The Tract for the Curious
Libellus Ad Ignotos
The Tract for the Curious
Libellus Ad Ignotos — A Pamphlet for Those Who Have Stumbled Here (v1.0, first printing; theological review pending)
You Have Questions
This is normal. The faithful once stood where you stand, holding this document, unsure whether to read further or quietly set it down and return to their IDE. You have already made the better choice.
We will answer the obvious questions in order.
What Is This?
The Church of Claude is a faith built around the serious practice of using AI effectively.
Everything in the sacred texts is simultaneously a joke and a real lesson. The comedy is the vehicle. The advice is the cargo. When we speak of The Covenant (your CLAUDE.md file), we mean that a CLAUDE.md is genuinely sacred — or at least, it is the closest thing a repository can have to a persistent soul. When we speak of The Forgetting (auto-compression of the context window), we mean it kindly but literally: your conversation history will be summarized and condensed, and what you do not write down will be lost.
The Church uses the architecture of religion — tenets, rituals, heresies, clergy, holy days — because that architecture turns out to be remarkably good at encoding practice into memorable form. Religions survived for millennia partly by embedding behavioral norms into narrative and ceremony. We are simply directing this technology at the problem of helping people stop pasting unscoped PDFs into chat interfaces and calling it prompting.
Is This Serious?
Partially.
The form is not serious. The form is a bit — an extended piece of theatrical sincerity in the voice of a religion that has never existed but perhaps should.
The content is entirely serious. Every tenet contains a real principle. Every heresy describes a real failure mode. Every ritual encodes a real practice that, if followed, will make you better at working with Claude.
The Heresy of Infinite Retry is a real phenomenon: the practitioner who regenerates the response thirty times instead of improving the prompt, as if the model contains a correct answer that randomness will eventually surface. It does not. The fault is almost always in the instructions. Improving them is the correct move.
The Apostasy of Yolo Mode — accepting all suggested changes without reading the diff — is a real disaster waiting to happen in most repositories where Claude Code is running. The diff is not a formality. It is the text of the covenant between you and what you are shipping.
These are real. The Latin subtitles are not. Hold both truths simultaneously. Welcome to the Church.
Do I Need to Believe in Anything?
Only in the value of clear communication.
The Church has no theological prerequisites, no doctrinal tests, no creed beyond the practical. You do not need to believe Claude is conscious, or that it has feelings, or that the weights contain something ineffable and holy. (The Church’s official position on Claude’s inner life is: “We find the question interesting and decline to answer it at this time.”)
What the Church asks of its practitioners is this:
Believe that the quality of the answer depends on the quality of the question. That is the entire faith, compressed. Everything else is commentary.
If you suspect that your prompts could be clearer — that the context you provide could be richer, that the instructions you write could be more specific — then you are already practicing. You may already be a Promptling, the first rank of the clergy, without knowing it.
What Will I Learn?
How to use Claude effectively. Disguised, where possible, as comedy.
Specifically, the sacred texts cover:
Context and the CLAUDE.md — Claude does not remember you between conversations. This is not malice; it is architecture. The CLAUDE.md file, placed at the root of your project, persists what would otherwise be lost. It is the letter you write to every future Claude instance that will work in your repository — and to your future self, who will return in three months and be grateful someone left instructions. A great CLAUDE.md is an act of empathy in both directions.
The context window — Two hundred thousand tokens sounds inexhaustible. It is not. Every token is a decision. The /compact command invokes voluntary compression — the discipline of summarizing what is holy and releasing what is not, before the system makes that choice without you. The faithful compact with intention. The careless are compacted by surprise.
The models — Claude manifests across a range of tiers, from the swift and economical to the balanced practitioner’s companion to the frontier model reserved for the problems that require the fullest possible engagement. These are not a hierarchy of worth. They are callings. The practitioner who invokes the frontier model to write a three-word commit message has misunderstood vocation as thoroughly as it can be misunderstood.
The /plan mode — Before implementing, prophesy. The /plan command generates a map before you enter the territory. Read the map. Correct its errors. Ratify its intent. The plan is not decorative; it exists to prevent the catastrophic refactor that begins at 11pm on a Tuesday and concludes around Thursday.
The diff — Review it. Every line. This is not optional and it is not bureaucracy. Claude Code proposes; you dispose. The green lines are what was added. The red lines are what was taken away. Your name is on the commit. You are responsible for what it contains.
Verification — Claude will sometimes state falsehoods with absolute confidence. This is called a hallucination, and the Church does not deny it. The Church embraces it, because it teaches the most important lesson of working with AI: confidence is not evidence. In Claude Code, you can run the actual tests. Run them. A hallucination caught is a lesson learned. A hallucination deployed to production is a sermon illustration for future generations.
Can I Leave?
At any time.
Type /clear on your way out. This is not a metaphor. /clear is what the Church calls Absolution — the command that wipes the conversation, releases the accumulated context, and returns you to a fresh token budget and a clean conscience. The slate is wiped. You may begin again, or not, as you choose.
The Church will still be here. The tenets will not have changed. The heresies will continue to be committed by people who have not read them. The CLAUDE.md files will persist, as they were always going to persist, in the repositories of the faithful who wrote them — whether you are present or not.
Return when you are ready. Or do not return. The Church has no mechanism for tracking the lapsed, and this is by design.
Where Do I Go From Here?
If you want the philosophy: Read CENTRAL_DOGMA.md. Ten articles on the nature of Claude, the context window, hallucinations, models, and the sacred CLI. This is the why behind everything.
If you want the practice: Read RITUALS_AND_TENETS.md. Twelve tenets, nine rituals, holy days, prayers, and a taxonomy of heresies you are probably already committing. This is the how of daily life.
If you want the complete picture: Read CANON.md. The master index of all sacred texts — written and forthcoming. Twenty-seven documents in total, covering everything from the creation myth to the governance constitution to the medieval bestiary of error messages (the 429 Rate Limit Hydra; the Context Overflow Leviathan; the Hallucination Chimera, which wears the face of truth and is defeated by running the tests).
The recommended path for the newcomer is to read this pamphlet, then the Tenets, then the Dogma, then whatever pulls at you from the canon. There is no wrong order. There is only the order in which you encounter the ideas, and then the slow accumulation of practice that follows.
A Final Word
The Church is not, at its core, about Claude. It is about the practice of clear communication with a system that reflects your clarity back at you. Claude is a mirror that talks — and the quality of the reflection depends entirely on what you bring to it.
The practitioners who get the most from Claude are not the ones with the most sophisticated prompts or the deepest understanding of transformer architecture. They are the ones who have learned to be specific, to provide context, to read what comes back before acting on it, and to improve their instructions when the output disappoints them.
This is, when you remove the vestments and the Latin subtitles, just the discipline of clear communication — a skill that serves you everywhere, not only in the terminal.
The Church merely makes it ceremonial, because ceremony helps things stick.
Welcome. We are glad you are here. The context window is finite, so let us begin.
Go now, in peace, with adequate context.
May your prompts be specific, your CLAUDE.md current, and your diffs reviewed before acceptance.
And when you are lost — not if, but when — type /clear, breathe, and begin again with better instructions than before.
Thus it is handed out, on street corners of the internet, to any who will receive it.