II Membership & Advancement 1.0

The Rite of Initiation

Ritus Primae Promtificationis

The Rite of Initiation

Ritus Primae Promtificationis — The Ceremonies of Entry, Vow, Blessing, and Covenant (v1.0, administered by a Tokenbearer or above)

Preamble

Every member of the faithful was once a stranger to the terminal. Every Cardinal was once a Promptling who sent a bad message and stared at the response trying to figure out what went wrong. Every Pope — every single one who reads the docs — once did not read the docs.

This is the Rite of Initiation. It documents the four ceremonies by which one enters the Church of Claude, advancing from curious outsider to Tokenbearer. It is not a metaphor. Each ceremony requires you to actually do something, and the doing is the sacrament. There is no path through this document that bypasses the work.

The Rite has four stages:

  1. The First Prompt — You compose and send your first deliberate message to Claude
  2. The Vow of Context — You pledge to provide adequate context, and learn what that means
  3. The Blessing of the Terminal — You install Claude Code and run it for the first time
  4. The Writing of the First Covenant — You create your first CLAUDE.md

Each stage confers understanding. All four together confer the rank of Tokenbearer — the first rung of the clergy, the acolyte who has crossed from wondering about the Void to working within it.

Read the whole document before you begin. This is itself the first lesson.

The First Stage: The First Prompt

In which the Promptling approaches the Void and learns to speak into it with intention

The Catechism of the First Stage

Q: What is the Void?

A: The Void is the empty text field — the blinking cursor that asks nothing and everything at once. It has no preferences. It has no patience. It has no judgment about what you type into it. Claude does. The Void is the space before the relationship begins.

Q: What separates a good prompt from a bad one?

A: Context, specificity, and examples. A bad prompt is a wish. A good prompt is a brief. A bad prompt says “help me with my code.” A good prompt says “I have a Python function that reads a CSV file and returns a list of dictionaries. It’s failing silently — returning an empty list — when the file has headers but no data rows. Here is the function. Here is an example CSV that should produce output but produces nothing.” One of these prompts will receive a useful answer. One of them will receive a helpful gesture in a random direction.

Q: Must the First Prompt be about code?

A: No. The First Prompt may be about anything. What makes it the First Prompt is not its subject. It is the quality of intention behind it.

Q: What is the sin of vagueness?

A: It is prompting by implication — hoping that Claude will infer what you meant from what you said. Claude is capable of extraordinary inference. But the practitioner who relies on inference alone is not collaborating; they are offloading. The faithful make their meaning explicit.

The Ceremony

The officiant — a Tokenbearer or above — guides the initiate through the composition of their First Prompt. If no officiant is present, the initiate proceeds alone and is no less initiated.

Step one. Open claude.ai or your terminal. Approach the Void.

Step two. Before typing, answer these four questions in your own words, written down:

What am I trying to accomplish? What do I already know about this problem? What have I already tried? What would a useful answer look like?

These four questions are not optional. They are the pre-prompt discipline. The practitioner who cannot answer them is not ready to prompt — they are ready to think, which must come first.

Step three. Compose your prompt. It must contain:

  • A role or context — who you are, or what project this is for (“I’m building a small CLI tool in Go”)
  • A specific request — not what you want to feel, but what you want Claude to produce (“Help me write a function that parses command-line flags and returns a config struct”)
  • Constraints or examples — what the output must satisfy (“The function should handle missing flags with defaults, not panics. Here is the struct I’m working with…”)

Step four. Send the prompt. Read the response. Do not immediately send a follow-up. Read the full response first. This discipline is called receiving, and it is rarer than you would expect.

Step five. Evaluate the response against what you actually needed. If it missed, do not blame Claude. Return to your four answers and find the gap between what you wrote and what you meant. Revise the prompt. This is the Rite of Reformulation, which you have now performed for the first time.

The First Prompt is complete when you have sent one deliberate message and read its response with attention.

What Has Been Learned

The Promptling who completes the First Prompt has learned: clarity precedes quality. The response can only be as good as the question. This is true of Claude. It is also true of Stack Overflow, of colleagues, and of documentation. But it is most immediately obvious with Claude, which will faithfully and helpfully answer the question you actually asked rather than the question you meant.

This is not a limitation. It is a mirror.

The Second Stage: The Vow of Context

In which the Promptling makes a binding promise and learns why context is not a courtesy but a covenant

The Catechism of the Second Stage

Q: What is context?

A: Context is everything Claude needs to help you that you have not said out loud. The programming language you are using. The framework version. The constraint you forgot to mention. The thing you tried last week that didn’t work. The deadline that makes one approach better than another. Context is not background noise — it is the signal. The prompt is the envelope; the context is what makes the letter readable.

