Pick one.
Claude's set up instantly.

AI training: For Claude subscribers, Anthropic's default is that your conversations are not used to train AI. You can check or change this at claude.ai → Settings → Privacy.

Government requests: Anthropic, like any US company, has to comply with valid court orders. They don't proactively share your data with anyone, and they notify users wherever the law allows.

If your account were compromised: The damage is limited — Claude stores your skill instructions (the "recipes"), not your actual emails, files, or personal content.

Most skills don't need Gmail access at all — they work entirely from what you paste in. Skills that act on your email do need a connection, and Claude walks you through setup when it's needed.

When Claude does access Gmail, it requests minimum permissions — usually read-only — and never sends, books, or submits anything without your explicit "yes" first. Claude reads emails in the moment and does not store or cache content.

You can revoke Claude's Gmail access at any time via Google account settings → Third-party apps.

Claude saves your instructions — the rules, format preferences, and tone you described. It does not store your emails, files, notes, or any personal content. Think of it like saving a recipe: Claude keeps the recipe, not the ingredients you used to test it.

Skills persist across every future conversation — you never have to explain your preferences from scratch again.

Three different things:

Regular chat — Claude starts fresh every time. It doesn't remember your preferences from the last conversation.

Context (sometimes called memory or profile info) — general background you've shared: your name, job, writing style. It helps Claude know who you are.

A skill — specific rules for one repeatable task. It tells Claude exactly how to handle that job, every single time you ask.

Analogy: regular chat is asking a stranger for help. Context is telling them who you are. A skill is training them on precisely how you want one specific thing done.

Yes, any time. Just tell Claude: "Update my [skill name] skill to also…" or "Delete the [skill name] skill." You can also refine it gradually over time — skills get better the more you work with them.

Most skills wait for you to ask — they don't run on their own. But there's a special type called a scheduled skill that runs automatically at a time you set (like every Sunday at 7pm, or every weekday morning).

The catch: your computer needs to be on and Claude needs to be running in the background. If your laptop is closed or off, it won't run until the next time you open it.

Think of it like a reminder alarm — it only goes off if the device is on.

Yes — this is where Connectors come in. Connectors are pre-built integrations that give Claude permission to access specific apps: Gmail, Google Calendar, Slack, your calendar, and others. You set them up once in Claude settings, and Claude can then read from or act within those apps.

Skills and Connectors work great together. A Connector gives Claude access to the app; your skill tells it exactly how to behave once it's there.

Example: you set up the Gmail Connector (so Claude can read your inbox), then create a "meeting-scheduler" skill (so Claude always finds a 30-minute slot, drafts the invite in your voice, and confirms before sending). Without the skill, Claude can access Gmail but will freestyle it. With the skill, it always follows your rules.

The "Connected" skills in this gallery all use Connectors.

Yes. If it's easier to talk than type, record yourself describing what you want — "whenever I ask for a client email, keep it under 150 words, start with something positive, always end with a clear next step…" — then upload the audio file into Claude. It'll transcribe what you said and build the skill from it.

This works especially well if you already have a voice memo where you described how you like things done, or if you want to capture a workflow while it's fresh in your head.

No to all three. Skills are built right into Claude — no terminal, no setup files, no API key required. You create a skill just by having a conversation.

Quick plain-English note on "API key": an API is a back door that developers use to connect their own code to Claude. As a Claude subscriber, you're using the front door — the app — and your subscription covers everything.

If you see "Skills API" mentioned anywhere online, that's the developer version — for engineers building products. Not relevant to you.

Using a skill counts the same as any other Claude conversation — it comes out of your normal usage. No extra charge just for having skills. Skills that generate files (like a spreadsheet or a PDF) might use a bit more since Claude's doing more work, but for most people it's well within what their plan covers.

Skills live in your personal account and can't be shared directly — yet. But you can share the recipe: copy the instructions you used to create it and send them along. They paste it into their own Claude, and the same skill is set up in under a minute. The shortcut isn't transferable, but the instructions to make it are.