rdmpw3275m changelog — CLOSE-OUT: fleet Claude Code 2.1.280 + default model rollout

[2026-09-23 14:27:20 → 14:32 EDT · rdmpw3275m · claude Opus 5.5 (1M) · session 5739b30e]

Rich: "resume/complete this work/document/close out accordingly when finished."

Final verified state (14:27:20 EDT)

• All 6 hosts on Claude Code 2.1.280 (claude-code@latest).
• No top-level model / effortLevel pin anywhere. The resolved default is claude-opus-5-5[1m] at medium (DEC-20260922-02), enforced every 30 min by rdmsm4x:~/.agent-coordination/canonical/patch_settings.py via configsync.
• permissions.defaultMode is bypassPermissions on all six, never written.
• 36 of 36 live interactive sessions on 2.1.280: rdmpw3275m 6, rdmsm4x 7, rdmbair15m5 5, rdmbair13m5 5, jdmbair13m5 6, rdmpw3265m 7.
• Every ticket this session opened is resolved: ISSUE-20260902-01, ISSUE-20260906-02, ISSUE-20260921-09, DEC-20260921-01, ISSUE-20260922-11, ISSUE-20260922-12, DEC-20260922-02.
• ISSUE-20260921-15 and ISSUE-20260831-19, where this session contributed, were later closed by other agents.
• The only related ticket still open is TASK-20260906-01, which belongs to codex.

Close-out actions (this pass)

• PROJECTS.md cc-restart row: "Open: jdmbair13m5 upgrade" became "CLOSED OUT 2026-09-23". Still 44 rows. The agent-coordinator row was left as it was, because another agent had already updated it (now v1.13).
• rdmsm4x:~/dev/fleet/cc-restart: added SESSION-STATE.md (counts, final state, lessons) and ISSUES.md (CCR-001: no automated tests; CCR-002: fold the post-reboot relaunch path into the tool). Commit 46a1630, pushed to fleet and backup, clean.
• Earlier today (10:11–10:15):
  • The fleet-notes-publish skill on the hub was corrected. notes_changelog.zsh v1.3 files to the WORK iCloud account [email protected] → llmlog, so verifying in [email protected] gives a false "missing".
  • Memory (here and on the hub): detach long follow-through with a double-fork setsid daemon or a launchd one-shot, not nohup … & disown from ssh.

Where everything is