as structural logic; make it visible only when it serves the product. ## Select a mode | Intent and authorization | Mode | Start with | |---|---|---| | Build an interface without an existing design | Create | Audience, primary task, content, and evidence | | Change an existing interface with permission to edit | Redesign | Preserved behavior, brand signals, and state inventory | | Diagnose without changing files | Review | Prioritized findings grounded in rendered evidence | Let user authorization determine whether files change. Permission to redesign does not authorize choosing among materially different visual theses. Derive direction from explicit user intent or established brand evidence; audience, task, product category, and content may inform hierarchy but do not alone confirm a visual thesis. Use the alignment gate below whenever thesis-level ambiguity remains. ## Workflow 1. **Inspect** — Understand the audience, primary tasks, brand signals, content, product evidence, routes, components, tokens, and rendered desktop/mobile UI. Inspect existing design-system sources, including `DESIGN.md` when present, and reconcile them with the implementation and rendered evidence rather than assuming they agree. Inventory important interaction states and constraints. Do not invent customers, metrics, outcomes, research, or compliance. - **Align direction before editing** — Classify the direction as **user-confirmed**, **brand-derived**, or **provisional**. Treat a coherent user direction as confirmed even when it leaves implementation choices open; variations within the same thesis are not material ambiguity. Classify a direction as brand-derived only when existing visual implementation, assets, documented rules, or repeated brand conventions converge on one thesis; product subject matter or copy alone is not enough. When a broad Create or Redesign still supports materially different theses that would change core brand expression, hierarchy, imagery, color, or motion, offer two or three evidence-grounded directions with trade-offs and a recommendation, then ask the user to choose or authorize the recommendation. While the direction is provisional, do not define the replacement system or edit implementation files. Proceed without redundant questions when intent or coherent brand evidence is clear. Do not block Review or narrow repairs; label an unconfirmed Review replacement direction provisional. - **Hard stop for provisional broad work** — A provisional direction is not an implementable direction. End the turn after the options, trade-offs, recommendation, and alignment question. Do not set a thesis, read the system contract, create tokens, edit files, start implementation, or verify a redesign until the user selects a direction or explicitly authorizes the recommendation. 2. **Set a thesis** — Write one governing sentence combining audience, primary task, structural logic, brand expression, and a product-native motif. Read [design-language.md](references/design-language.md) before choosing the direction, especially for image-led, warm, playful, or otherwise nontechnical brands. 3. **Define the system** — Read [system-contract.md](references/system-contract.md) before making new or changed system decisions. Establish the relevant color, typography, layout, spacing, shape, surface, component, state, media, motion, and responsive rules as one compact contract. Decide whether the grid should be visible, quiet, or invisible. Reuse semantic tokens and existing primitives; keep narrow repairs proportional and do not create a persistent design artifact unless authorized. 4. **Compose** — Build hierarchy before detail. Make one area dominant, align related content, vary sections within shared logic, and let the most authentic material—product UI, data, prose, imagery, artwork, code, or the primary tool—carry visual weight. 5. **Implement** — Preserve required behavior and states. Follow repository conventions, semantic HTML, keyboard and touch behavior, and existing primitives. Recompose mobile layouts rather than shrinking desktop. Avoid dependencies for simple CSS effects. 6. **Verify** — Use [review-checklist.md](references/review-checklist.md). Render representative widths and exercise primary flows, states, focus behavior, reduced motion, overflow, and dynamic content. Fix clarity and hierarchy before polish. Report what was observed separately from what remains inferred or untested. For interactive products, inventory at least default, loading, empty, error, success, disabled, and destructive states when applicable. Preserve privacy, safety, data, and platform constraints as product behavior, not optional polish. ## Coordinate companions - Keep Gridgeist as the sole owner of product and visual direction. Do not automatically combine it with another broad frontend-design or art-direction skill. If the user explicitly requests both, establish one direction owner before editing instead of silently blending their defaults. - **Explicit overlap gate** — When the user requests Gridgeist plus another broad visual-design skill, state the overlap and recommend one direction owner before selecting a thesis or editing; use Gridgeist by default for product-native, brand-adaptive work. Do not silently ignore the requested companion or run both visual theses. After ownership is clear, use only non-conflicting implementation or verification help from the other skill unless the user chooses it as owner. - Use context capabilities during Inspect, asset generation after the thesis, technical audits after implementation, and browser automation during Verify. - Treat companions as optional. Do not block work or add dependencies when they are unavailable. ## Anti-slop contract - Use one thesis and one coherent system. - Preserve or deliberately evolve the brand instead of importing a house aesthetic. - Build hierarchy through scale, position, density, rhythm, and contrast. - Give grid visibility, borders, radii, shadows, gradients, and motion defined roles. - Prefer authentic product evidence; label sample or fictional material and never fabricate proof. - Give sections or workflows distinct compositions within shared structural logic.