Se Dev Torch

SE Dev Torch Skill

Torch plugin development for Space Engineers version 1.

This skill is for Torch framework itself: plugin lifecycle, manifests, commands, managers, patch helpers, and server UI integration. Use se-dev-game-code or se-dev-server-code alongside it when task crosses into Keen internals.

Compatibility: Plugins built with this skill target Torch only — not compatible with Magnetar, which uses different patcher and SDK. For Magnetar server plugins use se-dev-plugin and se-dev-plugin-sdk skills. When starting from se-server-plugin-template, use its last-torch-compatible tag as basis, not current main (which dropped Torch target). See TorchPlugin.md.

CRITICAL: Commands run in a UNIX shell. Use bash syntax. On Windows this is BusyBox; on Linux/macOS use the native shell.

Examples:

• test -f Prepare.DONE && echo ready
• TORCH_ROOT=/path/to/Torch ./Prepare.sh
• if exist Prepare.DONE echo ready

Actions:

• prepare: Run one-time preparation (Prepare.bat on Windows, Prepare.sh on Linux/macOS, or run_prepare.sh as wrapper)
• bash: Run UNIX shell commands via busybox
• search: Search indexed Torch source using search_torch.py

Routing Decision

Check these patterns in order:

Requirements

Host system must have following on PATH:

• Python 3.11 or newer
• git command line client

If user already has local Torch checkout, set TORCH_ROOT to that repository root before preparation. Otherwise skill clones https://github.com/TorchAPI/Torch into its persistent data folder.

Getting Started

CRITICAL: Before running ANY commands, read CommandExecution.md.

If Prepare.DONE file missing in this folder, run one-time preparation first. See Prepare.md.

Essential Documentation

• CommandExecution.md - command execution rules and quick start
• QuickStart.md - first searches and common task entry points
• TorchPlugin.md - Torch plugin authoring patterns
• Architecture.md - where to inspect each subsystem in Torch source tree
• CodeSearch.md - complete code search reference
• Troubleshooting.md - common failure modes and what to check

Quick Search Examples

# Find the base plugin class
uv run search_torch.py class declaration TorchPluginBase

# Find the Torch plugin API interfaces
uv run search_torch.py interface declaration ITorchPlugin -n Torch.API.Plugins

# Inspect command definitions
uv run search_torch.py class declaration CommandModule -n Torch.Commands
uv run search_torch.py method signature RegisterPluginCommands

# Inspect manager usage patterns
uv run search_torch.py class declaration DependencyManager -n Torch.Managers
uv run search_torch.py interface declaration ITorchSessionManager -n Torch.API.Session

# Find patching infrastructure
uv run search_torch.py namespace declaration PatchManager
uv run search_torch.py class declaration PatchContext -n Torch.Managers.PatchManager

General Rules

• Search only indexed Torch repository selected during preparation.
• Prefer manifest.xml metadata over [PluginAttribute]; PluginAttribute is obsolete in Torch.
• Prefer torch.Managers.GetManager<T>() or session.Managers.GetManager<T>() over obsolete ITorchBase.GetManager<T>().
• Use TorchPluginBase as normal starting point unless task clearly needs lower-level implementation.
• If task is about Space Engineers game behavior rather than Torch framework behavior, use se-dev-game-code or se-dev-server-code as companion skill.
• Preparation builds a separate Graphify graph for the selected Torch checkout (or SE_DEV_TORCH_PLUGIN_ROOT). With the fast Rust clustering backend (Python 3.12 via uv, provisioned automatically) prepare builds it automatically (~1 minute for a Torch checkout). Where the fast backend is unavailable it stays opt-in with SE_DEV_GRAPHIFY=1; SE_DEV_GRAPHIFY=0 disables it. Read on demand — build via GraphifyPrepare.md, query via GraphifyUsage.md.

Remarks

Original source of this skill: https://github.com/CometWorks/skills