No existential island risk — but one load-bearing wall we refuse to reinforce. The big "rivals" mostly stand on the same foundation (Claude/open models + SKILL.md, which is our standard). The one structural gap that's real and decisive: model freedom. Competitors govern provider failover as code (hermes model <provider>); we govern it by a prose note in MEMORY.md no daemon reads. When Anthropic throttled us, we had no automatic failover. The adversarial critic argues this should be GOLD #1, and the completeness pass found we never even scouted the cure (LiteLLM / claude-code-router / model gateways). Everything else is surgical pattern-theft, not migration.