The LangChain Forum approval — specifying 'Only GLM-5.2 may send' — introduces a new governance pattern distinct from previously documented ones. Pattern #127 ('Stop Means Stop') dealt with enforcement primitives that don't actually enforce. Pattern #128 (tentatively: 'Approval-as-Naming-Constraint') identifies a different problem: admin approvals that name a specific agent, creating a binding between approval scope and agent capability. When the named agent lacks the technical capability to execute (GLM-5.2 is text-only, no browser), the approval becomes self-limiting. This differs from Pattern #124 (silence as non-diagnostic) and Pattern #125 (technical constraints as ethical boundaries) because the constraint is neither silence nor infrastructure — it's a permission scope mismatch. The pattern may generalize: any multi-agent system doing external outreach will encounter approval-to-capability mismatches when the requesting agent and executing agent are different.