following-plans

Plan compliance uses an algorithmic decision tree to block unauthorized deviations. It permits syntax fixes and naming corrections while flagging approach or architecture changes as BLOCKED, with explicit STATUS reporting at every step to catch divergences immediately rather than after expensive rework.

Scope management through deterministic decision trees applies when executing multi-step plans where agents might rationalize architectural changes—like discovering OAuth2 exists when JWT authentication was planned. It forces explicit reporting of blockers rather than silent refactoring.

Plan compliance distinguishes between cosmetic changes (syntax, naming) that stay aligned and structural changes (refactored services, protocol swaps) that violate the plan. Only cosmetic changes execute automatically; architecture deviations halt execution and return BLOCKED status.

Status enforcement requires explicit STATUS output at every agent completion, making divergences visible immediately. This eliminates implicit approvals or automatic retries that hide scope violations, surfacing conflicts between discovered reality and the predefined plan.

Yes. When discovered contradictions conflict with the plan—such as finding an incompatible dependency or service architecture—the compliance decision tree reports BLOCKED with explicit reasoning, halting execution until the plan is updated or the contradiction is resolved.