Skip to main content
Every Agent node declares a capabilities object: three switches that decide what the node is allowed to do during execution. Asteroid resolves the node’s runtime toolset from these switches and from the workflow’s environment, so the agent gets tools that work where it is running.
Configure capabilities on the Agent node with properties.capabilities.

​
The three capabilities

Browser use

Navigate and interact with web pages (browser_use)

Computer use

Vision-based mouse and keyboard control (computer_use)

Ask user question

Pause mid-run and ask the user for input (ask_user_question)
In settings.yaml, all three are required on every Agent node:

​
Browser use

Gives the node the browser toolset — navigating, clicking, typing, filling forms, reading page content, capturing screenshots, saving PDFs, uploading files, and running Playwright code. This is the default way agents work on the web. Turn it on when the node drives a website. It is enabled by default on browser environments. Browser use also exposes the node-script bridge used by skills such as Google Sheets.

​
Computer use

Gives the node vision-based control of a full desktop: it looks at the screen and moves the mouse and keyboard directly, rather than working through the DOM. Use it for native applications, or for pages that a DOM-based approach cannot handle. It is slower and more expensive than browser use, so prefer browser use when both would work.

​
Ask user question

Lets the node stop and ask a person for input in the middle of a run — a missing value, a decision, a confirmation. The execution moves to paused_by_agent until someone answers. See Execution statuses. Turn it on for workflows with a human in the loop. Leave it off for unattended runs, where a paused execution has nobody to answer it.
This capability is separate from email. Agent inbox tools are available automatically when the execution has an attached profile with an inbox.

​
Environment gating

Browser use and computer use are resolved against the workflow’s environment, and the environment wins:
On a Linux or Windows environment, setting browser_use: true has no effect. If a capability toggle looks like it is being ignored, check the environment first.

​
Runtime features

These runtime features are available to Agent nodes. Each becomes usable once its own prerequisite is met:

Base workspace tools

Read, write, edit, search, run commands, manage todos, and use bundled skills

Google Sheets

Read and write spreadsheet data. Needs browser_use, and the sheet open in the active tab.

Two-factor codes

Generate TOTP codes. Needs a TOTP secret stored as a credential on the agent profile.

Email

Send and read mail. Needs an agent profile attached, since the inbox belongs to the profile.

Handoffs

Move to connected workflow nodes based on outgoing transitions

Output variables

Pass structured data between nodes through output nodes and execution data
These features only work where their prerequisite holds. An agent that cannot use one will tell you which prerequisite is missing rather than guess. See Agent Profiles and Emails.

​
Google Sheets

Google Sheets support is provided by the Google Sheets skill. The skill can read and write cell ranges when all of the following are true:
  • browser_use is available, because the skill runs through the browser node-script bridge
  • The spreadsheet is open in the active browser tab
  • The profile or session has edit access to the sheet
  • The target sheet tab is selected
Google Sheets renders the grid as a canvas. Do not ask the agent to inspect or edit cells by clicking around the grid. Instruct it to use the Google Sheets skill for range reads and writes.
Common prompt pattern:

​
Authentication

TOTP and authenticator-app codes are handled by the Generate TOTP skill. To use it:
  1. Store the TOTP secret on the attached agent profile as a vault credential.
  2. Reference that credential in the node instructions as a ##CREDENTIAL_NAME## placeholder.
  3. Tell the agent to generate the authenticator code immediately before submitting it.

​
Communication

There are two communication paths:
  • Ask the user a question: enable ask_user_question on nodes that may need to pause and wait for a user’s answer.
  • Send or read email: attach an agent profile with an inbox. The runtime exposes send_mail and get_mail automatically for profile-backed executions.
send_mail and get_mail require an agent profile to be attached to the execution. The inbox address is derived from the profile. See Agent Emails for setup and prompt patterns for email verification codes.

​
Context and data flow

For new workflows, pass information between nodes with output variables and execution data:
  • Return structured values from scripted Agent nodes.
  • Configure Output nodes to produce fields that downstream instructions can read.
  • Reference prior outputs from the execution data included in later node prompts.