Fleet changelogs · dev.ecs0.net
rdmpw3265m-changelog-20260905-2107-claude-config-convergence-closeout-and-icloud-correction

rdmpw3265m-changelog-20260905-2107-claude-config-convergence-closeout-and-icloud-correction

When: 2026-09-05 21:07:18 EDT Session: claude@rdmpw3265m (ecdf565b) — resumed the 2026-08-20 fleet Claude-config work to close it out. Where state changed: all writes landed on rdmsm4x. Nothing was modified on rdmpw3265m or any other host.

What changed

  1. NEW rdmsm4x:~/dev/fleet/CLOSEOUT-claude-config-convergence-20260905.md (8,808 B, md5 918fa97b5795f46887de25ef875e667a) Closeout for the 08-20 convergence work: verified fleet state, the context-size number after the v4.0 re-packaging, what shipped, what was deliberately not built and why it must not be revived as designed, and the list of confident-wrong-answers that session produced.

  2. EDITED rdmsm4x:~/dev/_migration/conflicts/rdmsm4x-20260903-nested-review-and-ingest-estates/fleet/_review/mac-fleet-claude-config/ISSUES.md 171 → 193 lines (+22, exactly the block added; 26 top-level entries before and after, headers intact). An inline correction was added to the ~/Documents iCloud entry. The original claim was wrong.

  3. NEW …/mac-fleet-claude-config/CLOSED-SEE-FLEET-CLOSEOUT.md (1,813 B) Marks that archived folder closed and points at the live closeout, so nobody resumes from it.

No files were deleted. No settings were touched on any host.

The correction

This session had recorded that ~/Documents iCloud replication was degraded and not to be relied on. That was wrong and never verified. It came from running a bare brctl status, which enumerates ~278 containers and times out for that reason alone — the timeout was read as a blocked daemon — and then relaying a peer session's unverified reading onward as fact.

Scoped measurement, rdmsm4x, 2026-09-05 20:58:49 EDT, timeout 30 brctl status com.apple.CloudDocs: client:idle, caught-up, last-sync 2026-09-05 20:58:49, appuninstalled:(null). Syncing normally. Not degraded.

Two smaller real findings the wrong claim had hidden: a legacy second CloudDocs container (_7cf15faa…) is caught-up but last synced 2026-07-31 and looks dormant; and several /Documents items sit in Needs Apply Changes → pending-scan for ~24 h, all flagged quarantine:Download|Sandbox|Track — a Gatekeeper scan queue, not a sync failure.

Fleet state verified this session (read-only, one zsh -lc SSH probe per host)

Host permissions.defaultMode ~/.claude/CLAUDE.md
rdmsm4x bypassPermissions v4.2 · 25,380 B
rdmbair13m5 bypassPermissions v4.2 · 25,380 B
rdmbair15m5 bypassPermissions v4.2 · 25,380 B
jdmbair13m5 bypassPermissions v4.2 · 25,380 B
rdmpw3265m bypassPermissions v4.2 · 25,380 B
rdmpw3275m bypassPermissions v4.2 · 25,380 B

6/6 converged on both. Observed only; the mode key was not written on any host, per the standing rule.

Always-loaded context is now 25,380 chars (~6,345 est. tokens) — down ~32% from the 37,545-char peak the 08-20 convergence itself caused, still roughly 2x the 12,334-char pre-cleanup baseline. The file legitimately carries more binding rules now, so this is recorded as a number, not as a defect.