[FEATURE] Claude Rules/Memory should be Requestable
Problem Statement
By default claude code adds the full content of all existing rules (and memories) from .claude/rules into the context. This is very bad, since quite quickly the context can completely blow up.
The way it should work is similarly to how the cursor handles it, where Claude code gets a description of all the rules that apply (e.g., either via paths, or because they always apply) and then should be able to request the full rule (or just read it). This is also how cursor handles it and its great. It allows the model to figure out when to use rules, and doesn't uneccesarily bloat the context (https://cursor.com/docs/context/rules)
- You should add a description meta-field
- You should make it such that by default only the description of the rule gets added to claude's context.
Ideally you would also just literally adapt the same format for rules as cursor has. Lots of people use cursor + claude code and right now its super tedious to synchronize all the rules becuase of different meta-fields etc.
Proposed Solution
Add a description field to the metadata of each rule. By default, Claude should only see the descriptions of each rule and then be able to read the full rule if they want to.
Alternative Solutions
_No response_
Priority
High - Significant impact on productivity
Feature Category
Other
Use Case Example
_No response_
Additional Context
_No response_
This issue has 4 comments on GitHub. Read the full discussion on GitHub ↗