Skip to main content

Agents

An Agent is a governed execution participant with roles, skills, limits, context, and permissions.

What this means

In Subbasis, agents are execution participants inside runtime governance. They are not unrestricted bots.

Why it matters

Operational reliability depends on clear accountability. If any agent can do anything, approvals and policy become meaningless.

How Subbasis handles it

Agent behavior is shaped by:

  • role responsibility
  • skill profile / skill pack
  • permission scope
  • policy checks
  • task-level context visibility
  • approval dependencies

This model enables multi-agent execution while preserving control.

Example scenario

In a support escalation workflow:

  • Triage Agent classifies issue and gathers context.
  • Specialist Agent proposes remediation.
  • Operations Agent executes connector actions only if approvals are satisfied.

Each agent role has a different boundary and evidence trail.

What to configure

  • role definitions
  • skill pack assignment
  • permissions and restrictions
  • policy conditions
  • approved connector/model resources
  • escalation and handoff semantics

Limits and deployment notes

  • Agent count, skill profile limits, and advanced controls depend on plan/deployment.
  • Human review is still required for sensitive operations.