Skills

A skill is a workflow Brain has learned and can replay on demand. Sales follow-ups, weekly syntheses, coachee check-ins, deal-room digests — anything you do more than once and want consistent.

You should turn a workflow into a skill when:

• You do it more than once a week.
• You want consistent format / tone / structure each time.
• It involves multiple tools (mail + calendar + CRM, say).
• A teammate would benefit from the same recipe.

You don’t need a skill for one-off tasks — just ask Brain in plain English.

After Brain successfully helps you with a workflow, type:

/create-skill

A modal opens with:

• Name — Brain proposes one from the conversation; you can edit.
• Description — short, what this skill does in one line.
• Steps — markdown of the workflow Brain followed, pre-filled.

You edit the fields, click Save. The skill is now stored in your vault (Memory/.claude/skills/) and available next time you mention it.

Mention it by name in chat:

Run the “weekly Acme synthesis” skill.

Or describe the situation and Brain picks the right skill:

It’s Friday, time for the weekly synthesis.

You can also see your skills with:

/skills

When you correct Brain’s execution of a skill (“not that format — keep the bullet style”), Brain captures the correction in the skill itself. Next run reflects it. To rewrite a skill explicitly:

Update the “weekly Acme synthesis” skill — add a section for next week’s calendar.

V1: skills are per-user, kept in your encrypted vault. V2 (planned): export a skill as a Markdown file and share it with a teammate, who can /import-skill it into their own vault.

Anything Brain can do in normal chat:

• Read from your connected accounts.
• Draft mail, summaries, slides outlines.
• Create calendar events.
• Update HubSpot deals.
• Save notes to your vault.

A skill is not magic — it’s a recipe that uses the same building blocks as any other Brain turn.

• Send messages without confirmation (still draft-first).
• Use integrations you haven’t connected.
• Run on a schedule (no cron — you trigger them manually).

If a skill runs but produces a wrong result, treat it like any wrong Brain reply: correct in chat, and the next run incorporates the correction. If it consistently fails, ask Brain to show me the skill’s steps and edit them by hand.