Server Maintenance

Phased maintenance sweep for big-chungus. The expensive research (file locations, attribution patterns, baselines) is pre-paid in references/ and scripts/sweep.sh — spend agent judgment only on Phase 2 forensics, never on re-discovering the system.

Operating Principles

• Never start with ad-hoc ps/journalctl spelunking. scripts/sweep.sh gathers all of it in one ~2min pass into one report file. Run it first, read the report, work from there.
• The map is the baseline. An observation is only an anomaly relative to references/server-map.md. Compare before investigating.
• One fix at a time, verified. Parallel fixes destroy attribution.
• Read-only until Phase 3. Phases 0-2 must not change system state.
• Always ask both questions of a loop: what kills it AND what revives it. Kill loops on this box have historically been self-inflicted pairs (guardian script + restart policy).
• The skill is self-maintaining. If a sweep reveals the map is stale, updating it is part of the job (Phase 6). Stale maps are worse than no maps.

Phase Routing