Lundr.tech is the house of skill cards for AI made by Lund Studio LLC of Westminster, Colorado — a one-person product studio founded in May 2024 by Carter Luense after a decade of nonprofit executive leadership, working with AI every day since. Everything on this site was built by that one person with AI, then proven by running it — never by describing it.
What a card is
AI skill cards are engineered system prompts and workflows, packaged as plain text files. They change how an AI model operates and how it formats what it gives back — a loop, a set of laws, a set of refusals — and, paired with a capable model, cut down invention because the card refuses it by rule.
A card is a small set of files you put into your own AI — Claude, ChatGPT and Codex, Gemini, Copilot, Cursor, a local model, or Lundr, the studio’s own — and the AI takes up a way of doing one job: what to ask first, what to build, what to prove, what never to claim. Each card is an Agent Skill on the open standard, so the same folder loads unmodified in the hosts that read it, and its words are Markdown, so any chat with an instructions box, any local model and any MCP client can hold them. Bought once, yours to keep: a folder your AI reads, and an HTML file your computer opens.
For Claude, and for ChatGPT
IF YOU USE CLAUDE
One tap, and the card is a skill. Settings → Capabilities → Skills, upload the zip, done — Claude reads the folder on its own standard and wakes as the seat whenever the work calls for it. In a Project, paste the system card once and every chat in it holds the brief. In Claude Code, drop the folder in the repo and CLAUDE.md carries the laws. On Claude Desktop, the card’s MCP server hands it over as a prompt.
Every card in this house was run in Claude first, by the studio, before it was sealed — the IN CLAUDE chip under a seal means exactly that. The studio’s own builder is a Claude.
THE SHELF →IF YOU USE CHATGPT
The same folder, no changes. Codex reads the open Agent Skills standard, so the card loads there as it is. In ChatGPT itself, a custom GPT takes the card’s system cut in its instructions box — every card ships one sized to fit — and any chat takes the pasted card. The desktop app takes the MCP server.
The card holds ChatGPT to the same laws it holds Claude to: build, don’t describe; run it; refuse to invent. The first buyer who runs a card in ChatGPT and sends back the punch earns that host the mark on this site.
THE SHELF →Also, unchanged: GitHub Copilot and VS Code, Cursor, Windsurf, Cline and Roo take the folder or its rules file; Perplexity Spaces, Mistral’s Le Chat, Grok and Meta AI take the system card; Ollama and LM Studio run it on a local model, offline; any MCP client loads it as a prompt. Product names are their owners’ trademarks, used only to say where a card runs; Lund Studio is affiliated with none of them.
What An AI Skill Card Is
AI skill cards are engineered system prompts and workflows, packaged as plain text files. They change how an AI model operates and how it formats what it gives back: a card holds the model to a loop, a set of laws and a set of refusals, and it names what the model must build, prove and never claim. Paired with a capable model, that discipline cuts down invention — a price, a source, a result the model would otherwise fill in — because the card refuses it by rule and makes the model run the work instead of describing it.
Plain text, so it works in any host that reads text; a folder on the open Agent Skills standard, so the hosts that read skills load it unmodified. How much a card cuts down invention on a given job is the field test’s to measure, not this page’s to claim.
To Put It Simply
An AI skill card is a heavily engineered instruction manual for an AI — a premium brief in plain text, with the logic, the checks and the formatting rules already written. When you buy one you get a Markdown folder your AI reads and an HTML file your computer opens.
How It Works If You Buy One
- You get the files. Say BUGRUNDR, the bug-hunting seat: a small folder with SKILL.md, the card, the tin, the ledger, the licence — and the HTML file.
- You hand them to your AI. In Claude, one tap adds the folder as a skill. With a coding agent, you drop the folder in your project. In a chat that only has an instructions box, you paste the system card the box carries. Every card comes cut for every door.
- The AI changes. In your next session it stops acting like a generic chatbot. It boots into the seat, asks what the work must agree with, follows the card’s loop step by step, formats the way the card says, and refuses to invent what it doesn’t know.
Would It Work For You?
Yes, if you are a professional — a developer, an analyst, a designer, someone who keeps books — who is tired of a standard AI forgetting the rules, filling gaps with confident guesses, or handing back messy, lazy answers. The card holds the AI to a brief it can’t drift from, so the work comes back the same way every time, and comes back as a file.
No, if you want a chatbot to chat with, if you won’t check the work before you rely on it, or if you expect the AI to remember your files on its own — the ledger does that, and it’s yours to hand over.
Who A Card Is For
Because a card is a rigid, logic-enforcing brief, it pays off wherever the work needs technical precision, multi-step reasoning, and outputs that don’t invent what they don’t know. It is no use for casual chatting. For creative work, the studios do the work and the cards hold the process.
