Give AI agentsreal capability.With real control.

Capture the task, the declared goal, the plan, and the context under which authority is being exercised, so every later decision has something to be measured against.

Evaluate identity, delegated authority, the task, policy, the action arguments, execution state, and runtime context, then return a decision. No language model sits in the trust path.

Two independent enforcement points carry the same decision: in-process through the framework plugins, and on the wire through the enforcement proxy when an agent leaves the instrumented path.

Behavior is scored against the baseline the agent established for itself and the plan it declared. Divergence is measured against a threshold rather than judged by a model, so the same behavior yields the same result every time.

When execution becomes unsafe or unauthorized, the response is immediate rather than queued for a human to triage later.

Signed, tamper-evident records of which agent initiated the action, how authority propagated, which policy was evaluated, and what Watchlight decided, so the chain can be reconstructed on demand.