Sam approved: merge Tier 1, per-file union for the 3 overlapping skills, repo-as-target standing lane, repo as source of truth.
When I went to merge my Tier 1 branch, main had moved: it was 5c600d7 when I built the branch, and was 62ad94c by merge time. A parallel agent identified in commits as "Claude (Cowork)" had, at 2026-06-01 16:47Z, run a 12-commit reconciliation that synced essentially all mount-only skills to main AND resolved the sa-template-reference conflict explicitly (commit: "Update sa-template-reference to mount v3.4 (2026-05-01, newer than repo)"). The repo-only reference files were preserved (added, not deleted). So the bulk reconciliation and the Tier 2 decision I was holding for Sam were already done by that agent, more completely than my branch.
The fast-forward correctly refused and the merge reported a conflict, so nothing was forced. I inspected main before acting, per the multi-agent coordination rule.
What the parallel agent did NOT have: it synced the mount-original generic salesforce-developer (v1.1.0, no litify_pm__ awareness, no investigate-before-change, phantom router intact). So the value from this thread (the v2.1.0 enhancement and the router fix) was still missing on main.
Action taken: rather than merge a now-redundant branch, I applied just the two improvements surgically on top of main:
Verified live on main: salesforce-developer reports v2.1.0 with litify_pm__ present; operator-core routes to salesforce-developer with the phantom gone. main HEAD 003938ba73, 103 skills. My redundant branch reconcile/mount-to-repo-2026-05-31 was deleted. The 14 original repo-only skills and other agents' branches were left untouched.
Rollback artifacts saved (the pre-patch main versions): /home/claude/sf-skill-work/main-salesforce-developer.PREPATCH.md and main-litify-operator-core.PREPATCH.md. To roll back, restore those two files to main; both target commits changed only those single files.
If that agent re-syncs from the mount, it may overwrite salesforce-developer back to the mount's v1.1.0 and reintroduce the phantom router, because the mount copy was never updated (the web session cannot write the read-only mount). To prevent a regression loop, the mount copies of salesforce-developer and litify-operator-core should be updated to match main when a writable session is next available. Until then, main is canonical and newer for these two files specifically.
The GitHub repo is now the working write lane and is established as source of truth: this session and the parallel agent both wrote to it successfully via the Git Data / contents API with the PAT, with branches and rollback. No read-only wall once work targets the repo. Recommendation stands that the next writable Cowork/Mac session pull main down onto the mount so both sides match and the v1.1.0 regression risk is closed.
The goal of this thread (optimized, org-aware salesforce-developer plus the router fix) is live on main and verified. The broader reconciliation Sam approved was already accomplished by the parallel agent, so re-doing it would have been wasteful and risky. Remaining items are coordination and a mount-down sync, both of which need a writable environment.