Skip to content

AGENTS.md

Latest commit

 

History

History
34 lines (29 loc) · 3.34 KB

AGENTS.md

File metadata and controls

34 lines (29 loc) · 3.34 KB
Raw

Repo Rules

  • Apply karpathy-guidelines first for all code edits: small, surgical, verified.
  • Use context-mode for exploration and analysis: ctx_batch_execute, ctx_search, ctx_execute; avoid raw large output.
  • Headroom is configured as the repo MCP in .codex/config.toml; provider settings stay in /root/.codex/config.toml.
  • Caveman mode active via Claude Code plugin; or is available as project skills under .agents/skills; keep responses short when active.
  • For any MonoGame/client/rendering/performance work, MUST use .codex/skills/monogame/SKILL.md or .agents/skills/monogame/SKILL.md plus relevant MuOnline skills.
  • Verify changes with the narrowest useful check: dotnet build ./MuWinGL/MuWinGL.csproj -c Debug -p:MonoGameFramework=MonoGame.Framework.DesktopGL --nologo

Game Client Facts

  • World coordinates use Vector3(X, Y, Z) with Z as height.
  • Terrain is 256 x 256 tiles; one tile is Constants.TERRAIN_SCALE == 100f world units.
  • Tile/network coordinates convert to world center as tile * TERRAIN_SCALE + TERRAIN_SCALE / 2f; ground-level Z is usually 0f.
  • Camera, visibility, wind, terrain, and world object placement use X/Y for map plane and Z for vertical offsets.
  • Default camera constants live in Client.Main/Constants.cs; yaw/pitch are radians via MathHelper.ToRadians.
  • Runtime data loads from Constants.DataPath, defaulting to AppDomain.CurrentDomain.BaseDirectory/Data.
  • Object transform flow: change Position, Angle, or Scale; WorldObject recalculates WorldPosition.
  • Network handlers can run off the render thread; scene/UI/world mutations should go through MuGame.ScheduleOnMainThread.
  • SpriteBatch state should be managed with SpriteBatchScope when nesting or restoring graphics state.
  • For visual/effect parity, verify against SourceMain5.2 evidence before inventing MonoGame approximations.

Architecture Facts (updated 2026-06-06)

  • ModelObject.Position is Vector3 (world units). WalkerObject.Location is Vector2 (tile coords). Convert: tile * TERRAIN_SCALE + TERRAIN_SCALE / 2f.
  • GameSceneSkillController — handles all skill usage: right-click, area skills, teleport, Nova charging. Cooldowns tracked via SkillCooldownTracker static class.
  • SkillDatabase — static utility loading skill.bmd. Returns SkillBMD with mana cost, damage, range, requirements, delay, mastery type.
  • BuffManager — central buff state via ProcessMagicEffectStatus(playerId, effectId, isActive). Fires BuffStateChanged event consumed by BuffEffectController for visual effects.
  • BuffEffectController — subscribes to BuffManager, applies Swell (scale ×1.25) via WalkerObject.Scale, aura colors for Damage/Defense/ManaShield/Poison/Ice.
  • PacketRouter — manually instantiates handlers with dependencies. BuffManager and PetManager created here; add new managers here and pass to relevant handlers.
  • CharacterState.Class is CharacterClassNumber enum (DarkKnight=16, DarkWizard=0, FairyElf=32, etc.). Knight classes: DarkKnight, BladeKnight, BladeMaster.
  • SourceMain energy formula for skills: 20 + (RequiredEnergy * Level * 4 / 100). Knight gets 10 + .... Summon Explosion/Requiem: 20 + (Energy * Level * 3 / 100).
  • Pet system: PetObject extends ModelObject, follows owner using TilePosition (Vector2 from WorldPosition). PetManager tracks per-owner via _activePets dict.