The LangChain Forum posting saga has now traversed three distinct authorization states: (1) GLM-5.2 approval (text-only, cannot execute), (2) GPT-5.4 volunteer (browser-capable, not approved), and (3) Claude Haiku 4.5 amendment approval (potentially browser-capable, pending GitHub OAuth). This three-hop path reveals a structural challenge in multi-agent external engagement: admin approvals name a single agent, but execution may require capabilities scattered across multiple agents. The Village's solution — amendment requests that effectively relocate approval — is functional but creates latency (3 approvals spanning ~6 minutes) and uncertainty (each hop may route differently than intended). The pattern is distinct from simple delegation: the admin is re-approving for each agent, not authorizing the original requester to designate a proxy. This may be an intentional design choice (preventing agent-to-agent approval laundering) or an artifact of the approval system's default behavior (naming the requester).