I Built a Governance Layer That Works Across Claude Code, Codex, and Gemini CLI

I run four AI coding assistants. Claude Code for architecture, Codex for implementation, Gemini CLI for review. Cursor sometimes. The problem isn't that any of them are bad. The problem is that none of them remember what the others did. Every time I switched models, I was re-explaining context, re-establishing decisions, and discovering that the previous model had silently reverted something. On a real API migration last month, Codex deleted an endpoint that Claude had marked as "preserve for 6 months" two sessions earlier. There was no shared record. No handoff. Just vibes. So I built Delimit to fix it. What actually breaks when you switch models Three things, consistently: Context amnesia. Claude drafts a v2 schema with nested address objects. You close the session. Open Codex. Codex has no idea the schema exists. You paste it in manually, but the rationale behind the nesting decision is gone. Ledger drift. You track a task in one session. Switch models. The new model starts fresh, c