rdmbair15m5 changelog — 2026-09-01 02:03–12:50 EDT

Session: monitor local agy sessions, auto-resume at quota reset, reconcile all session output into the canonical root, and land the RDmenubar -> Voight rename. Host: rdmbair15m5 (macOS 27.0, arm64). Mode: prototype.

What changed on this Mac

• Created ~/dev/fleet/agy-session-watch/ — agy-lib.sh, agy-monitor.sh, agy-resume-at-reset.sh, theme-blocks.sh, README.md, SESSION-STATE.md, ISSUES.md.
• Ran the quota-reset auto-resume against 5 blocked agy sessions.
• Sent check-in messages to agy sessions ttys001, ttys004, ttys007, ttys008.
• Ran build.sh in ~/dev/apps/Voight to exercise the universal2 gate for the first time.
• Restarted the background agy monitor (~/Library/Logs/agy-monitor/watch.pid).

What changed on rdmsm4x (canonical)

• Promoted ~/dev/fleet/agy-session-watch/ (7 files, md5-verified both sides).
• Appended ONE row to ~/dev/PROJECTS.md for agy-session-watch. Verified append-only: file went 92 -> 93 lines and the md5 of head -92 was unchanged afterward.
• Pushed 249 pure-add .agents/ files into apps/replicantDB using --ignore-existing, so overwrite was structurally impossible; confirmed rdmsm4x SESSION-STATE.md mtime unchanged.
• Copied this session's changelogs into ~/dev/LLM/Claude/changelogs/.

Nothing else on rdmsm4x was written by this session.

Outcome

Auto-resume. Google quota reset at 05:17:21 EDT, derived from the 429 countdown text (five sessions agreed to the second). Resume fired 05:18:06; canary passed; 5 nudged, 0 skipped; all 7 sessions survived.

Reconciliation — 0 real differences vs rdmsm4x by rsync -rn -c, all .build* excluded: apps/cveDB, apps/Tyrell, apps/replicantDB, apps/Voight. issues/closed/ went 0 -> 48 and EPIC-20260831-02 reached status: closed.

Voight. Rich chose the name at 10:31 EDT. apps/RDmenubar renamed to apps/Voight; bundle identity com.eastcoastscience.Voight with CloudKit container iCloud.com.eastcoastscience.Voight — team-permanent, set before any release. Promoted to rdmsm4x (345 files, 0 differences). No stale RDmenubar directory or PROJECTS.md row remains, so fleet intake cannot ingest it twice under two names.

Findings worth keeping

1. rdmsm4x was AHEAD, not behind. rsync flagged 15 replicantDB Sources/*.swift as "local would overwrite remote" purely on mtime, but git showed rdmsm4x holding a commit on top of this host's HEAD. Pushing would have reverted it and looked like a clean sync. Direction of travel must be established with git, never with mtimes.
2. agy logs contain NUL bytes — grep without -a treats them as binary and suppresses matches, so a quota-blocked session scans as clean. A false all-clear.
3. Build-sandbox naming is inconsistent (.build, .build_auditor_m8, .build-worker1). Only a quoted .build* glob excludes them all; a narrower pattern inflated one diff by ~150k files and caused a correct agent claim to be wrongly doubted.
4. Crash logs are pre-created at 0 bytes for every live agy session; only size > 0 means a crash.
5. Terminal do script ... in tab 1 of (first window whose id is N) delivers text to the tab's foreground process stdin. Window ids are stable; window indices reorder on focus.
6. A relayed peer claim that the sessions were "halted awaiting approval" was false. All 7 were idle at a prompt in accept-edits mode; they had been quota-blocked and already auto-resumed. The claim was checked cheaply rather than acted on blindly.

Verification performed

• Send mechanism proven end-to-end against a throwaway cat before any live session was touched.
• Drift guard tested both directions: valid record passes; dead pid, wrong tty, wrong conversation and stale window id each refused.
• All sync claims re-measured with rsync -rn -c; agent self-reports were not taken on trust (one, the ticket closure, did not hold on first check and was corrected).
• Apple Notes publish verified by three independent reads.

No secrets recorded.