Agents — recorded, gated automation
Declare agents over your knowledge and learning ops. Every run is recorded and gated; content changes need a human approval; the execution runtime is not yet wired.
- Every agent run is recorded and auditable
- Content-mutating runs pause for a human approval
- Agents can never touch credential signing, status lists, or exam scoring
- Each run respects a budget cap
Recorded and gated, before it is live
An agent is a declared plan of steps over your knowledge and learning ops. The execution runtime is not yet wired — the step-executor is a stub while the framework is chosen — but every run is already recorded and gated, so the governance works today.
Three controls on every run
- Guardrail — a plan containing a forbidden action (credential signing, status-list changes, exam scoring) is recorded blocked and never runs.
- Approval gate — a plan that would change content pauses as awaiting approval until a human role-holder clears it.
- Budget cap — a run that would exceed its step budget is stopped and recorded blocked.
Approvals
Pending runs appear at the top of the Agents screen. An owner, admin or enablement manager can Approve & run or Reject — the decision is recorded either way.
Agents make no accreditation or competence claim, and can never issue a credential.
Frequently asked
Are agents live?
Not yet. The step-executor is a stub (noop/echo) while the agent framework is being chosen. Runs are still recorded and fully gated — so the governance (approvals, guardrails, budgets) works today — but no live automation runs until the framework decision is made.
What is the approval gate?
Any run whose plan would change tenant content pauses as 'awaiting approval' and waits for a human role-holder (owner, admin or enablement) to approve it. Approving executes the run through the stub; rejecting records the decision and never runs it.
What can an agent never do?
Sign or revoke credentials, change a status list, or score an exam. Those actions are blocked outright — a plan containing one is recorded 'blocked' and never runs. Credential signing is an isolated, human-governed boundary; no agent may cross it.
What is the budget cap?
A per-agent limit on how many steps a single run may take. A run that would exceed it stops and is recorded 'blocked', so a misconfigured agent can never run away.