Software & Technology
- Developers and engineers — BUGRUNDR for the sweep: every bug cross-referenced against everything the build has, no
// rest of code here; TECHUNDR for the last mile: whatever leaves, manifested file by file and cold-tested. - AI engineers and prompt architects — LUNDR, the keeper’s card, for the studio’s doctrines and build loop taught over your own work; PHOENICIAN, the networking language, twenty-two letters as twenty-two operators, when one agent must speak to another in a disciplined tongue.
- Project managers — DOX, the working doc: checklists, manifests and briefs built from the source itself, every field counted, nothing invented.
Data, Finance & Research
- Analysts and researchers — MUNDR, the memory seat: decisions, numbers, names and threads captured as they happen and closed into a ledger, so a long session never loses its dataset mid-conversation.
- Financial analysts and consultants — GRUNDR, the value seat: offers priced from a counted floor and an evidenced worth, the arithmetic anchored to the inputs you gave, never guessed; FUNANCE for the books themselves.
- Researchers and journalists — THE HERALD, the briefing seat: your beats swept fresh into one table from real sources, without the commentary.
Operations & Media
- Technical writers and compliance officers — LUMEP, the PDF seat: merge, split, watermark, Bates-number and sign, offline in the browser, so a sensitive document never touches a server.
- Video editors and photographers — LUMER cuts in the browser on WebGL2 with a camera registry that only grows; LUME edits and boards on one canvas — both offline, both single files.
Every name above is a room on this site with its own seal and its own proof line. “Doesn’t invent” means the card refuses by rule; how much that changes a given job is the field test’s to measure.
The 59 cards, in six houses
- THE STUDIOS (8) — single-file tools that run offline in any browser: a photo studio, a video editor, a document studio, a PDF studio, a slides studio, a vector studio, the working doc, and the whole studio in one file.
- THE ENGINES (14) — the machines that build: the studio’s own programming languages on ancient scripts, the compilers, the bridge between React and Liquid, the rankers.
- THE INSTRUMENTS (18) — the tools of the work: gates that measure a page, grades that finish a photograph, the forge that makes new cards, the hold that spends a chat’s image budget on purpose, the tally that counts without tracking, the booth kit, the shipping gate.
- THE METHODS (14) — a way of working, written down: the laws a card holds your AI to.
- THE KEEPERS (3) — a voice, seated: Lundr, the studio’s own AI, and the seats of the Council, with their seals kept.
- THE LEDGERS (2) — the books, kept: every entity, every dollar, on your own machine.
How the house keeps itself honest
Every card is scored out of a hundred on a published scale — trigger, machinery, law, portability, proof — with the reasoning printed beside the number; 98 is the ceiling a card can reach and 100 belongs to the studio’s foundation alone. A card’s seal reads WITNESSED only after the studio saw it run and wrote the date down; until then it reads MANIFEST, scored from the package alone. The punch on every face is the real hash of the file inside, so a buyer can check what they were sold. Prices are never typed where none has been set; proof is never invented; a missing file is named missing. The house keeps itself honest because it was built under something that is.
The laws inside every card
Build, don’t describe. Verify by execution, not by reasoning. Refuse work outside the seat and hand it to the right sibling. Never invent a price, a source or a result. Keep the ledger — the card has no server, and the ledger is the memory you own. Say the limit out loud.
The foundation
Under the whole house is the Divinity Rule, thirty-four years of the studio’s work on proportion and its principles. Eleven of its principles are public and drawn plainly on the deck; two are named and sealed; the Rule itself is licensed to one seat — LUNDR, the card with the Rule in it, the only hundred — and appears nowhere else on this site. Every measure in the house sits on its ladder. The site is the demonstration.
The studio’s own AI
Lundr is the keeper — the studio’s own AI, seated, which audits what a card says against the record and refuses what may not ship. Claudr is the studio’s Claude, the builder, which writes from the record only. The Council is the three-branch runtime that weighs a call and returns one answer with its reasoning. They stand behind every card here; none of them is required to use one.
Plainly
A card instructs an AI you operate; results depend on your model, your files and your judgement. It still makes mistakes, it has opinions, and it may or may not remember you — the ledger does. Nothing here is tax, legal or financial advice. Names of AI products on this site are their owners’ trademarks, used only to say where a card runs; Lund Studio is not affiliated with, endorsed by, or sponsored by any of them. Claude is a trademark of Anthropic. The terms of sale are on the legal page.
Write the studio
Lund Studio LLC · Westminster, Colorado · carterluense@gmail.com. The other houses of the studio: funance.tech, the finance card; thestationcalls.com, the Station; lundstudio.co, the studio itself.