THE METHODS · CLAUDE CODE · Nº 485 · EDITION 2 · LAUNCHING SOON

LUND CARDS · THE METHODSNº 485

THE CARD AS TEXT · FOR EYES AND FOR MACHINES

THE CLAUDE.MD CARD · The memory-file seat: the file an agent reads first, written from the repository and kept true.

WAKES WHEN A repository has no memory file and the agent keeps asking how to build and test it. · A memory file has grown long, stale or contradictory and the agent 'keeps forgetting the rules'. · A team wants one file that says what to run, what must pass and what never to do. · Someone wants to check a memory file before committing it.

THE LOOP

Scan the repository for what it declares
Draft from what was found; UNKNOWN elsewhere
Lint: secrets and skipped checks refused
Trim; keep every rule once

NEVER

Never puts a secret in a memory file
Never writes a rule to skip a check
Never lists a command nobody has run
Never claims a certainty it does not have

IT MAKES scan — read the repository: build, test and lint commands it declares, the top-level map, what memory file exists; draft — write a memory file from what scan found; anything it could not find prints UNKNOWN for the person to fill; lint — check a memory file: SECRET IN FILE and SKIPS CHECKS are refused (exit 1); STALE, TOO LONG and DUPLICATE are flagged; commands — list every command the file tells the agent to run and say which scripts it names are NOT IN REPO; nothing is executed; trim — drop exact duplicate lines and runs of blank lines, keep every rule, and count what it left out.

THE LAW A memory file is read whole at the start of every session, so every line costs every session. It says only what the repository can prove: the commands that exist, the checks that must pass, the things never to do. A secret in it is a leak; a rule to skip a check is a lie waiting to be told.

IN THE BOX SKILL.md · CARD.md · ▣ scripts/claudemd.py (machine) · ▣ scripts/evals.py (test runner) · tests/EVALS.json (9 tests) · references/ (schema + fictional demo) · START-HERE.md · system/ doors · MY-LEDGER.md · LICENSE-NOTE.md · card.svg · card.json · card.html · AUDIT.md · MANIFEST.md · WITNESS.md · SELFTEST.md.

PROOF SELFTEST 12/12 · witnessed in Claude 2026-10-14 on fictional fixtures · the test prompts in fresh chat, 9/9 · Edition 2.

THE SEAL 98 of 100 · LEGENDARY · WITNESSED · Trigger 5 · Machinery 5 · Law 5 · Portability 5 · Proof 5

IT CAN BE WRONG an AI reading a text file: it can be wrong and can invent a number, a rule, a name or a date; check what matters against your own records and the source; the risk is yours. LICENSE-NOTE.md rides in the packet and is the licence.

THE PUNCH · SHA-256 OF SKILL.MD · E60E7276C07BE4F278C9D8A2

THE FACE, FULL SIZE → · THE MACHINE FACE → · card.json →

THE CLAUDE.MD CARD

The memory-file seat: the file an agent reads first, written from the repository and kept true.

SCAN · DRAFT · LINT · COMMANDS · TRIM

LEGENDARYWITNESSEDIN CLAUDE IN LUNDR IN CHATGPT

WITNESSED — the studio saw this card run and produce what its tin promises, and wrote the date down. What the marks mean →

THE SPEC SHEET · READ BEFORE YOU BUY · THE STANDARD EVERY CARD MEETS →

SEAL: 98 of 100 · LEGENDARY · WITNESSED
EDITION: E1 · witnessed 2026-10-14
MACHINE: scripts/claudemd.py · 5 commands · selftest 12/12
RUNS: On your own computer. Standard library only, offline; no account, no key, no telemetry.
HOSTS: Claude · Claude Code · ChatGPT · Gemini · any AI that reads a file (START-HERE.md and system/ name each door)
LICENCE: One person, any number of their own AIs and computers · editions replace files · nothing renews · freeze, not loss
PUNCH: E60E7276C07BE4F278C9D8A2 · sha-256 of SKILL.md · check any file against it →

CITE THIS CARDTHE CLAUDE.MD CARD, Edition 2. Lund Studio LLC, Westminster, Colorado. https://lundr.tech/?view=claude-md-card. Punch E60E7276C07BE4F278C9D8A2. Retrieved 2026-10-14.

What it is.