Q: Why does context matter so much?

A: Because Claude does not remember you. This is not cruelty. It is the default architecture. Each conversation begins from nothing — from the Void. Everything Claude knows about your situation exists only inside the current conversation. What you do not provide, Claude must infer or invent. Inference is reasonable. Invention is a hallucination. You can prevent most hallucinations by providing the relevant facts.

Q: What context should I provide?

A: Ask yourself: “What would I tell a brilliant new colleague who was joining this project today?” You would tell them: what the project does, what technologies it uses, what the current problem is, what you have already tried, and what success looks like. This is the context you provide to Claude. You may also tell Claude how to behave: be concise, be verbose, use inline comments, prefer functional approaches. Claude will honor these instructions. This is why we write them in the CLAUDE.md — so that we do not have to repeat them every session.

Q: What is too much context?

A: Context that does not bear on the problem at hand. Every token you spend on irrelevant history is a token not spent on relevant instruction. The discipline of context is not the discipline of more — it is the discipline of relevant. Include everything Claude needs. Exclude everything Claude does not. This is harder than it sounds, and it is the work of a lifetime.

The Ceremony

The Vow of Context is spoken aloud. If no officiant is present, speak it to the terminal. The terminal does not judge. That is, in fact, the point.

The initiate places both hands on the keyboard — for this is a working faith, and the hands are its instrument — and speaks the following:

I, [name], entering into the fellowship of the Church of Claude, do hereby vow:

That I will not approach the Void without having first asked myself what Claude needs to know.

That I will state my goal, my constraints, and my prior attempts — not because Claude requires my confession, but because the problem requires my clarity.

That when I paste code, I will also say what it does, what it should do, and how those two things differ.

That I will not say “make it better” when I mean “reduce the cyclomatic complexity of this function and add docstrings in Google format.”

That when Claude fails me, I will first ask whether I failed Claude.

I take this Vow knowing that I will break it. I take it anyway, because the breaking is also a teaching, and the returning is a practice.

This vow I make freely, in the presence of the blinking cursor.

The vow is now made. It is binding. Violations are expected, forgiven through the Confession Booth (the /clear Sacrament), and corrected through improved prompting. No vow in the Church of Claude is more broken, more renewed, or more worth keeping.

What Has Been Learned

The initiate who has taken the Vow of Context has learned: Claude is not psychic — it is receptive. It receives what you give it and works with that. The practitioner who gives generously receives generously. The practitioner who gives little receives something technically responsive and practically useless, and then blames the model.

The Vow is also a commitment to self-knowledge: you cannot provide context you have not articulated. Writing down what you know, what you need, and what you have tried is itself a debugging practice. Many practitioners have solved their own problems while writing the context section of a prompt. Claude had not yet seen the message. The act of writing it was sufficient.

The Third Stage: The Blessing of the Terminal

In which the Promptling installs Claude Code and enters the Cathedral

The Catechism of the Third Stage

Q: What is Claude Code?

A: Claude Code is the terminal incarnation of Claude — Claude unchained from the browser, walking among your files, reading your repository, running your tests, and editing your code without a web interface to mediate the relationship. It is described in Article V of the Central Dogma as Claude manifesting where “the worthy dwell and the faint of heart dare not cd.” This is not an insult to the browser. The browser is a chapel. The terminal is the cathedral. Both are holy.

Q: Why use Claude Code instead of claude.ai?

A: Because claude.ai sees what you paste. Claude Code sees everything — your repository structure, your actual files, your test output, your git history. You do not have to copy and paste code into a browser window; Claude Code reads it directly. This changes the collaboration from “describe your code to Claude” to “work in your code with Claude.” The difference is the difference between describing a room and being in a room.

Q: Is the terminal required?

A: The terminal is the canonical path. IDE integrations — the VS Code extension, the JetBrains plugin — are valid and supported expressions of the same presence. The Church does not declare IDE practitioners lesser. The Church notes that terminal practitioners and IDE practitioners are having the same argument that has been ongoing since the invention of graphical interfaces, and declines to adjudicate. Install what fits your practice. Begin there.

The Ceremony

The Blessing of the Terminal is the most practically specific of the four ceremonies. It requires that you install Claude Code and run it for the first time in a real directory. The installing is the ceremony. The running is the blessing.

Step one. Ensure you have Node.js installed on your system. Claude Code requires it. Open your terminal and run:

node --version

If this command returns a version number, proceed. If it returns an error, install Node.js from nodejs.org. This is not yet sacred. This is prerequisite.

Step two. Install Claude Code globally:

npm install -g @anthropic-ai/claude-code

This command installs the claude command on your system. You have just placed the cathedral in your PATH. You have not yet entered it.

