AMSDAL Claude Code Plugin

Claude Code plugin that gives Claude full knowledge of the AMSDAL framework: models, REST API, ETL, ML, deployment, testing, and the plugin ecosystem.

Prerequisites

• Claude Code installed and working

That's it. The plugin is fully self-contained.

Quick Start

1. Clone the plugin

git clone git@github.com:amsdal/claude-code-amsdal.git ~/claude-code-amsdal

2. Install the plugin

Add it as a local marketplace and install:

claude plugin marketplace add ~/claude-code-amsdal
claude plugin install amsdal@amsdal-plugins

This makes the plugin available in every Claude Code session (CLI and Desktop Code tab).

3. Verify

Start Claude Code in any directory and ask something like:

> "How do I create an AMSDAL model with a foreign key?"

Claude should know the answer without any extra context.

Alternative Installation Methods

Per-project scope

Install for a specific project only (writes to .claude/settings.json):

claude plugin install amsdal@amsdal-plugins --scope project

One-time session (for development/testing)

claude --plugin-dir ~/claude-code-amsdal

What's Included

Skills

Skills are contextual knowledge packs that Claude loads automatically based on what you're working on.

Auto-loaded (Claude picks these up from context):

| Skill | When | What it knows | |-------|------|---------------| | amsdal-overview | Always | Architecture, project structure, how to create apps from scratch, conventions | | amsdal-models | Working with models | Fields, relationships, CRUD, QuerySets, transactions, hooks, migrations | | amsdal-server | Working with server | REST API, authentication, permissions, events, health checks | | amsdal-glue | Working with ETL | Connections, CQRS, queries/commands, planners, executors | | amsdal-ml | Working with ML/AI | Embeddings, AI agents, MCP server, semantic search | | amsdal-ecosystem | Using existing plugins | S3 storage, email, CRM, LangGraph, integrations | | amsdal-frontend-configs | Working with frontend | Dynamic forms, controls, conditions, actions, dashboards | | amsdal-testing | Writing tests | pytest patterns, fixtures, utilities | | amsdal-plugins | Creating plugins | AppConfig, events system, custom routes, middleware |

Manual (invoke explicitly when needed):

| Command | What it does | |---------|-------------| | /amsdal:amsdal-deploy | CLI reference, cloud deployment, monitoring |

Agent: amsdal-expert

Deep research agent for questions that require understanding internal behavior — e.g. "why does QuerySet.get() return a LegacyModel?", "what does AmsdalTransactionManager.commit do on failure?", traceback analysis.

The agent cross-references three sources:

1. Knowledge base (bundled with this plugin) — behavioral descriptions of Cython-compiled modules. 2. .pyi stubs (from your installed AMSDAL packages) — exact API signatures. 3. Source code of pure-Python packages (from your venv's site-packages) — amsdal_server, amsdal-glue, amsdal_cli, etc.

Knowledge Base

Core AMSDAL packages (amsdal, amsdal_models, amsdal_data) are compiled via Cython — you only have .so binaries and .pyi stubs, not Python source. This makes debugging harder because internal behavior isn't visible.

To bridge this gap, the plugin includes a knowledge/ directory with behavioral documentation for every Cython-compiled module. Each file describes step-by-step what the code does internally, edge cases, error conditions, and side effects.

Structure

The directory mirrors the package layout 1:1:

knowledge/
├── amsdal/                              # amsdal_framework internals
│   ├── fixtures/manager.md
│   ├── manager.md
│   ├── services/transaction_execution.md
│   └── ...
├── amsdal_models/
│   ├── classes/
│   │   ├── constants.md
│   │   ├── enums.md
│   │   ├── helpers/reference_loader.md
│   │   └── utils.md
│   ├── querysets/base_queryset.md
│   └── utils/
│       ├── files.md
│       ├── schema_converter.md
│       └── specific_version.md
└── amsdal_data/
    ├── lock/implementations/
    │   ├── redis_lock.md
    │   └── thread_lock.md
    └── transactions/manager.md

How the agent uses it

When you share a traceback like:

amsdal_data.transactions.manager.AmsdalTransactionManager.commit

The agent finds knowledge/amsdal_data/transactions/manager.md, reads the behavioral description of commit(), and explains what happens inside — including the order of operations, when REVERT is issued, and which errors are raised.

The knowledge base is kept up-to-date with AMSDAL releases.

What Can Claude Do With This Plugin?

• Create AMSDAL apps from scratch — models, transactions, config, fixtures
• Write models with fields, relationships, validation, hooks
• Set up authentication and permissions
• Build ETL pipelines with multi-database queries
• Integrate ML features — embeddings, AI agents, MCP server
• Use existing plugins — S3 storage, email, CRM, LangGraph
• Write tests following AMSDAL patterns
• Create custom plugins with events, routes, middleware
• Help with deployment — CLI commands, cloud, monitoring
• Debug production issues — trace through internal behavior of Cython-compiled modules via the bundled knowledge base