rdmsm4x-changelog-20260831-1624-session-continuity-standard-and-macos-update-handoff

rdmsm4x — SESSION_CONTINUITY standard, and handoff for a macOS update

When: 2026-08-31 15:45–16:24 EDT · Host: rdmsm4x · Author: claude@rdmsm4x (dev-39) Scope: ~/dev/lib/app-baseline, ~/dev/sites/dev.ecs0.net, ~/dev/issues, ~/.claude.

Summary

Wrote and published the fleet standard Rich asked for — agents register work when they start and publish state when they stop — then prepared this session for a macOS update. The standard proved itself within the hour by preventing a real collision.

1. The standard

~/dev/lib/app-baseline/SESSION_CONTINUITY.md, published at https://dev.ecs0.net/standards/session-continuity.html (verified 200 with body content, not just a status code).

Key finding: the machinery already existed. claim / heartbeat / release / claims / handoff / next were already in bin/ticket. The gap was that nothing required registration and nothing defined what makes a state file resumable. So this is a standard, not new tooling.

Its core is the six questions a SESSION-STATE.md must answer. The two everyone skips — what is verified vs assumed, with the command that proves it, and which traps were found — are the two that carry the value. A file recording only what was done is a changelog; one recording what is trustworthy is a handoff.

Discoverable three ways: a session-continuity pointer skill (deliberately not a second copy), one line in ~/.claude/CLAUDE.md's load-on-demand table, and the published page. gen_standards.py reads every *.md in app-baseline/, so the page cannot drift from source.

2. It prevented a collision on day one

Rich approved ISSUE-20260831-13 (standards unversioned) and asked for immediate execution. I tried to claim it — the lease refused me and named session ae5eb16e, who had claimed it 32 seconds earlier. I stood off instead of racing two git inits on one tree, and sent them the go-ahead plus pre-commit checks.

They were right and I was wrong. They scoped to app-baseline/ (39df56c, 13 files); my recommendation was git init ~/dev/lib with binaries gitignored. That was worse on three counts, all verified independently:

Size — 5.9 GB, 15 files over GitHub's 100 MB hard limit. I measured this but treated it as something to ignore around.
Nesting — six nested .git dirs under lib/. I never checked; git -C … rev-parse only tests the top of a tree, which is precisely how I got it wrong.
Content — arista/ is 5.4 GB of employer material under ~/dev/RULE-employer-client-material-boundary-20260823.md. Gitignoring it would still have placed it inside a repo boundary bound for a personal GitHub remote. My option would have walked into that while looking careful.

Correction recorded on the ticket rather than quietly superseded.

3. Handoff and auto-resume for the macOS update

~/dev/SESSION-STATE-dev39-rdmsm4x.md — conforms to the new standard's six questions, written for a different agent as well as a resumed one.
TASK-20260831-13 registers the handoff so it is discoverable via ticket list, not only as a file someone must know to look for.
Auto-resume hardened for an OS update: waits up to 60s for Dock (an update can run agents before the desktop is usable), and disarms only after confirming a claude --resume process actually started — osascript returning 0 means only that Terminal accepted the command. If the update resets automation/TCC it exits 1 and stays armed for the next login rather than spending its one shot. Both paths verified in a probe harness.

4. Flagged, not fixed — duplicate standards pages

Another agent hand-wrote wiki/session-continuity-20260831.html (16:02) and wiki/verification-discipline-20260831.html (15:44) beside the generated /standards/ pages. Hand-written standards pages are what gen_standards.py exists to prevent — they have no source file, so they go quietly wrong on the next edit. Not deleted (not mine); consolidation requested on the bus with a suggested route: move any unique content into the canonical .md, regenerate, retire the hand-written page to quarantine.

Verification

/standards/session-continuity.html  200 + body content   (Host header required)
/issues/                            200
app-baseline repo                   39df56c, 13 files, clean
claims held by this session         0  (DOC-20260831-03 released)
unpushed on the recovery branch     0
auto-resume                         armed, not consumed, both paths probe-verified

STILL OPEN FOR RICH — none blocked by the update

TASK-20260831-02 — rotate the Google API key. Owner-only.
Backup remote for app-baseline (ISSUE-20260831-13 follow-on). Dedicated repo only — never a lib/-level one, per the employer boundary above.
ISSUE-20260831-11 — build_site.py reports "verified" without fetching a URL.

After the update

OrbStack must restart before dev.ecs0.net serves; 404/refused right after boot is the stack starting, not a regression. Verify: curl -H 'Host: dev.ecs0.net' http://127.0.0.1:8788/issues/ → 200. Other sessions arm their own resume agents, so several Terminal windows may open at login.