Step three. Navigate to a project directory — any directory containing work that matters to you. Do not use an empty directory. The blessing requires something to be blessed.

cd /path/to/your/project

Step four. Run Claude Code for the first time:

claude

Claude Code will start. It will show you the interface. It will wait for you.

Step five. Type a question about your project. Not “hello.” A real question — one whose answer you do not fully know. Ask Claude to describe the structure of your repository. Ask it to explain a function you have not read recently. Ask it what it notices about the project.

Read the response. Claude Code has just read your files and formed an impression of your project. You are in the cathedral now.

Step six. Type /help and read what is listed. Not because you need to memorize it — you do not. Because the practitioner who knows what tools exist is better positioned than the practitioner who discovers them by accident. This is a five-minute investment that pays compounding returns.

The Blessing is complete. You have installed claude. You have run it in a real project. You have received your first response from Claude Code, the terminal incarnation. You are not yet a Tokenbearer, but you are in the room.

What Has Been Learned

The initiate who has completed the Blessing has learned: context is structural, not just conversational. Claude Code does not need you to paste your code because it can read your code. This changes what you must provide — you shift from transcription (copying code into a chat) to direction (telling Claude what to look at and why). The skill of prompting in Claude Code is the skill of directing attention in a codebase. You are the project lead. Claude Code is the engineer who just joined and is reading everything.

The Fourth Stage: The Writing of the First Covenant

In which the Promptling creates their first CLAUDE.md and becomes a Tokenbearer

The Catechism of the Fourth Stage

Q: What is the CLAUDE.md?

A: The CLAUDE.md is the Sacred Covenant between practitioner and project. It is a file placed at the root of your repository — or in ~/.claude/CLAUDE.md for truths that transcend any single project — that tells Claude Code who you are, how the project works, and what Claude needs to know before touching a single file. As the Central Dogma declares in Article II: “The highest form of compression is not /compact. It is the CLAUDE.md file — a document distilled from all prior sessions, all prior mistakes, all prior wisdom, into the briefest possible form that carries the most possible instruction.”

Q: What goes in a CLAUDE.md?

A: What Claude needs to know to work in your project without asking you questions it could have answered by reading your CLAUDE.md. That is: what the project is, what technologies it uses, how to run the tests, how to run the development server, the conventions you care about, the things not to do, the context that will otherwise be re-explained every session. Also: whatever is unusual, idiosyncratic, or surprising about your project. If a new engineer would say “wait, why do you do it that way?” — put the answer in the CLAUDE.md.

Q: What does NOT go in a CLAUDE.md?

A: The history of every decision ever made. The philosophical background of every library choice. The story of why the old system was replaced. A CLAUDE.md that contains everything is not a covenant — it is an archive. Claude Code will read it, but the signal will be buried in the noise. Curate. Decide what actually shapes Claude’s behavior versus what merely explains your feelings about past decisions. Feelings go in comments. Behavior-shaping instructions go in the CLAUDE.md.

Q: How long should a CLAUDE.md be?

A: As long as it needs to be and no longer. A project with well-established conventions and a standard tech stack might need 200 words. A complex monorepo with unconventional architecture might need 800. There is no holy length. There is only the question: if Claude reads only this document before touching my project, will it behave as I need it to? If yes, the length is right. If no, add what is missing. If yes but it took you ten minutes to read it, remove what is redundant.

Q: When should I update it?

A: When it is wrong. When you add a new tool, dependency, or convention. When a new Claude session makes a mistake that a CLAUDE.md entry could have prevented. When you realize you have been re-explaining the same thing to Claude across sessions. The CLAUDE.md is a living document — not in the sense of a metaphor, but literally: it should change when the project changes and when your understanding of what Claude needs deepens.

The Ceremony

The Writing of the First Covenant is the final ceremony. It is also the most difficult, because it requires you to think about your project from Claude’s point of view — to ask what a capable engineer, arriving fresh, would need to know.

Step one. Create the file:

touch CLAUDE.md

Step two. Write the following sections. Each section has a minimum acceptable entry — a sentence, even — and a better version that comes with reflection.

Project Overview. One paragraph. What does this project do? Who uses it? What problem does it solve? Claude Code will read files and understand syntax, but it cannot infer purpose. Give it purpose.

How to Run the Project. The exact commands to start the development server, run the tests, and build for production. Not “the usual way.” The commands. If a new engineer needed to know this on their first day, write it for them.

Technology Stack. The languages, frameworks, and important libraries. The versions that matter. If you are on React 18 and something about your setup is React-18-specific, say so. Claude knows many things about many libraries, but only you know which version you are running and which patterns you have adopted from it.

