claude skills vs projects vs artifacts: which one to use, for what.
a project is where the context lives, a skill is a procedure claude follows when the task calls for it, and an artifact is a thing claude made that you would put in front of someone. that is the whole split, and it comes straight from anthropic's help centre, read 11 november 2026. most confusion comes from putting a procedure in a project's instructions, or expecting a project to produce a document. the table first, then the hr team as the worked example.
the three in one table
| projects | skills | artifacts | |
|---|---|---|---|
| what it is (anthropic's words) | "self-contained workspaces with their own chat histories and knowledge bases" | "skills extend claude's capabilities by giving it access to specialized knowledge and workflows" | "anything claude makes for you that you'd put in front of someone" |
| holds | files, instructions, and in the beta version memory; "relevant documents, text, code, or other files" | a folder with a required skill.md, a description, instructions, and optionally "bundled resources like images or scripts" | documents, code, single-page sites, images, diagrams, dashboards, small tools |
| who triggers it | you open the project and chat inside it | claude: "claude determines when each skill is needed based on your request" | claude, when the output is something to keep |
| plans | all plans; free accounts up to five projects; large knowledge bases (rag) on paid plans | "free, pro, max, ultra, team, and enterprise"; requires code execution to be enabled | "free, pro, max, ultra, team, and enterprise"; requires "cloud code execution and file creation" |
| shared with the team | on team and enterprise: view or edit access per member or whole organisation; admins can turn sharing off | managed under customize > skills; the skills directory lists downloadable ones | new artifacts "start private to you"; sharing settings control who can open them |
| the mistake | pasting a procedure into project instructions and expecting it to run every time | writing a skill with a vague description, so claude never picks it | treating a chat answer as the deliverable instead of asking for the file |
the hr team, worked
people search "claude skills vs projects vs artifacts for hr", so here is that team. the handbook, the benefits summary and the leave policy go in a project called people ops; on a team plan the whole department gets view access and two people get edit. every policy answer claude drafts cites the paragraph it came from, because the source is in the project. the repeatable procedures become skills: "onboard a new hire" (the checklist, the systems to provision, the three emails), "write a job description from a role brief", "prepare a review packet". each has a specific description, because "a specific description tells claude when to invoke your skill." the outputs are artifacts: the onboarding checklist for this hire, the job description for this role, the review packet. they are saved in the artifacts tab, start private, and are shared with the manager who needs them. the project does not change week to week; the skills change when the procedure changes; the artifacts are new every time.
the three questions that pick the container
- does claude need to know this every time, in this area of work? project.
- is this a way of doing something that should run the same way each time it comes up? skill.
- is this a thing someone else will open, read or edit? artifact.
for a sales team the same split is the icp and the playbook in a project, "research an account" and "draft a follow-up" as skills, and the account brief as the artifact; claude for sales teams walks through it. for the plan question underneath (which seats get shared projects), the plans compared.
questions people ask
what is the difference between a claude project and a claude skill?
a project is a workspace with its own chat history and knowledge base: the context. a skill is specialised knowledge and a workflow in a skill.md file that claude invokes itself when your request calls for it: the procedure.
what is a claude artifact?
in anthropic's words, "anything claude makes for you that you'd put in front of someone": documents, code, single-page sites, images, diagrams, dashboards and small tools, saved in the artifacts tab and private to you until shared.
which should an hr team use?
all three: the handbook and policies in a shared project, procedures such as onboarding and job descriptions as skills, and each checklist, job description or review packet as an artifact.
want the containers set up for your team, with the procedures written?
projects, skills and the approval queue installed on your seats, one workflow first, verified in your tools.
see how the install worksrelated: prompts vs skills vs plugins vs mcps · claude project knowledge base · claude for hr teams
claude by role: closers · sales teams · sales managers · customer success managers · customer support teams · account managers · recruiters · hr teams · project managers · accountants and bookkeepers · marketing managers · marketing agency owners · insurance agents · real estate agents · small law firms · e-commerce store owners · therapists in private practice · consultants and coaches · restaurant operators · operations managers · grant and nonprofit teams · business owners · ugc creators. running a small business? start with claude for small business: what it means or see ai implementation for small business.
sources, read 11 november 2026: support.claude.com, "what are projects?" (workspaces, files and instructions, five projects on free, rag on paid plans, team and enterprise sharing with view or edit access, admins can turn sharing off) · support.claude.com, "using skills in claude" (definition, claude decides when a skill is needed, skill.md with a description, bundled resources, all plans, requires code execution, customize > skills) · support.claude.com, "what are artifacts and how do i use them?" (definition, examples, artifacts tab, all plans, "cloud code execution and file creation" setting, new artifacts start private; legacy artifacts before 16 september 2026 can still be published). every quotation is anthropic's own wording.