Skip to content

Developer Toolkit (jdt)

The Jarvis Developer Toolkit is a CLI that handles the full package lifecycle: scaffold, validate, test, generate manifests, and deploy. It is the primary way to create, iterate on, and ship Jarvis packages.

Installation

jdt depends on the Jarvis command SDK (jarvis-command-sdk), which is not on PyPI, so install the SDK from git first:

pip install git+https://github.com/alexberardi/jarvis-command-sdk.git
pip install git+https://github.com/alexberardi/jarvis-developer-toolkit.git

This installs the jdt command. Verify with:

jdt --version

Commands at a Glance

Command What it does Speed
jdt init Scaffold a new package with working stubs Instant
jdt test Three-phase validation (manifest + AST + imports) ~2s
jdt validate Fast manifest-only check Instant
jdt manifest Generate/update jarvis_package.yaml from code ~1s
jdt deploy Install to a node (local or SSH; Docker on the legacy install), then restart the node ~3s

Typical Workflow

# 1. Create a package
jdt init my_weather --type command

# 2. Implement your logic
cd my_weather
# ... edit commands/my_weather/command.py ...

# 3. Validate as you go
jdt test .

# 4. Deploy to a node, then restart the node so it loads the package
jdt deploy local .

For a complete walkthrough, see the Development Workflow guide.

What Gets Scaffolded

When you run jdt init, you get a complete, valid package that passes all tests out of the box:

my_weather/
├── commands/
│   └── my_weather/
│       ├── __init__.py
│       └── command.py        # Working IJarvisCommand stub
├── jarvis_package.yaml       # Valid manifest
├── CLAUDE.md                 # AI assistant context
├── README.md
├── LICENSE
└── .gitignore

The stubs are not empty shells --- they implement every required property and method with sensible defaults. You can jdt deploy the scaffold immediately and iterate from a working baseline.

Component Types

jdt scaffolds six component types. Five of them install on a Jarvis Automation node; prompt providers belong to the old Docker install only:

Type Flag Interface Entry File
Command --type command IJarvisCommand command.py
Agent --type agent IJarvisAgent agent.py
Device Protocol --type protocol IJarvisDeviceProtocol protocol.py
Device Manager --type manager IJarvisDeviceManager manager.py
Routine --type routine JSON schema routine.json
Prompt Provider (legacy only; not supported on jarvisd) --type prompt_provider IJarvisPromptProvider provider.py

You can combine types in a single package:

jdt init smart_lights --type command,protocol,manager

This creates a multi-component bundle with all three stubs and a unified manifest.

Validation Pipeline

jdt test runs the structural checks Pantry runs on a submission, in three phases. Passing locally means your package passes those checks; Pantry also runs an AI security review of the code, which can flag a package that jdt test passes.

Phase 1: Manifest Validation
  ├── Schema version, YAML syntax
  ├── Semver format (X.Y.Z)
  ├── Required fields (name, description, version)
  ├── Valid categories and component types
  └── Component paths exist on disk

Phase 2: Static Analysis (AST)
  ├── Correct base class (IJarvisCommand, etc.)
  ├── Required methods/properties present
  ├── No dangerous imports (subprocess, os, shutil, ctypes, importlib)
  ├── No dangerous calls (eval, exec, os.system)
  ├── No raw database access (sqlite3, sqlalchemy)
  └── No shared directory name collisions

Phase 3: Import Checks
  ├── Module imports successfully
  ├── SDK subclass found and instantiated
  ├── Properties return correct types
  └── Examples methods return lists

Next Steps