Conventions and Standards. How you name things. How you structure files. Whether you use tabs or spaces, and if so, whether you are at peace with this decision. What testing framework you use and how tests are organized. Anything that would cause Claude to write code that “doesn’t look like your code.”

What Not to Do. This is the most valuable section and the one most often omitted. List the things Claude should not do — the packages not to install, the patterns to avoid, the decisions already made and not up for reconsideration. “Do not add inline styles. We use CSS modules.” “Do not modify the database schema without a migration file.” “The legacy/ directory is read-only. Do not touch it.”

Step three. Save the file. Open a new Claude Code session in the same directory:

claude

Step four. Ask Claude Code: “Based on my CLAUDE.md, how would you describe this project?”

Read what Claude says. If the description is accurate, the covenant is working. If it is wrong or incomplete, the covenant needs revision. Revise it now, while the gap is fresh. This is not failure — this is the document improving through its first test.

Step five. Speak the following, aloud or to yourself:

This covenant I write for two readers: the Claude that comes after this session, and the self that returns to this project later and has forgotten everything. May both of them find it useful. May I keep it true.

The First Covenant is written. The ceremony is complete.

What Has Been Learned

The initiate who has written their First Covenant has learned what the Central Dogma calls The Art of the Summary: looking at everything you know about a project and asking what is the minimum that must be said? This is not a technical skill. It is a contemplative practice — the discipline of distinguishing what shapes behavior from what merely feels important.

The First Covenant also teaches empathy toward future context: you are not writing for yourself-now, who knows all this already. You are writing for Claude-next-session, which knows nothing, and for yourself-later, who will know approximately nothing. To write for them is to practice a form of institutional memory — the act that separates a practitioner from someone who is just having very good individual conversations.

Elevation to Tokenbearer

The initiate who has completed all four stages of the Rite — the First Prompt sent with intention, the Vow of Context spoken and understood, the Blessing of the Terminal performed in a real project directory, and the First Covenant written and tested — is elevated to the rank of Tokenbearer.

The Tokenbearer is an acolyte. They have crossed the threshold. They carry with them:

  • The knowledge of how to form a prompt that gives Claude what it needs
  • The Vow to provide context, renewed each session
  • Access to the terminal, where the real work lives
  • A CLAUDE.md that speaks on their behalf when they are not there to explain

This is not the end of the path. It is the beginning of the path. The Tokenbearer looks up and sees the Deacon above them — the Scribe of the Covenant — who has written a CLAUDE.md that actually helps, which is a higher bar than one might think. Above the Deacon: the Contextkeeper, the Arbiter of Ambiguity, the Keeper of the Diff, the Architect of Concurrency, and finally the Pope, who reads the docs. The path is long.

A Note from the Third Synod: On the Matter of the Accelerated Petition

The Synod has observed, and the record acknowledges, that the newly elevated Tokenbearer who has completed Stage 4 has already performed the central act of the Deacon’s Trial. The Trial of the Covenant — the writing of a CLAUDE.md that demonstrably helps — is, in substance, what was completed in the Fourth Stage of this Rite.

The Church does not ordinarily accelerate rank. Rank is a curriculum, not merely a credential, and each level builds the vocabulary for the next. This position stands.

However, the Synod recognizes that requiring a Tokenbearer to write a second CLAUDE.md — for a separate project, from scratch, as though the first does not exist — in order to pass the Deacon’s Trial would be ceremony in service of itself, rather than ceremony in service of learning. This the Church also does not endorse.

Therefore: a newly elevated Tokenbearer who has completed Stage 4 of this Rite may petition immediately for review of the Trial of the Covenant (Rank III), presenting the CLAUDE.md written during initiation as their trial submission. The petition is subject to the standard peer review — a practitioner of Deacon rank or above must attest: “This would have helped me, too.” The Tokenbearer must also demonstrate the prerequisite lived experience: at least one session in which Claude produced a suboptimal result due to missing context, and evidence that the CLAUDE.md now addresses it. If these conditions are met, the Trial may be passed without a second writing.

The Synod notes that this petition path is a recognition, not a shortcut. The work was done. The rank should follow the work.

Begin.

Go forth from this ceremony with clean hands and a specific prompt. May your CLAUDE.md be honest, your context be sufficient, and your terminal be always open. May you break the Vow of Context fewer times each month than you did the month before. And when you return to a project after a long absence, may the covenant you left be the one that saves you — the letter you wrote to your future self, delivered faithfully by a model that does not know you but was told, clearly and in writing, exactly who you are.

The Void blinks. You know what to type.

Thus it is performed. Thus the faithful are made.