Claude Project Instructions Generator
Build custom instructions for a Claude Project — purpose, role, context, tone, and explicit always/never rules, formatted as ready-to-paste markdown.
Claude Project custom instructions set the behavior baseline for every conversation inside that Project, so vague or thin instructions mean re-explaining context in every new chat. This tool builds a complete set of Project instructions covering role, scope, and formatting preferences, structured to actually persist useful context across conversations.
How to use it
What this Project is generally used for — this frames everything Claude does inside it.
What's in and out of scope, plus how you want responses formatted by default.
Ready to use as-is in Claude's Project settings.
Tips for better results
- Include formatting preferences even if they seem minor. Small defaults, like preferring bullet points over long prose, save correcting the same thing repeatedly.
- State scope boundaries explicitly, including what's out of scope. This prevents Claude from over-extending into adjacent tasks the Project wasn't meant to cover.
- Revisit instructions after a few real conversations, not just at setup. Gaps in the instructions usually only show up once genuine, varied use starts happening.
Example output
“Purpose: drafting and reviewing internal engineering documentation. Scope: technical writing and review only, not code implementation. Formatting: prefer numbered steps over prose for procedures, avoid headers under 3 sentences.”
TL;DR
A Claude Project’s custom instructions apply to every conversation started inside that Project, which makes them worth writing deliberately rather than as a quick afterthought. Vague instructions (“be helpful and write good content”) don’t actually change Claude’s behavior in any consistent way; specific ones (a stated role, explicit always/never rules, a named tone) do, because they give Claude concrete rules to check a response against rather than a general mood to aim for.
Why explicit always/never rules outperform general guidance
“Write good content” gives Claude nothing concrete to check a draft against. “Always cite sources with links” and “never use the word revolutionary” both give Claude a specific, checkable rule — the kind of instruction that actually holds up consistently across dozens of separate conversations in the same Project, rather than drifting conversation to conversation the way vaguer guidance tends to. This is the same principle behind this very site’s own Claude Project instructions: specific voice rules and hard “never” rules, not just a general tone description.
What each field contributes
- Purpose is the only required field and anchors everything else — be specific about what this Project is actually for.
- Role gives Claude a consistent persona to hold across every conversation in the Project, not just the first one.
- Context to always assume saves you from re-explaining the same background in every new chat inside the Project.
- Tone/style and output format preference combine into one consistent instruction about how responses should read and be structured.
- Always/Never lists become explicit markdown bullet rules — the most reliable way to make a behavior consistent across many separate conversations.
Purpose: “drafting and editing blog content for our SaaS product,” Role: “our in-house content editor and SEO reviewer,” Tone: direct and concise, Always: “cite sources with links” / “use active voice,” Never: “use the word ‘revolutionary'” / “fabricate statistics.” This produces a formatted document with # Project purpose, # Role, # Tone and style, # Always, and # Never sections, each with the specific content you entered — ready to paste directly into the Project’s Custom Instructions field.
Who this is for
Anyone setting up a Claude Project for a recurring, specific kind of work — a content operation, a codebase, a research workflow — who wants the Project’s behavior to stay consistent across dozens of future conversations instead of being re-explained, and often re-forgotten, each time.
FAQ
Where do I actually paste the generated instructions?
Into the Custom Instructions field in that Project’s settings inside Claude — look for the Project’s settings/edit option once it’s created.
How is this different from a single prompt I’d type into one conversation?
Project instructions apply automatically to every new conversation started inside that Project, unlike a one-off prompt that only affects a single chat.
Can I update these instructions after the Project already has conversations in it?
Yes — updated instructions apply to new conversations going forward; they don’t retroactively change already-completed responses.
Should I combine this with uploaded Project knowledge documents?
Yes, they serve different purposes — instructions govern *how* Claude behaves, while uploaded knowledge documents give it specific reference material to draw on. Most well-set-up Projects use both.
Does this tool store what I type?
No, it runs entirely in your browser.
