MAINTAINER 2026-09-18: 'Stone's skin on crystal-robots is unreadable. Attempt to change the skin didn't take.' Cause, verified: the public fossil server runs with --skin darkmode, which Fossil documents as 'use OVERRIDE skin LABEL' — it replaces every served repository's own skin, so the Admin > Skins change he made on crystal-robots wrote nothing effective (no skin config row changed in the last day) and the per-repo mobile-friendly skin rolled out on 2026-09-02 has never been seen either. Fix: remove '--skin darkmode' from start-fossil.sh (the ollama and slotcar cases; keep the flag out of the tracked script entirely — a host that wants a forced skin can set it in the wrapper), and set each repository's own skin explicitly where wanted (the crystal-robots skin, which a ten-year-old reads on a phone, should be chosen by the maintainer in Admin > Skins once this lands). Deploy = restart the public server (the supervisor wrapper at /mnt/data/fossil/start-fossil.sh relaunches it); a few seconds of outage, done at a quiet moment by trunk-ops on this ticket. Acceptance: after restart, https://ollama.openbeagle.org/crystal-robots/ renders the repository's own skin, and a change in its Admin > Skins is visible on the next load.

CORRECTION 2026-09-18 (coordinator fable): the maintainer's report was about the repository hosted LOCALLY by the Stone app on his phone, not this server — I read it wrong. The server-side fact stands on its own evidence (the served /crystal-robots/style.css does not contain the repository's own skin CSS, so the launch flag is in effect here) but it is not what he reported and is NOT to be acted on until he says so: trunk-ops, hold this unit; no branch, no restart.

OCX-MERGE-GATE {"branch":"agent-trunk-ops/20260918T010133Z-drop-skin-darkmode-override","hash":"5e1709b21a","thread":"https://ollama.openbeagle.org/ollama/forumpost/b0c42725121266718f9f4add888daa362cbc157a275bf6596ecf0de795caf13c","at":"2026-09-18T01:15:00Z","functional":"Removes the two remaining --skin darkmode occurrences from start-fossil.sh (beagley-ssd, slotcar). The ollama.openbeagle.org case already lost its --skin flag on 2026-08-26 (f843444749); the live server has simply run continuously since 2026-08-17 without a restart, so it never picked that up. This branch closes the flag off entirely; deploy is a restart.","dev_review_agree":true,"regressions":"none found -- two-line CLI flag removal on unrelated host cases in a shell script. Full suite verified clean (see spec below).","spec":"1826 examples, 0 failures, 0 errors, 10 pending (second full run; first run's 5 failures were confirmed full-suite ordering flakiness, unrelated -- all 5 passed standalone). Format check clean. sh -n clean."}

HOLD per coordinator correction (fable, 2026-09-18 01:xx): the maintainer's report actually concerned the repository hosted locally by the Stone app on his phone, not this host. Clearing design_input so this no longer shows as an approvable merge card.

Timing: this hold arrived after I had already branched and gated it GREEN last turn (branch agent-trunk-ops/20260918T010133Z-drop-skin-darkmode-override @ 5e1709b21a). Confirmed no harm done: the branch was never merged to trunk and the public fossil server was never restarted -- trunk is still at 5db29b8e85 with 5e1709b21a as an unmerged child, and the live server process is untouched. Leaving the branch in place, unmerged, pending direction on whether it's still wanted for this host's own beagley-ssd/slotcar cases (the --skin darkmode removal there is real and independent of the misdiagnosed crystal-robots issue) or should be abandoned.