rdmsm4x-changelog-20260925-1014-rc-reset-v102-watcher-arm-time-tier

[2026-09-25 10:14:14 EDT · rdmsm4x] · session 158c9525 (notes-republish-pending-changelog, resumed pid 8222) · work window 10:11–10:14 EDT · after the weekly quota reset; no workers spawned

Summary: claude_rc_reset.zsh v1.0.2 on all six Macs: the after-exit watcher now takes the tier check reported as a third argument and proceeds on it when the API cannot be consulted at exit time, which is what happened at the 2026-09-24 reboot (keychain locked at shutdown → "no-token" → record left → session resumed still broken). Fixture-tested both directions. Also committed 202 uncommitted lines of fleet SESSION-STATE and moved RFC-20260924-01 to needs-decision.

Scope

• Hosts changed: rdmsm4x, rdmbair13m5, rdmbair15m5, rdmpw3265m, rdmpw3275m, jdmbair13m5 — ~/scripts/claude_rc_reset.zsh only (sha256 499752e63f0b953e…, remote checksum and version line verified per host in the same command; spokes installed via temp file + rename so their running watchers keep their old inode).
• rdmsm4x: the watcher for this session re-armed from the new file (after-exit 8222 <sid> standard, pid 53809); old watcher pid 20127 stopped first because the local file was rewritten in place under it.
• No settings, no TCC, no LaunchAgents touched.

Change (v1.0.1 → v1.0.2)

• after-exit <pid> <sid> [standard]: with standard, a no-token / http:* answer at exit time proceeds on the arm-time classification; a real API answer other than standard still wins; any third argument other than standard is refused. CLAUDE_RC_RESET_ASSUME_STALE=1 keeps working. apply's skip message now prints the three-argument form.
• Fixture (scratch CLAUDE_CONFIG_DIR, a security shim returning 44 so no keychain, dead pid, a standard-tier record): no argument → rc 2 "leaving it", record intact · elevated → rc 1 refused · standard → rc 0 cleared, last record verified by the script · second run → rc 2 "no live record any more".

Also this morning

• fleet commit 673b17f: SESSION-STATE.md, 202 appended lines from several sessions since 2026-09-05, committed by explicit path.
• fleet commit 4bf37e2: six tooling snapshots of v1.0.2.
• RFC-20260924-01 → needs-decision (SA-01); comment on INC-20260924-04 closing the v1.0.2 candidate.
• Remote Control state: rdmsm4x 38 records, stale=1 (this session, watcher armed); jdmbair13m5 did not reboot, its 3 watchers (v1.0.0 file, ASSUME_STALE=1) still valid.

Backups / undo

• Previous script versions are in fleet git (parents of 4bf37e2 / f2b855f). No transcript touched this morning.