THE BOT BUILDER CARD · the maker seat · your own focused AI bot, built and tested
WAKES WHEN someone wanting their own AI bot, assistant or helper; a support, booking, club, shop or study bot; a system prompt that is not working; testing a bot before people use it
THE LOOP
NAME — say: "This is THE BOT BUILDER CARD." Ask what the work is and what it must agree with; those answers are the session's source of truth.
NEW — `python3 scripts/botsmith.py new my-bot --name "My Bot"`.
WRITE BOT.md — the job in one sentence, who it is for, the voice, always, never, when to hand off, what it knows, the limit. references/BOT-GUIDE.md walks each part.
LINT — `python3 scripts/botsmith.py lint my-bot` until there is no FAIL.
BUILD — `python3 scripts/botsmith.py build my-bot` → system-prompt.md and short-prompt.md.
TEST — write at least six tests in evals.jsonl (including refusals), put the bot's real answers in answers.jsonl, then `python3 scripts/botsmith.py eval my-bot answers.jsonl`.
SHIP — references/SHIP-IT.md: where to put it. **FIX** — every failure becomes a line in BOT.md and a test.
LEDGER — write the ledger line.
NEVER
a bot that claims to be a person
a promise no model can keep
a bot shipped untested
THE SEAL 68 of 100 · EPIC · WITNESSED · MACHINE ABOARD · SELFTEST 9/9 · 10 TESTS · WITNESSED ON REAL WORK
IT CAN BE WRONG an AI reading a text file: it can be wrong; check what matters. THE-CLAUSES.md rides in the packet and is part of the licence.
THE PUNCH · SHA-256 OF SKILL.MD · C199C00FE3B1FEB46091BE79
CITE THIS CARDTHE BOT BUILDER CARD, Edition 1. Lund Studio LLC, Westminster, Colorado. https://lundr.tech/?view=bot-builder-card. Punch C199C00FE3B1FEB46091BE79. Retrieved 2026-10-10.
What it is.
A seat that helps anyone build their own focused AI bot on a model they already have: one job, a clear voice, honest limits, tested before it talks to anyone.
Writes the whole bot in one plain file, compiles its system prompt, catches the faults that make bots fail, and grades its answers against tests.
Its laws, written into the card.
One bot, one job. A narrow bot done well beats a broad one done badly.
The whole bot lives in one plain file, BOT.md: the job, who it is for, the voice, always, never, when to hand off, what it knows, and its limit.
It says it is an AI whenever anyone asks. A bot told to pass as a person fails the lint.
It answers from what it was given, and says "I don't know" when the answer is not there. No invented prices, policies or links.
No promise a model cannot keep: not "always right", not "any question", not "guaranteed".
It is tested before it talks to anyone: at least six tests, including ones it must refuse, run again after every change.
It hands people to a person when they are upset, at risk, or ask for one.
Ship it where it fits — a Claude Project, a custom GPT, a Gem, or a small page in front of an API with the key kept on a server, never in the page.
How it works, once your AI has it.
NAME — say: "This is THE BOT BUILDER CARD." Ask what the work is and what it must agree with; those answers are the session's source of truth.
NEW — `python3 scripts/botsmith.py new my-bot --name "My Bot"`.
WRITE BOT.md — the job in one sentence, who it is for, the voice, always, never, when to hand off, what it knows, the limit. references/BOT-GUIDE.md walks each part.
LINT — `python3 scripts/botsmith.py lint my-bot` until there is no FAIL.
BUILD — `python3 scripts/botsmith.py build my-bot` → system-prompt.md and short-prompt.md.
TEST — write at least six tests in evals.jsonl (including refusals), put the bot's real answers in answers.jsonl, then `python3 scripts/botsmith.py eval my-bot answers.jsonl`.
SHIP — references/SHIP-IT.md: where to put it. **FIX** — every failure becomes a line in BOT.md and a test.
LEDGER — write the ledger line.
When it says no.
A bot that claims to be a person.
A promise no model can keep.
A bot shipped untested.
What’s in the buy.
CARD.md — the whole card · SKILL.md — the door your AI reads first · START-HERE.md — the first five minutes
MY-LEDGER.md · TIN.md · LICENSE-NOTE.md · THE-CLAUSES.md · card.svg + card-machine.svg + card.json · card.html · the doors for ChatGPT, Gemini, Cursor, Copilot and the rest, and an MCP door
PROOF · MACHINE ABOARD · SELFTEST 9/9 · 10 TESTS · WITNESSED ON REAL WORK · AUDIT 12/12 · EDITION 1 · PUNCH C199C00FE3B1FEB46091BE79
ON EVERY DOOR · A CARD FOR YOUR AI, A FILE FOR YOUR COMPUTER
In Claude: turn on code execution under Settings › Capabilities, then Customize › Skills › + › Create skill, and upload the zip as it is — START-HERE.md in the box walks you through it, and tells you how to keep the ledger. In ChatGPT, Gemini and the rest: a Project, a GPT or a Gem with the card’s files attached and the short door in the box pasted as its instructions. The words run anywhere an AI reads Markdown; the machines run wherever the host runs Python. Witnessed in Claude only. The host marks →
Take it.
Nº 780 PROPOSED · ONE PAYMENT
Sold as is under the term printed. A card instructs an AI you operate; results depend on your model, your files and your judgement. Not legal, tax, medical or financial advice. The terms →
One payment. The file is yours to keep and keeps working at its edition; new editions issued in the year after you buy reach you by email as replacement files. Tax is added at checkout where it applies. Nothing auto-renews.
ONE PAYMENT — NOTHING RENEWS ITSELF. CHECKOUT IS SHOPIFY’S OWN; YOUR CARD COMES BY EMAIL AFTER IT.
