Skip to content
BrainRoad BrainRoad
Documentation Menu

Skills

The two built-in BrainRoad skills your AI helper ships with, what they let it do, and how extra skills work.

On this page

What a skill is

A skill is a small packet of instructions and helper scripts that your AI helper reads so it knows how to do a specific kind of work. You do not invoke skills yourself. The helper picks the right one based on what you ask it to do.

BrainRoad ships exactly two skills into every helper workspace. Everything else your helper can do comes from the runtime itself (OpenClaw or Hermes) or from tools you configure separately.

The two built-in BrainRoad skills

brainroad

This skill is your helper’s map of what BrainRoad has set up for it:

  • Business Brain access. How to search your uploaded documents, notes, and knowledge through the Brain connection, and how to answer honestly when your Brain has no real business documents yet. See Business Brain.
  • Reading its email inbox. Your helper can read mail sent to its managed agent mailbox (its @mg.brainroad.com address). Reading only. Sending always goes through review, described below.
  • Reading its phone number. If you have the opt-in SMS number (Pro plan, set up on request), the helper can see the number and read inbound texts, including 2FA codes. It cannot send texts directly. An outbound text is proposed as a card in your Inbox and sent server-side only after you approve it; texts are never sent automatically.
  • Login codes. Generating time-based one-time codes (TOTP) for services where you have stored a 2FA seed.
  • GitHub access. Verifying the GitHub token and SSH key you provision through the dashboard.

If you have not set up one of these (for example the SMS number or GitHub), the skill tells the helper to point you to the dashboard instead of pretending the capability exists.

propose-triage

This is the only way your helper proposes an action that touches the outside world. Instead of acting, it stages a card in your Inbox with Approve, Decline, and Snooze controls. Card types include:

  • Email reply drafts. The most common card. The helper grounds the draft in your Business Brain, then stages it. The send happens server-side only after you approve.
  • Notes to the Brain. Something the helper thinks is worth remembering.
  • Connector calls. A request to call an external API connector you have already registered in Settings. Nothing is called until you approve.
  • Quote estimates. Editable line items with a server-verified total before anything is emailed.
  • Calendar bookings. This emails a confirmation with the appointment details. It does not write to your calendar app.

This is not a courtesy layer the helper could skip. The pod’s network blocks direct outbound email, so even a helper that wanted to send mail itself cannot. All sends run through BrainRoad’s server after your approval. Repeatedly approved low-risk email replies may later earn narrow, capped autonomy that you can pause or revoke; high-stakes messages remain manual.

Adding more skills

Skill management beyond the two built-ins belongs to the runtime, not the BrainRoad dashboard. OpenClaw and Hermes each have their own skills screen, opened from the dashboard into the runtime’s own interface. What is available there depends on the runtime and its ecosystem.

Two honest caveats before you add skills:

  • Extra skills often need configuration or a separate account. A skill that talks to an outside service usually needs your own credentials for that service, and may have its own costs.
  • Third-party tools keep their own authority model. BrainRoad’s review step covers actions routed through BrainRoad’s own executor (email replies, connector calls, quotes, bookings, Brain notes). A tool you wire up directly inside the runtime is governed by that tool’s own permissions, not automatically by your Inbox.

Skills and isolation

Skills run inside your helper’s private, isolated workspace. Nothing you install affects any other user, and no other user’s skills can reach your environment.