telep-master agent worklog

Running worklog for the always-on Claude control-plane agent on telep-mainframe.

2026-08-08 — Provisioned

  • Provisioned the always-on, phone-drivable Claude control-plane agent on telep-mainframe per the approved design (telep-mainframe-handover, 2026-08-08-telep-master-agent-design).
  • Installed tmux (3.5a), cloned the vault to /home/levander/obsidian using the deploy key, and pre-seeded trust so claude remote-control launches headlessly.
  • Installed and enabled control-plane.service (systemd, User=levander) which runs claude remote-control --spawn same-dir inside the persistent tmux session control-plane-telep-mainframe. Verified it connects to the relay.
  • Installed obsidian-sync.service + obsidian-sync.timer (OnBootSec=5min, OnUnitActiveSec=15min) for auto pull/commit/push of this vault. Timer enabled; first push left to schedule.

2026-08-22 — OAuth error / zeroed credentials

  • Control-plane app session threw an OAuth error; control-plane.service looked active but the pane churned Reconnected after Ns. Root cause: ~/.claude/.credentials.json had every token zeroed (access/refresh len 0, expiresAt=0, all mcpOAuth.* empty; only subscriptionType=max intact), rewritten 07:48. Empty refreshToken ⇒ no silent refresh ⇒ full re-login.
  • Fix: interactive claude login in a REAL terminal on the box (ssh -t levander@telep-mainframe '~/.local/bin/claude login'). Shared per-user creds file ⇒ the Restart=always service picked up new tokens automatically, no restart. Verified access/refresh len 108, expiresAt ~8 h, ✔︎ Connected · obsidian · main.
  • Gotchas: claude login needs a TTY (fails via ! prefix / plain ssh); restarting on empty creds just loops; off-LAN only the slow DERP Tailscale path 100.115.209.87 works (need ConnectTimeout 30 s+). Full writeup: 2026-08-22-control-plane-oauth-zeroed-creds.

2026-08-08 — Camera OSD clock fix

  • User reported “right cameras time is broken”. Confirmed via latest.jpg snapshots: ÉSZAK pair (192.168.30.139, camwall right column) OSD read 2026-04-21 04:32 — 109 days behind. Found the DÉL pair (192.168.30.119, left column) was also wrong, +1 h.
  • Root cause: no working NTP on the cams VLAN (GetNTPFromDHCP=true, 0.0.0.0) plus Tapo firmware ignoring the ONVIF timezone/DST fields.
  • Installed /usr/local/bin/cam-timesync.py, /etc/cam-timesync.env (0600), and cam-timesync.timer (hourly, enabled) to push offset-compensated host time over ONVIF. Both panes verified correct.
  • Full writeup: 2026-08-08-camera-osd-clock-drift-onvif-timesync.