Skills
Saved prompts as runnable cards
Describe the recurring ask once, then it is one click.
Anything you find yourself retyping into the assistant is a skill waiting to be saved. A skill is a card with a name, a prompt, the tools it may use, and a Run button.
What a skill holds
- The prompt
- Written once, in your words. Review my pipeline: stale deals, missing close dates, next steps.
- A time frame
- Day, week or month, substituted into the prompt at run time.
- Tools
- Which of the assistant’s tools the skill may call. Read-only for a report, more for an action.
- Visibility
- Shared with workspace peers, with starter templates provided.
Where to run one
- From the Skills page.
- From the command palette: type the skill’s name.
- From a record’s Research card, with the person already in context.
- On a schedule, as a routine.
Around skills
The parts of the workspace this one works with.
Routines
The assistant on a schedule. Every Monday at eight, summarise last week’s pipeline and post it.
A routine is a prompt with a clock. It runs in its own chat, with your permissions, and tells you when it is done. Anything you would ask the assistant every week is a routine waiting to be written.
The assistant
Your agent, wired into your workspace. Claude, Codex or Cursor’s. It reads what you can read and does what you can do.
The panel is your agent with your workspace in front of it. Claude, Codex or Cursor’s, whichever you connect. It can read records, conversations, meetings and reports, and it can act: create and update records, send and schedule messages, build sequences, run reports, change settings. With your permissions, in your workspace, and nowhere else.
Questions
How is a skill different from a routine?+
A skill runs when you click it. A routine runs on a clock. A routine can run a skill.
Can I use a skill from Claude Desktop?+
Yes. Skills are listed to any connected MCP client.
Can a routine send email?+
Yes, with your permissions. Most people have it draft and let the brief show the drafts for approval.
What if it fails?+
The run is marked failed with the reason, and you are notified.
Can it send mail without asking?+
It can, with your permissions. Ask it to draft and you review, or let it send and it tells you what went out.
Does it see other people’s data?+
Only what your permission profile allows. A member’s assistant sees a member’s view.
Which agent?+
Whichever you connect: Claude, Codex, or Cursor’s. It runs in the app, or locally through the desktop app when one is awake.
Try skills on your own data
Connect your email, then open a person. Most of the product shows up on that one page once the mail is in.
More in Agent
An assistant with your permissions, and the workflows, routines and forms that run without you.
It reads what you can read and does what you can do. With your permissions, in your workspace, and nowhere else.
Every Monday at eight, summarise last week’s pipeline and post it.
A reply arrives, a deal changes stage, a form is submitted. A condition decides, and the workspace does what you would have done by hand.
A block-based builder. A submission creates or updates a person and can trigger a workflow.