https://gitlab.synchro.net/main/sbbs/-/commit/c2d999815f54bdaad2964763
Modified Files:
.claude/skills/logs/SKILL.md
Log Message:
.claude/skills/logs: lessons from FTP rate-limit testing session

Three small additions captured while testing the d7c823c9d rate-limit auto-filter rollout against the live FTP server on vert:

- "Common mistakes" gains a note that `ls`/`stat` mtime lies when
reading sbbsctrl-written `data/logs/{TS,MS,FS,WS}<MMDDYY>.LOG` over
SMB/Samba while sbbsctrl still holds the file open. Observed: `tail
-F` over the mount happily streamed new lines while `ls -la` reported
a 20-minute-old mtime. Local NTFS on the same Windows host updates
mtime as expected, so this is an SMB metadata-caching gotcha, not a
Windows-general one.

- "Quick map" entry for active rate-limit deny pressure updated to
match the new log format the rollout introduced ("Too many requests
per rate limit (NN over NNs) for <IP-or-CIDR>") -- the trailing
`for <IP-or-CIDR>` makes attribution by host or subnet trivial when
RateLimitSubnetPrefixN > 0.

- "Investigation cookbook" gains a recipe for live-watching a Windows
sbbsctrl disk log with `tail -F | grep --line-buffered`, including
a small patterns table. Notes that the post-shutdown `#### Xxx
Server thread terminated (...)` summary line only reaches the
sbbsctrl GUI Log control, not the disk file (just describes the
symptom; the underlying bug is now tracked as gitlab issue #1146).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
--- SBBSecho 3.37-Linux
* Origin: Vertrauen - [vert/cvs/bbs].synchro.net (1:103/705)