declaring-intent

To require an agent to declare intent before a critical action, you mandate an explicit intent record containing the action, target, reasoning, and rollback plan. This forces the agent to expose its thinking as a proposal for human review before execution.

An intent declaration before a database migration must include the explicit action, target, evidence, expected result, abort criteria, verified rollback, decision rights, and backup. This output creates a complete intent record for a reviewer to challenge.

Declaring intent before an irreversible deploy allows a reviewer to challenge the reasoning and verify preconditions before execution. It forces the actor to explicitly define the 'why' and the 'how will I know' rather than just evaluating the outcome.

Yes, intent declaration prevents unauthorized public claims by requiring an explicit proposal with abort criteria and rollback before the action. The agent must state its reasoning and await review, ensuring the stated intent never serves as evidence of understanding.

No, intent declaration requires a verified rollback to ensure reversibility or safe halting. If a data change lacks a rollback mechanism, it fails the explicit abort criteria and backup requirements needed to output a valid intent record for review.

The best way to review an agent's reasoning before a model swap is to generate an intent document stating the action, target, and expected result. This declaration exposes the thinking so a reviewer with decision rights can challenge the proposal before execution.