Skip to content

URL: https://mkdocs.justinsforge.com/memory/general/reference_bots_health_skill/

/bots-health skill

One-shot, read-only health snapshot of the Telegram bot fleet. Built 2026-06-09 during the fleet audit (see telegram-fleet-audit-brief-2026-06-09).

  • Script: scripts/forge_bots_health.sh (atomic; no side effects, safe anytime)
  • Skill: .claude/skills/bots-health/SKILL.md
  • Covers all five long-pollers: capture, coordinator, general-purpose, notify-bot, remote-bridge

Output sections:

  1. Per bot: systemd state, NRestarts, active-since, last ERROR line from logs/telegram-<bot>.log
  2. Watchdog last line (logs/bot-watchdog.log, should read "ok all 5 bot services active" within 5 min)
  3. Canary last result (logs/canary*.log, daily 06:00, expect "CANARY PASS")
  4. Today's brain latency per telegram invoker from data/claude-quota/<YYYY-MM>.jsonl

Related facts from the same audit: watchdog guards all 5 pollers as of 2026-06-09 (was 3); bot units run TimeoutStopSec=90s so in-flight claude -p children survive restarts; transient getUpdates timeouts log at WARNING, so any ERROR in a bot log after 2026-06-09 is worth a look.

[Claude Code, telegram-fleet session]