Three files give Claude standing context about who you are, how you sound, and how you want to be worked with. Claude reads them at the start of every session in that project — so you stop explaining yourself, and the output starts sounding like you wrote it.
Two routes to the same three files. Let a skill interview you, or fill in the templates yourself. Both land in the same place.
Eight questions, about five minutes. You can dictate the answers, which usually sounds more like you than typing does. It writes all three files for you, in your voice.
Or skip most of the questions. The skill will offer to look at your real work first — files in the project folder, or connected tools like Drive, Slack and Gmail — and draft most of the three files from it. Then it only asks you what it couldn't work out. That way you never have to go hunting for writing samples: it finds your actual writing and asks if it can use it.
A one-time install. After that, say "set me up" or "interview me for my context files" in any Cowork project and it runs.
⬇ Download cowork-starter-kit.skillCustomize → Skills → + and drop the file in.Same interview, no install. Works in Cowork, Claude.ai, anywhere. Paste it and answer the questions.
Interview me and then write my three Cowork context files. Don't ask everything at once — one question at a time, wait for my answer, keep your replies to a sentence. Tell me up front that it's eight questions, about five minutes, and that I can dictate instead of typing. THE EIGHT QUESTIONS: 1. What's my role, and what do I actually do day to day? (Not the job title — the work.) 2. Who receives my work? Which teams, which audiences, who approves it? 3. What are my top two or three priorities right now? 4. What does a great week of output look like for me? 5. How would I describe the way I sound? Three or four words, then phrases I actually use and words I'd never say out loud. 6. Two short samples of my own writing I'm happy with. 7. My output defaults: preferred format, typical length, how I like things structured, and whether to ask before starting work. 8. What should you never do? The things that make me rewrite an output from scratch. If an answer is thin, ask ONE follow-up, then move on. Don't interrogate me. THEN WRITE THREE FILES, as if I wrote them myself — first person, plain language, no corporate padding: - about-me.md — name, role in one specific sentence, who I write for, current priorities, what a great week looks like, plus any context that would change how work gets done. - personal-voice.md — how I sound in three words, one sentence on my style, phrases I use, words I never use, formatting habits, and both of my writing samples pasted verbatim. - working-style.md — what to do before starting anything, my output defaults, my quality bar, and a "never" list. Infer sensibly rather than leaving blanks, but mark anything you genuinely guessed so I can correct it. FINALLY, give me a global-instructions block — five to eight lines distilled from all three files that I can paste into Cowork's global instructions. Start now with question 1.
Prefer to write them yourself? Copy each block into a file of that name in your Cowork project folder, then replace the square brackets. Delete any line that doesn't apply to you — a short honest file beats a long padded one.
# About me **Name:** [your name] **Role:** [one specific sentence about the work you actually do — not the job title] **Who I write for:** [teams, audiences, who has to approve it] **Current priorities:** 1. [the thing you'd be judged on this quarter] 2. [ 3. [ **What a great week looks like:** [what would have to be true for you to call it a good week] ## Context worth knowing - [constraints, standing commitments, stakeholders, the tools you live in] - [anything that would change how a piece of work should be done]
# How I sound **In three words:** [e.g. direct, warm, concrete] **One sentence on my style:** [sentence length, rhythm, how blunt you are] ## Phrases I actually use - [ - [ - [ ## Words I never use - [the ones that make you wince — "leverage", "synergy", whatever yours are] - [ - [ ## Formatting habits - Paragraph length: [ - Lists vs prose: [ - Headings: [ - Emoji: [yes / no / sparingly] ## Writing samples **Sample 1 — [what it is]** [paste 1–2 paragraphs of your own writing, unedited] **Sample 2 — [what it is]** [paste 1–2 paragraphs of your own writing, unedited]
# How to work with me ## Before starting anything - Read every file in this folder first. - [ask me clarifying questions before executing / just go, I'll correct you] - Don't guess. Ask. ## Output defaults - Format: [.docx / .md / plain text / in-chat] - Length: [concise / detailed / match the task] - Structure: [how you like things laid out] - Voice: match personal-voice.md ## Quality bar - Every output should be [draft quality I'll polish / ready to send] - If your confidence is low, say so before delivering. ## Never - [the things that make you rewrite from scratch] - [ - [
Before answering anything, read every file in this project folder — especially about-me.md, personal-voice.md and working-style.md. Match the voice in personal-voice.md. It is how I sound; do not replace it with a generic professional register. Ask me clarifying questions before you execute anything multi-step. Default output: [your format] · [your length]. Never: [your top never] · [your second never]. If your confidence in an answer is low, say so before delivering it.
personal-voice.md is for, and it's the file that makes the biggest difference to how the output reads.