Good direction! But "skill" itself cannot promise what you would actually use it for.

Problem: Each Odoo major version has different way of managing data and different field names sometimes. I am not even talking about the customisations. Odoo studio is other big non-context data.

Solution: 

1. Full context document on How these tables/fields are related to each other (their metadata and context). Which fields are monetary fields. What are the hidden filters Odoo use (like bills and invoices are stored in a same table. Odoo knows but you get all those dumped in a single table). Other Business term and data mapping. This we have already done in apps.odoo.com/apps/modules/19.0/niyu-odoo-ai and the full context is available inside module code.
2. Fine-tuned/trained Qwen or similar model with tonns of Odoo data, use-cases and scenerios. This would then enable us to not only rely on OpenAI wrappers but to have an actual fully local AI model capable of understanding full extend of our Odoo. We are working on creating Odoo only LLM for AI solution needing unlimited calls.
3. 
   
4. Again "skill" could be used as instruction for chain-of-thoughts or formatting but is very limiting way.
5. For further discussion in this regard you may reach out directly at info@niyulabs.com
6. 
   Edit: Replying to Rutger's reply - This is implemented in "apps.odoo.com/apps/modules/19.0/niyu_mcp_server" MCP Server for Ooo with Context. So whenever you ask AI something it first check the knowledge and then call proper tools go get data and re-verify. Without this, you. might have to write model names and maybe field names in prompt. With this MCP you can one-shot any query even in your business terms.
   
   Good catch on context.