Enforcement & First-Start#

operating-mode.observe-only in config.yml defaults to true and operating-mode.enforcement-enabled defaults to false.

Both must be deliberately changed by an administrator before FlagForge can take any preventable action (CANCEL, SETBACK, KICK, or COMMAND) — this is enforced in code (PolicyEngine/ActionDispatcher), not just by default configuration, so no combination of checks.yml/actions.yml tuning can bypass it.

The hardcoded exceptions#

Two safety guards refuse invalid data regardless of these switches. Neither punishes anyone: each only refuses input no unmodified client sends.

The crash guard in protection.yml drops client packets carrying values no vanilla client sends (an inventory slot outside every menu, a command suggestion nested deep enough to stall the parser, an oversized anvil name), before the server handles them, for every player including flagforge.bypass holders. It is configured and switched off in protection.yml, not here — see Configuration § Crash guard.

packet.payload.a's NaN/infinite-movement-coordinate guard cancels the offending move unconditionally, regardless of observe-only/enforcement-enabled — a non-finite position is never safe to apply, the same way the server-version compatibility guard always applies. This is a hardcoded safety check outside the normal policy pipeline, not a configurable response: it isn't gated by any checks.yml/actions.yml setting, and there is no way to turn it off short of disabling packet.payload.a entirely. See Known Limitations for how this compares to every other check's CANCEL response, which currently has no live trigger at all.

  1. Run with both defaults untouched for at least a few days. Detection, alerting, and staff commands all work fully in this mode — only the actual punishment step is inert.
  2. Review /ff history <player> and Discord alerts to confirm thresholds fit your server before touching anything.
  3. When ready, set operating-mode.enforcement-enabled: true in config.yml (or toggle it live with /ff enforcement on <reason>) and set observe-only: false. A response only ever executes if a check's own response-ceiling in checks.yml and a matching rule in actions.yml both permit it — see the Tuning Guide.