Really thorough breakdown of the current landscape. The part that stood out to me is how each tool handles authentication and session state differently. When you switch between Claude Code, Codex, and OpenCode you end up managing API keys, OAuth tokens, and config files in three different places. It is not just a convenience problem. If you want an agent to run unattended or move between machines, there is no portable identity layer that travels with it. A few teams are working on credential vaults and programmable wallets designed for agent runtimes, so the agent carries its own auth context instead of relying on your shell environment. That would change how these harnesses handle setup and deployment. Did you run into any particular auth workflows that were harder to automate than others?