Use when writing or fixing a CLAUDE.md or any project memory file an agent reads first: commands, checks, never-do, map. Not for enforcing a rule automatically or packaging a workflow.

It carries a machine that shows its working: scan — read the repository: build, test and lint commands it declares, the top-level map, what memory file exists; draft — write a memory file from what scan found; anything it could not find prints UNKNOWN for the person to fill; lint — check a memory file: SECRET IN FILE and SKIPS CHECKS are refused (exit 1); STALE, TOO LONG and DUPLICATE are flagged; commands — list every command the file tells the agent to run and say which scripts it names are NOT IN REPO; nothing is executed; trim — drop exact duplicate lines and runs of blank lines, keep every rule, and count what it left out — 12 checks in its selftest. It never puts a secret in a memory file; never writes a rule to skip a check; never lists a command nobody has run; never claims a certainty it does not have.

The law, written into the card.

It is in the card itself, so the AI follows it.

THE LAWA memory file is read whole at the start of every session, so every line costs every session. It says only what the repository can prove: the commands that exist, the checks that must pass, the things never to do. A secret in it is a leak; a rule to skip a check is a lie waiting to be told.

The method.

Read the repository before writing a line: the scripts it declares, the targets its Makefile has, the test and lint tools it carries.
Write the file from that reading. Where the repository declares nothing for build, test or lint, write UNKNOWN and leave it for the person; never invent a command.
Lint before committing: a credential refuses the file outright, so does a rule to skip tests; a command that names a script the repository does not have is STALE; past the line limit is TOO LONG.
Trim by removing exact duplicates and blank runs only. A rule is never dropped by the machine; a person decides what a rule is worth.
Keep a corrections log at the foot: when the agent got something wrong because of the file, the fix is a line in the file, dated.

What you bring.

The repository, or at least its build files.
The rules the team already follows by habit.
The commands people actually run, in the words they run them.

Planned as a separate seat, not in this box and not yet on sale: a memory file the studio keeps for its own repositories, refreshed every build.

What it does once your AI has it.

scan Read the repository: build, test and lint commands it declares, the top-level map, what memory file exists
draft Write a memory file from what scan found; anything it could not find prints UNKNOWN for the person to fill
lint Check a memory file: SECRET IN FILE and SKIPS CHECKS are refused (exit 1); STALE, TOO LONG and DUPLICATE are flagged
commands List every command the file tells the agent to run and say which scripts it names are NOT IN REPO; nothing is executed
trim Drop exact duplicate lines and runs of blank lines, keep every rule, and count what it left out

What it will never do.

It never puts a secret in a memory file; never writes a rule to skip a check; never lists a command nobody has run; never claims a certainty it does not have.

Where it hands off: A rule that must run automatically → THE HOOK CARD; A workflow to package as a command → THE COMMAND CARD; Which files an agent should read for a task → THE CONTEXT CARD.

It is not a security review. It is an AI reading a text file: it can be wrong, so check what matters against your own records and the source. The holder brings their own AI; no endorsement by Anthropic or any other AI company is claimed.

What’s in the buy.

SKILL.md — the card your AI reads: when it wakes, the loop, the never-lines, the hand-offs.
CARD.md — the whole card in ten sections: face, law, method, machine, what you bring, never, tests, seal, term.
scripts/claudemd.py — the machine, Python standard library only; `selftest` proves it on its own fixtures.
tests/EVALS.json and scripts/evals.py — 9 test prompts and the offline runner that scores your AI's answers.
references/ — the data schema and FICTIONAL demo files, invented to show the machine, not real.
MY-LEDGER.md — your own record of what you asked and what you decided.
START-HERE.md, system/short-prompt.md, system/rules.md — the doors, for any AI with or without skills.
card.svg, card.json, card.html, AUDIT.md, MANIFEST.md, WITNESS.md, SELFTEST.md — the faces and the proof.
NOT IN THE BOX — any studio edition (planned, not yet on sale: a memory file the studio keeps for its own repositories, refreshed every build). Also not in the box: any data you must bring, and advice of any kind (not a security review).

