Knowledge versus procedure
Anthropic’s skills article draws the line: Projects supply static background knowledge for the chats within them; Skills activate only when needed and work across Claude. One nuance from the Projects article: on paid plans, large project knowledge is searched with RAG instead of read in full. In plain terms, a Project answers “what should Claude know here?” and a Skill answers “how should Claude do this kind of task?”
| Project | Skill | |
|---|---|---|
| Holds | Knowledge files, instructions, chats, memory | Instructions, scripts and resources for one procedure |
| When it loads | Instructions apply throughout; knowledge is read or retrieved as needed | Only when Claude judges the task needs it |
| Where it works | Inside that one Project | Across Claude: chats, Projects, Cowork, Claude Code |
| Typical content | Client brief, product sheet, contract, codebase notes | “How we write a proposal”, brand guidelines, a report format |
| Team distribution | Share the Project (Team and Enterprise) | Owner provisions it for everyone, or ship it in a plugin |
| Plans | All; Free up to five Projects | Free, Pro, Max, Ultra, Team, Enterprise; needs code execution |
How a Skill loads
When you give Claude a task, it reviews the available Skills, loads the relevant ones and applies their instructions. It loads only what each task needs, which Anthropic says helps prevent overloading the context window. That is the main practical difference with a Project: a Project with forty pages of instructions spends that context in every chat, while forty Skills cost almost nothing until one is needed.
Find them under Customize > Skills. Click “+” and “Browse skills” for the directory. There are four kinds: Anthropic’s own (for example Excel, Word, PowerPoint and PDF handling), custom Skills you write, Skills your organization provisions, and partner Skills built to work with a partner’s connector.
Pick the right one
- The same client, product or codebase keeps coming back: a Project.
- The same way of working applies across clients: a Skill. Writing the proposal format into ten Projects means ten copies to keep in sync.
- Both: a Project per client for the facts, plus one Skill for the procedure. The Skill loads inside the Project when the task matches.
- A whole role’s toolkit (Skills plus connectors plus sub-agents): a plugin, which bundles them and installs as one unit.
- Rules that must apply to every single chat in your account: custom instructions, which Anthropic contrasts with Skills because they apply broadly instead of per task.
The misunderstanding: a Skill is just a saved prompt
A Skill can be only Markdown, but it can also carry scripts and resources that Claude runs, which is why Skills require code execution. That power cuts both ways: Anthropic notes Skills may include or instruct Claude to install third-party packages, so take them from sources you trust. On Enterprise, organizations can turn on skill scanning for uploaded Skills and plugins.
The opposite mistake is also common: putting a procedure in project knowledge as a document and hoping Claude finds it. With large knowledge, Claude switches to searching the files, and a procedure buried in a file may not come up. A Skill is designed to be found by task.
What this means for a team
Skills are the closest Claude comes to sharing how a team works: an Owner on Team or Enterprise can provision a Skill so it appears in every member’s list, enabled or disabled by default. What a Skill does not do is learn. When someone corrects Claude, the correction lands in that person’s memory, not in the Skill; someone has to edit the Skill by hand. And Projects stay bound to the people they are shared with. More on Projects in how to use Claude Projects effectively and what Claude Artifacts are versus Projects.
Where Bombos fits
Skills hold procedures someone wrote down; Projects hold knowledge someone uploaded. Bombos is a plugin that gives every AI tool your team uses, Claude, Claude Code, ChatGPT, Copilot, Cursor and your own agents, one shared memory that also keeps the rules your team teaches it through corrections and has them ready for the next question, with every answer traceable to its source. One person can try it first for their own agents.
Related questions
Can I use a Skill inside a Project?
Yes. Skills work across Claude, so a Skill can load in a chat inside a Project when the task calls for it. The Project gives the background, the Skill the procedure.
Which plans have Skills?
Anthropic lists Free, Pro, Max, Ultra, Team and Enterprise. Skills require code execution to be enabled.
Can I share a Skill with my team?
On Team and Enterprise, Owners can provision Skills for all members. Skills can also be bundled in a plugin, which members can share or publish to the organization when the owner allows it.
Are Skills only for Claude?
No. Skills follow the Agent Skills specification, an open standard published at agentskills.io, so the format is not limited to Claude.
