mjlab-skillkit

mjlab-skillkit

mjlab-skillkit is a promotion-ready skill kit for AI coding assistants (Codex, Claude Code, Gemini CLI, Cursor, OpenCode) built around two core capabilities for mjlab:

• IsaacLab Migration Skill — port IsaacLab projects into clean, mjlab-native implementations while preserving task behavior.
• mjlab Native Skill — author new mjlab-native tasks, configs, sensors, RL wiring, and mesh workflows directly from local or bundled mjlab docs and examples.

Installation

Interactive Mode

Run the installer to launch the interactive TUI:

cd mjlab-skillkit
bash scripts/install.sh

The TUI supports:

• Multi-select target tools (Codex / Claude Code / Gemini CLI / Cursor / OpenCode)
• Choose install method: copy (production) or symlink (development)
• Preview target paths before confirming

CLI Mode

# Install to a single tool
bash scripts/install.sh --tool claude

# Install to all tools at once
bash scripts/install.sh --tool all

# Use symlink (convenient when iterating on rules)
bash scripts/install.sh --tool claude --method symlink

# Specify a project directory (Cursor / OpenCode)
bash scripts/install.sh --tool cursor --project /path/to/your/project

Install Paths

codex, claude, and gemini install to global user paths by default. cursor and opencode also support --project for project-local installs.

Skill Suite

1) IsaacLab Migration Skill

Use this skill when the job is to move an IsaacLab environment, task, or project into mjlab without losing behavioral intent.

• Preserves parity for rewards, observations, actions, commands, reset/events, terminations, and curriculum.
• Maps IsaacLab concepts to mjlab APIs, managers, sensors, terrains, RL config, and task registration.
• Enforces mjlab-native outputs instead of compatibility shims.
• Includes migration rules, API mappings, gotchas, checklists, and complex-task playbooks.

2) mjlab Native Skill

Use this skill when the job is to build new mjlab-native code directly.

• Authors new tasks, EnvCfg / scene configs, manager terms, sensors, terrain setup, RL config, and task registration.
• Reuses local mjlab examples first, bundled references second, online docs only as a last resort.
• Supports direct mesh-import and asset authoring workflows.
• Helps assistants write code that already matches mjlab structure and public APIs.

What’s Included

• Migration references — rules, mappings, patterns, gotchas, checklist, and task migration playbooks.
• mjlab API pack — focused docs for envs, managers, scene, sensors, simulation, terrains, RL, viewer, and tasks.
• Authoring workflow — step-by-step native mjlab guidance for new code.
• Assistant adapters — installation surfaces for Codex, Claude Code, Gemini CLI, Cursor, and OpenCode.

Who It’s For

• Teams migrating IsaacLab task stacks to mjlab
• Engineers building new mjlab environments from scratch
• AI-assisted coding workflows that need precise, bounded mjlab guidance
• Internal platform or research teams standardizing mjlab development patterns

Scope

• Migrate IsaacLab projects to mjlab-native code.
• Author new mjlab-native tasks, components, and configurations from scratch.
• Preserve behavior parity where migration is the goal.
• Keep outputs aligned with public mjlab APIs and real task examples.

Repository Structure

mjlab-skillkit/
├── SKILL.md                             # Main skillkit entry point
├── agents/openai.yaml                   # Codex/OpenAI agent config
├── references/                          # Domain-specific reference docs
│   ├── migration-rules.md               #   Migration rules
│   ├── mapping.md                       #   Field mapping table
│   ├── patterns.md                      #   Common migration patterns
│   ├── checklist.md                     #   Migration checklist
│   ├── migration-gotchas.md             #   Common pitfalls
│   ├── mjlab-api-index.md               #   API index
│   ├── mjlab-api-*.md                   #   Per-domain API references
│   ├── mjlab-mdp-builtins.md            #   MDP built-in functions index
│   ├── mjlab-authoring-workflow.md      #   Authoring workflow
│   ├── mjlab-authoring-recipes.md       #   Authoring recipes
│   ├── mjlab-mesh-import-guidelines.md  #   Mesh import guide
│   ├── complex-task-migration-playbook.md
│   └── tracking-case-study.md
├── shared/mjlab-skillkit-rules.md       # Shared cross-tool rules
├── adapters/cursor/mjlab-skillkit.mdc   # Cursor adapter format
└── scripts/
    ├── install.sh                       # Installer
    └── package.sh                       # Release packager

Release Packaging

bash scripts/package.sh v0.1.0

Generates:

• dist/mjlab-skillkit-v0.1.0.tar.gz
• dist/mjlab-skillkit-v0.1.0.zip

Promotion Notes

• Use PROMO.md for short-form and long-form launch copy.
• Present the project as a skill kit rather than a single migration helper.
• Lead with the two-skill story: migration + native authoring.

License

MIT License — see LICENSE.