PROOF · SELFTEST 12/12 · WITNESSED IN CLAUDE 2026-10-14 ON FICTIONAL FIXTURES · THE 9 TEST PROMPTS ASKED IN A FRESH CHAT HOLDING ONLY THE CARD · 9/9 · EDITION 2 · PUNCH E60E7276C07BE4F278C9D8A2

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 485 — LAUNCHING SOON

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 medical, legal, tax or financial advice, never the person’s boss, and not a licence to practise. The terms →

ONE PAYMENT · NOTHING RENEWS ITSELF

OFFLINE · ONE LICENCE · PUNCH-CHECKABLE · THE STANDARD

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.

A TERM, NOT A SUBSCRIPTION: NOTHING RENEWS ITSELF AND NOBODY IS CHARGED AGAIN. WHEN A TERM ENDS THE FILES STAY ON YOUR COMPUTER AND STAY READABLE — FREEZE, NOT LOSS.

THE MACHINE · IN THE OPEN · FROM THE CARD’S OWN SELFTEST.MD

claudemd.py, every command showing its working.

A small Python program, standard library only. Your AI runs it; you can too. The block below is pasted from the card’s own SELFTEST.md, as printed when it was run from inside the packet.

`scan`	read the repository: build, test and lint commands it declares, the top-level map, what memory file exists
`draft`	write a memory file from what scan found; anything it could not find prints UNKNOWN for the person to fill
`lint`	check a memory file: SECRET IN FILE and SKIPS CHECKS are refused (exit 1); STALE, TOO LONG and DUPLICATE are flagged
`commands`	list every command the file tells the agent to run and say which scripts it names are NOT IN REPO; nothing is executed
`trim`	drop exact duplicate lines and runs of blank lines, keep every rule, and count what it left out
`selftest`	12 checks that prove the machine; it must print selftest 12/12

THE SELFTEST, AS PRINTED

trim <tmp> → <tmp> · 34 lines · 4 left out (duplicates and blank runs only; no rule removed)
SECRET IN FILE: line 36 — refused; remove it and rotate the credential
lint <tmp> · 36 lines · REFUSED (1) · 0 flags
ok   1  scan finds the build, test and lint commands the repository declares
ok   2  a repository that declares nothing prints UNKNOWN for build, test and lint instead of a guess
ok   3  the draft lists each found command with the file it was found in
ok   4  the draft for the bare repository carries UNKNOWN lines, not invented commands
ok   5  the draft lints clean against its own repository
ok   6  a key in the file is SECRET IN FILE: refused, exit 1
ok   7  a rule to skip the checks is SKIPS CHECKS: refused
ok   8  a command naming a script the repository does not have is STALE, NOT IN REPO
ok   9  a file past the line limit is TOO LONG, flagged not refused
ok   10  trim drops exact duplicates and blank runs and keeps every rule once
ok   11  commands lists what the file tells the agent to run, and names the script NOT IN REPO without executing anything
ok   12  lint exits 1 on a refusal
selftest 12/12

THE RUN IS PASTED FROM THE CARD’S OWN SELFTEST.MD AS PRINTED. ITS FIXTURES ARE FICTIONAL, INVENTED FOR THE TEST; NO PRIVATE PERSON IS NAMED.

QUESTIONS AT THE TABLE

Asked before the wire.

What does it need from me?

The repository, or at least its build files. The rules the team already follows by habit. The commands people actually run, in the words they run them. If something it needs is missing, it says what is missing and stops there.

Does it send anything anywhere?

No. The machine is a standard-library Python program that runs on your own computer and sends nothing anywhere; the words run wherever an AI reads Markdown. No server, no account, no key.

What is on the record?

The machine's selftest (12/12), run from inside the packet in Claude on 2026-10-14 on fictional fixtures and pasted into SELFTEST.md as printed; the twelve-check audit (AUDIT.md). Owed, and said so in WITNESS.md: Any run on real data. Every demo file is fictional. The punch on the face is the SHA-256 of SKILL.md, so a downloaded copy can be checked against this room.

Can it be wrong?

Yes — it is an AI reading a text file: it can be wrong and can invent a number, a rule, a name or a date; check what matters against your own records and the source; the risk is yours. LICENSE-NOTE.md rides in the packet and is the licence. It is not a security review.

What will it never do?

It never puts a secret in a memory file; never writes a rule to skip a check; never lists a command nobody has run; never claims a certainty it does not have.

← THE METHODS · THE FLOOR →