跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
设计与 UI

godot-adapt-single-to-multiplayer

Expert patterns for adding multiplayer to single-player games including client-server architecture, authoritative server design, MultiplayerSynchronizer, lag compensation (client prediction, server reconciliation), input buffering, and anti-cheat measures. Use when retrofitting multiplayer, porting to online play, or designing networked gameplay. Trigger keywords: MultiplayerPeer, ENetMultiplayerPeer, SceneMultiplayer, MultiplayerSynchronizer, rpc, rpc_id, multiplayer_authority, client_prediction, server_reconciliation, lag_compensation, rollback.

thedivergentai · 已有 AI 审核记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add thedivergentai/GD-Agentic-Skills --skill godot-adapt-single-to-multiplayer

技能说明


name: godot-adapt-single-to-multiplayer description: "Expert patterns for adding multiplayer to single-player games including client-server architecture, authoritative server design, MultiplayerSynchronizer, lag compensation (client prediction, server reconciliation), input buffering, and anti-cheat measures. Use when retrofitting multiplayer, porting to online play, or designing networked gameplay. Trigger keywords: MultiplayerPeer, ENetMultiplayerPeer, SceneMultiplayer, MultiplayerSynchronizer, rpc, rpc_id, multiplayer_authority, client_prediction, server_reconciliation, lag_compensation, rollback."

NEVER Do (Expert Multiplayer Rules)

Security & Authority

  • NEVER trust client-reported state — Clients own their 'Input', NOT their 'Position' or 'Health'. Server must validate every coordinate and health change.
  • NEVER use get_tree() groups for authority checks — Use is_multiplayer_authority(). Group registration is non-deterministic in high-latency joins.
  • NEVER allow unrestricted RPC rates — A malicious client can call a 'FireWeapon' RPC 10,000 times per second. Always implement rate-limiting (net_rpc_rate_limiter.gd).

Movement & Lag

  • NEVER skip Client-Side Prediction — Movement without prediction feels 'heavy' and unresponsive. Predict movement locally, then correct only on server disagreement.
  • NEVER sync peers at 60Hz — Sending entire state every frame will saturate client bandwidth. Use a lower tick-rate (20-30Hz) and interpolate between packets.
  • NEVER snap peer positions — Abrupt position updates cause 'jitter'. Store a buffer of past states and lerp between them with a 100ms delay.

Bandwidth & Sync

  • NEVER sync 'Full Floats' if possible — Quantize Vector3 data (truncating decimals) to save 50%+ bandwidth. Use MultiplayerSynchronizer with delta-sync enabled.
  • NEVER ignore 'Late Joiners' — Players who join mid-game won't see existing environmental changes. Broadcast a full world-state 'Snapshot' on peer connection.
  • NEVER test on 0ms ping — Everything works on localhost. Use a simulator (net_latency_simulator.gd) with 150ms ping to identify sync bugs.

Available Scripts

MANDATORY: Architecture decision tree first, then golden-path scripts. Deep latency workflows → references/latency-testing.md.

Authority / transport bridges

multiplayer_sync.gd

MANDATORY when adding MultiplayerSynchronizer interpolation for remote peers. Trigger: authority owns transforms; non-authority interpolates.

rpc_bridge.gd

MANDATORY signal→RPC bridge. Trigger: gameplay emits local signals; bridge validates authority and fans out RPCs.

Prediction / lag / lobby

net_prediction_reconciliation.gd

CharacterBody prediction + input-buffer replay for server reconciliation.

net_snapshot_interpolation.gd

Snapshot interpolation / jitter buffers for remote peers.

net_auth_server_validator.gd

Authoritative validation (position, speed, actions).

net_rpc_rate_limiter.gd

RPC flood / macro protection.

net_interest_management.gd

Distance-based visibility to cut bandwidth.

net_delta_compression_sync.gd

Quantization + significance checks for delta sync.

net_lag_compensation.gd

Server-side rewind for hit registration.

net_lobby_late_join_sync.gd

Late-joiner world snapshot bootstrap.

Diagnostics

net_latency_simulator.gd

MANDATORY before ship — see references/latency-testing.md.

net_debug_overlay_monitor.gd

RTT / loss / jitter overlay.

net_upnp_discovery_logic.gd

UPNP port mapping for listen-server / P2P hosts.


Architecture Patterns

PatternWhenScript golden path
Authoritative serverPvP, economies, cheat riskrpc_bridge.gd → net_auth_server_validator.gd → prediction/recon
P2P lockstep2–4 co-op, low cheat riskDeterministic inputs + net_upnp_discovery_logic.gd
Hybrid / host authorityParty games 4–8Host authority + late-join snapshot

Migration golden path (no inline host/join tutorials)

  1. Separate input (client) from simulation (authority).
  2. Set multiplayer_authority per player node; clients send intents only.
  3. MANDATORY multiplayer_sync.gd for property replication / remote interpolation.
  4. MANDATORY rpc_bridge.gd for gameplay events that cross the wire.
  5. Add prediction / lag compensation scripts only for the genres that need them.
  6. Validate with latency-testing reference + net_latency_simulator.gd at ~150 ms RTT.

Expert insights (WHY — keep in body)

  • Client prediction — WHY: without local sim, RTT doubles perceived input lag. Replay buffered inputs after server correction (net_prediction_reconciliation.gd).
  • Interpolation buffer — WHY: raw sync packets jitter; lerp between snapshots with ~100 ms delay (net_snapshot_interpolation.gd).
  • Hit rewind — WHY: clients fire at past world state; server rewinds RIDs via PhysicsServer3D before raycast (net_lag_compensation.gd).
  • Input send rate — WHY: 60 Hz input RPCs saturate uplink; batch at 20–30 Hz with significance checks (net_delta_compression_sync.gd).

Decision Tree: Which Architecture?

| Factor | Authoritative Server | P2P Lockstep |

阅读完整来源 ↗