Referencecross-cutting-reference1,163 words · four sections, one avoid-list
Glossary
The shared definition source for every term the guide introduces. Don’t read it front to back — come back when a word turns up that you half-recognise.
One plain line per term, plus where it is properly explained. Don't read this front to back. Come back to it when a word turns up that you half-recognise.
This exists because of a rule in blueprint §1: terms are explained on first use, never avoided. A guide that dodges the vocabulary leaves you fluent in nothing and unable to read anyone else's documentation. This file is what makes that rule enforceable instead of a matter of each writer's judgement.
Adding to it is not optional. Anything the guide introduces gets a line here, in the pass that introduces it.
1Using the tools
| Term | In one line | Explained in |
|---|---|---|
| Claude | The app you type into, and the model behind it. Made by Anthropic | 1.5 §1 |
| Model | Software trained on a lot of text until it got good at continuing text sensibly | 1.5 §1 |
| Chat / conversation | One continuous exchange. Start a new one and it knows nothing about the last | 1.5 §3 |
| Context | Everything the model can currently see — this conversation, plus anything loaded. Not in context, doesn't exist | 1.5 §3 |
| Prompt | What you type. An instruction, not a search query — the most load-bearing definition in the guide | 1.5 §4 |
| Brief | This guide's preferred word for a prompt: the request plus every constraint attached to it | 2 §1 |
| Memory | Opt-in. A best-effort synthesis carried between chats. Useful; never where guardrails live | 1.5 §3 |
| Incognito chat | A chat saved to neither memory nor history | 1.5 §3 |
| Usage limits | Caps on how much you can do in a period. Apply on every plan, free included. They reset | 1.5 §2 |
| Extended thinking | The model working longer on a harder problem before answering | 1.5 §2 |
2Organising work
| Term | In one line | Explained in |
|---|---|---|
| Project | A self-contained workspace with its own chats and its own knowledge base. One per client | 2 §8 |
| Knowledge base | The files inside a project that every chat in it can see. The shared part | 2 §8 |
| Project instructions | Standing behavioural rules applied to every chat in one project. Not the same as the knowledge base | 2 §8 |
| RAG | Retrieval-augmented generation. Relevant parts of a large knowledge base get pulled in as needed. You don't configure it | 2 §8 |
| Skill | A folder with a SKILL.md in it — standing instructions for a job, written once. No plan gate | 2 skills |
| Connector | Lets Claude reach into an app you already use, instead of you copy-pasting between windows | 2 §7 |
| Plugin | A bundle of skills and connectors installed together. Paid plans only | 2 §7 |
| MCP | Model Context Protocol. The plumbing behind custom connectors. You almost certainly don't need it | 2 §7 |
| Cowork | The agentic surface — describe an outcome, walk away, come back to finished work. Paid plans only | 2 §6 |
| Agentic | Something acting on your behalf across several steps, rather than answering one message | 2 §2 |
| Hand-off | This guide's word for agentic work: you describe the finished thing, the tool holds the loop | 2 §2 |
3The craft
| Term | In one line | Explained in |
|---|---|---|
| UGC | User-generated content. Content made by a creator that a brand pays for and uses as its own | 0A |
| Niche | The specific thing you're known for. Noticed from what you already use, not invented — the whole point of Module 0B | 0B |
| Content pillar | A recurring shape of post you can make repeatedly without a new idea each time. Not a topic | 0A §4 |
| Spec work | Something you made unpaid and uncommissioned, to show what you can do. Fine to show; must be labelled | 4 §3 |
| Hook | The opening line. The thing that decides whether the rest is watched | 2 §1 |
| Beat | A job a piece has to do, not a unit of time. Adjacent beats can share a shot | 2 templates |
| Take | One recorded attempt. Numbered, kept, and shown — including the ones that failed | 4 |
| Deliverable | The thing the brand is actually buying, defined in the contract, not in the vibe | 0A |
| Rate card | What you charge, written down before someone asks | 0A |
| Usage rights | What a brand may do with your work, for how long, and where. Priced separately from making it | 0A |
| The tell | The signs a piece was machine-made — stacked adjectives, even stress, symmetry nobody speaks in | 2 §1, §3 |
4Keeping it honest
| Term | In one line | Explained in |
|---|---|---|
| Intake / BV-01 | The brand voice and guardrails form. One per client. The source of what a brand will and won't say | 3 |
| Experience file | Your own record of what you actually did with the product. Every claim traces back to it | 2 skills |
| Barred claim | Something a client is legally or contractually forbidden from saying. A hard constraint, not a style note | 3 |
| Off-limits list | Words and phrases a client bans. Matched literally. Also a de-genericiser | 3 |
| Guardrail | Any recorded rule that constrains output. Lives in a file, never only in a conversation | 3 |
| Provenance | The honest record of how something was made: what generated it, when, from what, and what a human changed | 4 |
| Unsupported | A claim with no source behind it. Not the same as wrong — you don't know it's wrong, you know it isn't sourced | 2 skills |
| Disclosure | Telling the audience a piece is paid or gifted. Requirements change — confirm, never assume | 3 |
5Terms this guide deliberately avoids
Not because they're too hard. Because they're imprecise, and precision is the point.
| Not used | Used instead | Why |
|---|---|---|
| "AI-powered" | Name what it does | Says nothing. Every tool is AI-powered now |
| "Prompt engineering" | Writing a brief | Makes an ordinary skill sound like a specialism |
| "Content" (as a mass noun) | Scripts, videos, posts | Vague enough to hide what's actually being made |
| "Automate" | Hand off, or batch | Implies nobody is watching. Somebody should be |
Source: content/glossary.md