The need
Odoo changes between versions, so an AI assistant working from memory will confidently produce APIs that do not exist. Most AI coding guides make this worse, because they freeze details that go out of date into their own text.
What I owned
The plugin’s design and contents: the concepts guide, the research skill, the subagents, the commands and the conventions. I also refined the design through a detailed architecture audit.
The solution
One bet drives the kit: understand through durable concepts, then verify against the real source. The guide says what to look for and where, and the installed code says what is true. If no source is configured, the agents refuse to answer rather than guess.
Key modules
Concepts guide
Thirteen areas of an Odoo module, from the ORM and security to migrations. Each gives the idea, the vocabulary to search and where to look, and it deliberately leaves out line numbers that would go stale.
Subagents
One answers API questions from the real source and cites path and line. One reviews a diff against conventions and known pitfalls. One runs module tests in Docker, never against production.
Commands
Session start, source setup, module scaffolding, migration review, full code review and exporting a Studio view as a deployment script.
Setup
A setup command scans the project, confirms what it finds, writes the configuration and validates it with a test search, so the plugin works on any developer's machine.
How it was built
It is packaged as a Claude Code plugin with no absolute paths, so developers install it from a Git repository and receive updates through it. Non-trivial changes follow a spec-before-code workflow using the template in the kit.
Results
- Developers get answers tied to the code they actually run, with a path and line they can check.
- An architecture audit rated the concept A−, and called the concept and source split the kit’s best decision.