Safety Model

Safety Model

What this tool refuses to do, what it double-checks before doing, and why.

Protected files

A fixed set of files can never be installed over, no override, no exception:

• Any .exe — this is a blanket rule, not a named list, so anything unexpected still gets caught.
• Explicitly named launcher/service/anti-cheat files even if they lacked the .exe extension for some reason: GTA5.exe, GTA5_Enhanced.exe, GTAV.exe, PlayGTAV.exe, GTAVLauncher.exe, Launcher.exe, GTAVLanguageSelect.exe, RockstarService.exe, RockstarSteamHelper.exe, BattlEye.exe, GTA5_Enhanced_BE.exe, BEService_x64.exe.

If a mod's plan targets any of these, install/inspect reports it as a hard failure. There is no --force for this category.

Conflict handling

Every planned install is checked against the ownership of files already tracked in the database (which mod owns which file), in three tiers:

1. Protected hit — see above. Always fatal.
2. Self-update — the new files overlap mostly with that same mod's own previously installed files (you're reinstalling/updating something). Flagged, but doesn't block you — a one-click "yes, continue" rather than a hard stop.
3. Foreign conflict — the new files would overwrite a file owned by a different, currently-active mod. This requires an explicit override before install proceeds — never silent.

Malware scanning — what it actually is

gtavmm scan shells out to whatever OS-native antivirus is already on your system (Windows Defender's command-line scanner, or clamscan if you have ClamAV on Linux). There is no bundled scan engine or signature database — maintaining one is out of scope for a mod manager, and pretending to have one would be worse than being upfront about not having it.

A specific, real caveat from actual testing: Windows Defender's MpCmdRun.exe -Scan exit codes aren't reliably documented, and in practice exit code 2 can mean either "threat found" or "the scan itself failed" — a completely harmless test file produced exit code 2 with a plain scan error in the log, not a detection. To avoid a false-positive risk serious enough to damage trust in the tool, Defender-backed scans only ever report Clean or Unavailable — never a false "threat detected." clamscan's exit codes are properly documented by ClamAV and are trusted as reported.

Backups before anything destructive

See Backup and Restore — automatic per-mod backups on install, a recycle bin (not instant deletion) on uninstall, and manual full-folder backups for bigger changes.

The AI assistant never acts without a Plan you can see

The AI assistant is opt-in, off by default, and never executes anything directly. Every action it suggests — from a known-fix rule or a diagnosis — expands to an explicit Plan (which mods, which actions) that's shown to you before anything runs. Only two low-risk, reversible action kinds (disable_mod/enable_mod) can ever be whitelisted to skip the manual --yes step; the CLI itself refuses any other action kind on that whitelist, not just the UI. See Advanced Tools § AI Assistant.

This project never downloads mods on your behalf

Source integration is manual-link-only — you paste a URL, the tool doesn't fetch or auto-update from GTA5-Mods.com or LCPDFR.com. Profile import/export and reinstall both require you to supply the actual file yourself.

.rar and .oiv are refused, not guessed at

See Installing Mods — both are explicitly detected and rejected with a clear reason, rather than silently mishandled. A real .oiv sample that appeared to be "simple" was found, during testing, to actually modify RPF archive contents internally in a way the tool couldn't safely detect from assembly.xml alone — that incident is why .oiv is now refused across the board rather than attempting to classify "simple" vs. "complex" packages.