Imported from ericwu13/tai-robot (.claude/skills/log-scan/SKILL.md). Install upstream with npx skills add ericwu13/tai-robot --skill log-scan. Copyright stays with the author.

Log Scan

C:/Python313/python.exe -s scripts/monitor/check_logs.py

Scans, per bot directory, the newest two debug_YYYYMMDD.log files from the last 7 days (older logs are archaeology from retired test bots), plus n8n-service.log and ~/.n8n/n8nEventLog.log.

Pattern → severity

The app's log handler prints no levelname, so severity comes from the message body:

Findings are deduped per (pattern, file) and capped at 5 example lines, each redacted before it is printed.

Gotchas

• No levelnames in the app log — grep ERROR finds nothing useful. Match message bodies (the table above) instead.
• Debug logs are per-TPE-date and never rotated; a busy bot's daily log reaches ~9 MB. The helper tails them (2 MB) rather than reading whole files — do the same if you go in manually.
• Log lines carry two clocks: [<TPE> / <local>]. The first timestamp is TPE. The OS clock is UTC-7, so the second one is 15 hours behind and must never be used for reasoning about sessions.
• swap_strategy refused is often legitimate (one of the 5 safety gates: not flat, veto, in replay, timeframe mismatch, insufficient bars) — read the reason before treating it as a bug.
• A Traceback in a log from months ago is a retired test bot, not an incident; the 7-day window exists for exactly that reason. If you widen it manually, check the bot is one that still runs.

Next steps

• traceback in a live bot → the repo's debug skill (root-cause → fix → review → release)
• n8n.workflow.failed hits → the n8n-debug skill for the actual execution record (execs --errors, then detail <id>)
• Discord send failed → confirm notifications.* wiring with deploy-check