I had Ghostlight first on the strength of "actively developed". That is
the reason to keep out, not to go in: a world-elaboration and ontology
rebuild is running in it now, with commits landed 2026-09-05 across 178
branches. Adding a recipe to that tree collides with live work.
codex-connector leads instead -- one service, small blast radius, and it
already consumes the core stack from CultLib rather than forking it.
gjallar second as the cheapest end-to-end proof of the systemd-transient
workload driver: quiet repo, one unit, no compose indirection.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>