se-dev-torch

Original:🇺🇸 English
Translated
10 scripts

Torch plugin development for Space Engineers version 1. Search TorchAPI/Torch source code, plugin API, managers, commands, and server UI patterns.

25installs
Added on

NPX Install

npx skill4agent add cometworks/skills se-dev-torch

Tags

Translated version includes tags in frontmatter

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:
PriorityPatternExampleRoute
1Empty or bare invocation
se-dev-torch
Show this help
2Prepare keywords
se-dev-torch prepare
,
se-dev-torch setup
,
se-dev-torch init
prepare
3Bash/shell keywords
se-dev-torch bash
,
se-dev-torch grep
,
se-dev-torch cat
bash
4Search keywords
se-dev-torch search
,
se-dev-torch find class
,
se-dev-torch lookup
search

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

bash
# 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