Projects & Artifacts Best Practices
Ten practical habits for getting real, lasting value out of Projects and Artifacts, rather than setting one up once and letting it slowly drift into something less useful.
How to Use This List
- Treat the Project habits (A) as ongoing maintenance, not a one-time setup task.
- Treat the Artifact habits (B) as habits to apply every time you're about to ask for something substantial.
- Come back to group A periodically - Projects tend to drift gradually, not all at once.
- Adapt these to your own workflow; what matters is that scope, instructions, and edit requests all stay specific.
A - Keeping Projects Organized
- Scope every Project around a single, stable purpose. Name it after the specific task and audience it serves, not a broad category, so it's obvious later whether new work still belongs there.
- Write custom instructions as checkable rules, not general advice. "Under 150 words, one clear next step" holds up better over time than "keep it concise."
- Upload only files that answer a real, recurring need. Start small and add files as actual gaps show up in test conversations, rather than uploading everything that seems potentially relevant.
- Review a Project's scope when its instructions start needing exceptions. Frequent "unless it's about X" clauses are usually a sign the Project is covering more than one real purpose.
- Split a Project rather than let it sprawl. When two distinct audiences or rule sets have crept into one Project, separate them into their own Projects instead of continuing to patch a shared one.
B - Keeping Artifacts Easy to Review and Share
- Reserve Artifacts for genuine deliverables, not every long reply. A substantial output you'll likely revise belongs in an Artifact; a long explanation you'll read once is often fine in the chat.
- Make edit requests specific and scoped. Naming the exact section, row, or part to change produces a targeted revision; a vague request like "make it better" risks a broader rewrite than you intended.
- State what should stay the same, not just what should change. This matters more as an Artifact grows longer or more complex, where an unscoped edit has more room to ripple.
- Use version history as a normal part of iterating, not a last resort. Stepping back to a prior version and re-asking with a tighter request is a routine move, not a sign something went wrong.
- Start a new Artifact when the request is really a new deliverable. Folding an unrelated new output into an existing Artifact's edit history makes both harder to follow later.
FAQs
Which group of practices matters more when I'm just getting started?
Group A first. A well-scoped Project makes every conversation inside it better from the start, whereas Artifact habits only matter once you're producing substantial outputs to revise.
How often should I revisit a Project's setup?
There's no fixed schedule, but revisit it whenever the custom instructions start needing frequent exceptions, or whenever the Project's actual use seems to have drifted from its original purpose.
What's the single biggest mistake people make with Projects?
Letting scope creep in gradually - adding one more instruction or one more file at a time until a once-narrow Project covers several unrelated purposes and produces less consistent answers.
What's the single biggest mistake people make with Artifacts?
Giving vague edit requests. A general instruction like "polish this up" tends to produce a broader, less predictable revision than a specific one that names the exact part to change.
Should every Project have uploaded knowledge files?
No. Some Projects depend only on tone and behavioral rules and don't need reference files. Only upload files that answer a real, recurring need for that specific Project.
Is it bad to revert an Artifact to an earlier version?
No, it's a normal part of iterating. Reverting and re-asking with a more scoped request is a routine way to recover from a revision that went further than intended.
How do I know if a Project has grown too broad?
Watch for instructions that need frequent exceptions, files that only apply to some conversations, or answers that feel like they could have come from a plain chat with no Project context at all.
Does splitting a Project into smaller ones create extra work?
Some setup overhead, yes, but it usually pays off in more consistent, specific answers compared to one broad Project trying to serve multiple purposes at once.
Why does the list recommend saying what should stay the same in an edit request?
Because unscoped edit requests have more room to ripple into parts of the Artifact you didn't intend to touch, especially as the Artifact grows longer or more complex.
When should I create a new Artifact instead of continuing to edit the current one?
When the new request is really a different deliverable, not a revision of the existing one. Keeping unrelated outputs in the same Artifact's history makes both harder to follow.
Related
- What Makes a Good Project: Scope and Shared Context - the reasoning behind group A.
- Editing and Iterating on an Artifact - the reasoning behind group B.
- Project Setup Checklist: Instructions, Files, and Scope - a step-by-step checklist for setting up a new Project.
- Projects and Artifacts: Two Ways to Organize Claude Work - the foundational distinction both groups build on.
Stack versions: Written against the Claude model lineup current as of ~June 2026 - Claude Fable 5, Claude Opus 4.8, Claude Sonnet 5 (the default), and Claude Haiku 4.5. Model names, pricing, and product features move quickly - verify current specifics at platform.claude.com/docs before relying on them.