AI

ambiguous-interactive/fortress-rollback-godot

Developer tools

Эта страница переведена автоматически. Технические детали смотрите в оригинале. Открыть оригинал

0 stars Качество 59 Тренд 20MIT

A of fortress-rollback — a fortified, deterministic, GGPO-inspired rollback networking library — targeting .

Обзор

A of fortress-rollback — a fortified, deterministic, GGPO-inspired rollback networking library — targeting . This project is in an early development phase. Local sync-test sessions and an external-input rollback core are implemented. Fixed-membership byte sessions include bounded input exchange and rollback. Optional SignalFish binary relay and direct WebRTC byte delivery are implemented. Browser release validation remains planned. This repository is developed with heavy AI-assistance and carries a generic (inspired by unity-helpers): - Every agent front-end (Claude Code, Codex, Copilot, OpenCode, Cursor, Gemini CLI, Windsurf, Cline, ...) reads a that delegates to a single canonical context document. - The canonical context and all skills live under .llm/. - Skills follow the open Agent Skills SKILL.md standard, with generated discovery stubs under .claude/skills/ and .agents/skills/. - Every file in the harness is capped at (hard limit), enforced by tests, git hooks, and CI.

README

Показан фрагмент оригинального README. Открыть оригинал.

Fortress Rollback Godot

A GDScript-only port of fortress-rollback — a fortified, deterministic, GGPO-inspired rollback networking library — targeting Godot 4.x.

[!WARNING] This project is in an early development phase. Local sync-test sessions and an external-input rollback core are implemented. Fixed-membership byte sessions include bounded input exchange and rollback. Optional SignalFish binary relay and direct WebRTC byte delivery are implemented. Browser release validation remains planned.

AI-Assisted Development Notice

This repository is developed with heavy AI-assistance and carries a generic agentic harness (inspired by unity-helpers):

  • Every agent front-end (Claude Code, Codex, Copilot, OpenCode, Cursor, Gemini CLI, Windsurf, Cline, …) reads a thin pointer file that delegates to a single canonical context document.
  • The canonical context and all skills live under .llm/.
  • Skills follow the open Agent Skills SKILL.md standard, with generated discovery stubs under .claude/skills/ and .agents/skills/.
  • Every file in the harness is capped at 300 lines (hard limit), enforced by tests, git hooks, and CI.

See .llm/context.md for the full operating manual. GitHub operations use authenticated MCP first and authenticated gh CLI for capability gaps.

Quick Start

Run the offline example with Godot 4.3 or newer:

godot --path . res://examples/offline/offline.tscn

Move the two players with WASD and the arrow keys. The example uses the shipped driver and complete integer snapshots while sync-test repeatedly rewinds gameplay.

Run the packet-loss example with two independent byte sessions:

godot --path . res://examples/peer/peer.tscn

Move players with WASD and the arrow keys. Toggle packet loss, reordering or a partition to observe prediction stalls and recovery. Delivery runs in process.

Try the delayed spectator example to watch a corrected host stream while the host keeps playing. Press P to pause and catch up the viewer. Delivery runs in process; live spectator networking remains planned.

For matches that continue after peer loss, enable the graceful coordinated drop policy. The guide separates supported behavior from the remaining physical-departure and upstream-parity gates. Hot join and session reactivation remain unavailable; see supported features.

For an addon-only install, build and verify a candidate archive.

The supported features page states the supported Godot

Открыть полный README на GitHub

Рекомендуемые инструменты

Попробуйте другой запрос или уберите фильтр.

Установка

npx skillfish add ambiguous-interactive/fortress-rollback-godot
Новичок в MCP? Начните с руководства на mcpskill.org →