Which agent harness should we support next?

[uipreliga]

Currently built in: Claude Code, Codex, Antigravity, OpenCode and Pi. Agents register through the plugin SPI, so a new one needs no changes to the core.

If you want another one, reply with which — and, more usefully, whether your harness exposes a per-turn token count and which skills engaged. Those two are what decide how much of the eval surface works out of the box; see HARNESS_PARITY.md for what varies today.

Out-of-tree implementations are welcome too.