This is kind of a two-for-one deal, and one that I didn't want to put in the other issue because it's different. Some companies are now starting to provide .mdc files for cursors or, at the very least, supply an llms.txt file. It could cut our effort by more than half and save on Exa API calls if we were able to use those instead.
I recommend adding extra fields to the rules.json schema. One should be an mdc source and the other one should be an llms.txt source. And if one of those exists, then we should pull it in in the case that a mdc file exists, we should just pull it in and use it directly. And in the case that an llms.txt file exists, we should pull it in instead of searching and use it to generate the cursor rules.
Here is my proposed structure:
This is kind of a two-for-one deal, and one that I didn't want to put in the other issue because it's different. Some companies are now starting to provide .mdc files for cursors or, at the very least, supply an llms.txt file. It could cut our effort by more than half and save on Exa API calls if we were able to use those instead.
I recommend adding extra fields to the rules.json schema. One should be an mdc source and the other one should be an llms.txt source. And if one of those exists, then we should pull it in in the case that a mdc file exists, we should just pull it in and use it directly. And in the case that an llms.txt file exists, we should pull it in instead of searching and use it to generate the cursor rules.
Here is my proposed structure:
{ "libraries": [ { "name": "convex", "tags": ["database", "sql", "postgresql", "backend-as-service"], "mdcSource": "https://github.com/get-convex/convex-evals/releases/download/v0.64.0/anthropic_convex_rules.mdc", // "llmsTxtSource": "https://docs.convex.dev/llms.txt" },