How I built an append-only cryptographic audit trail for AI agent actions

## The problem AI agents are making real, consequential decisions — sending emails, calling APIs,...