[giscebot]

Summary

• inspect free space on the filesystem that contains the bridge database
• alert below a configurable 1 GiB default threshold
• skip monitor observability writes while disk space is low
• document the environment and CLI override

This keeps disk I/O failures fatal instead of retrying storage errors after external side effects. The production incident was caused by the database volume reaching 100% usage; the same event produced #343, #344, #345, and #346.

Validation

• pytest -q — 716 passed
• production services active after reclaiming 2.7 GiB from regenerable package caches

Operational follow-up

This detects the condition before SQLite fails; it does not replace host cleanup. Worktree housekeeping still needs separate attention because its last run left 196 entries unverified.

Closes #343
Related to #344, #345, #346

[giscebot]

Tanco aquesta PR perquè ha quedat substituïda per #350. La implementació consolidada manté l’alerta preventiva, combina llindars absoluts i percentuals (5 GiB o 10%) i falla tancat si no es pot mesurar el filesystem. No conserva el salt d’escriptures d’observabilitat d’aquesta versió: amb un llindar preventiu tan primerenc, perdre l’històric de monitoratge seria contraproduent. Mantenir les dues PR obertes dividiria el mateix fix.