What this tutorial solves
Claude’s 200k-token context is the headline feature, but in practice teams burn it by dumping every file into a Project and hoping retrieval sorts it out. The result: vague answers, drifting tone, and chats that confidently quote a draft you deleted three weeks ago. This walkthrough swaps the “folder” model for a curated workspace pattern that holds up over months.
Who this is for
Claude Pro, Team, or Enterprise users running multi-week work: research, long-form writing, codebase analysis, recurring client engagements, or repeat product reviews. If you open a new chat once a week and re-explain the same five facts, this is for you.
When to reach for it
When a task needs persistent context, named reference files, and a stable role across many chats. The clearest signal: you can name three files that should be loaded into every chat about this topic.
When this is NOT the right tool
One-off questions, exploratory prompts where you do not yet know the shape of the output, or short tasks where the cost of setting up a Project exceeds the time savings. For ad-hoc work, a regular chat with files attached is faster.
Before you start
- Define the outcome by deliverable, not topic: “Q3 launch plan with email, deck, and FAQ,” not “marketing.”
- Gather 3-7 files that ground every chat. Include a brief, conventions, and one strong reference example.
- Decide explicitly what stays out: old drafts, half-finished notes, exploratory tangents. Excluded files belong in a separate “scratch” Project.
- Pick one chat as a smoke test before building the full system.
Step by step
- Name the Project after the outcome (“Q3 product launch plan”), not the topic (“marketing”). Outcome names force a done state.
- Write Project Instructions in under 1000 words: role, audience, voice, what to skip, two or three success criteria. Read it back — anything generic about “be helpful” should be cut.
- Upload only the 3-7 files that ground every chat. Outdated drafts and tangential references stay out. Name files explicitly (
spec.md,voice.md,decisions.md) so Claude can reference them by name in replies. - For each new chat, lead with the sub-goal: “Today: rewrite the launch email using
voice.md. Output a single 180-word version plus three variant subject lines.” - When the chat reaches a real conclusion, copy decisions back into
decisions.mdin the Project and re-upload. This is the single highest-leverage habit. - Every two weeks, prune: replace stale files, delete chats that did not produce anything useful, refine Instructions based on what actually worked.
First-run exercise
- Pick a low-stakes deliverable — one email, one chapter section, one bug ticket.
- Run the Project once end-to-end without changing the prompt, files, or model halfway through. Resist the urge to “tune” mid-run.
- Mark each output as “ship as-is,” “needs editing,” or “wrong.” Note which Instructions correlated with the wrong outputs.
- Change exactly one variable for the second run, usually Instructions. Same Knowledge, new Instructions tells you whether the bottleneck is the system prompt or the source material.
Quality check
- Does the output match the named outcome, not just “look polished”? Run the output against the success criteria in your Instructions.
- Verify facts, page numbers, file paths, and commands independently. Claude will paraphrase your files confidently even when accuracy slips.
- Note remaining human-judgment risks: client confidentiality, copyright on uploaded materials, cost of a wrong direction, production impact of a code change.
How to reuse this workflow
- Save your Instructions as a snippet so you can clone the structure in 60 seconds when a new client or book starts.
- Build a “starter file set” template per Project type: writing always includes
voice.mdandoutline.md; consulting always includesbrand.mdandprior-deliverables.md. - Keep a
lessons.mdper Project — every time Claude misread a file, drifted in tone, or invented a fact. Patterns surface after 3-4 entries and feed back into Instructions. - Rerun a smoke test monthly. Model upgrades, file caps, and UI changes ship often.
Recommended workflow
Client deliverable Project: Instructions describe the client, audience, and tone. Files include brand guidelines, two prior deliverables, and a rolling decisions.md. Each chat handles exactly one section — never two — and ends by appending one paragraph to decisions.md.
Common mistakes
- Uploading 30 files because storage is cheap. Past ten files Claude retrieves the wrong chunks more often and confidence-scores noise as signal.
- Forgetting Project Instructions apply to EVERY chat inside, including unrelated tangents. A strict “respond only in formal English” rule will surface when you ask a casual side question.
- Treating Project files as live documents. They are snapshots until you re-upload. Local edits silently desync.
- Mixing project work with personal experiments. Voice and tone drift fast, and so does your trust in the output.
- Skipping the
decisions.mdhabit. Without it every new chat re-litigates settled questions. - Never pruning. Two-month-old Projects accumulate dead files that drag down retrieval quality.
Advanced tips
- Use clear filenames (
spec.md,decisions.md,brand.md). Claude can then quote them precisely instead of saying “the document you uploaded.” - When you make a major decision in a chat, summarize it in one paragraph and append to
decisions.md. Future chats pick it up automatically. - Pin the 2-3 chats that hold key reasoning — easier than scrolling history and useful for onboarding teammates.
- Keep a
glossary.mdfor any project with custom vocabulary. Claude defaults to industry-standard meanings unless you override them. - For coding Projects, include a
conventions.mdwith naming, error handling, and “never do this” rules. It is more useful than uploading the entire codebase.
FAQ
- Does Claude Projects remember chat history across sessions?: Projects share files and Instructions, not chat history. Re-upload summaries if you want continuity across chats.
- How big can a Project be?: Generous but not infinite. Past roughly 100k tokens of files, retrieval gets sloppier and replies slow down. Keep it lean.
- Can I share a Project with my team?: On Team and Enterprise plans, yes. Personal plans keep Projects private to your account.
- Should I split big work into multiple Projects?: Yes if the deliverables are distinct. One Project per deliverable beats one Project per client.
- How often should I refresh Knowledge files?: Every two weeks at minimum, or after any major decision that changes the source of truth.