Description
This is an issue I’ve found based on How to: Show current context size
When I try to create a rule that logs the current {context_size}
, {model_name}
after each chat request:
- I get results that differ from the
...
chat dropdown’s context value. - the model I’m using to make the request does not always match the one that’s logged.
tldr: Logging env context size doesn’t match the Chat dropdown’s context size.
Reproduction:
- Add a rule that logs the context size and model after each request (My rule here: spike: adds rule to display info about context size, account info by usrrname · Pull Request #8 · usrrname/cursorrules · GitHub)
- Check the chat dropdown context size
- Change the model and check it again.
Further investigation to inquire as to why including {context_size}
produced a value at all while chatting with gemini-flash :
So accordingly, context_size
is a read-only environment variable… but it differs from the one on the chat!
System info:
Version: 0.49.5
VSCode Version: 1.96.2
Electron: 34.3.4
Chromium: 132.0.6834.210
Node.js: 20.18.3
V8: 13.2.152.41-electron.0
OS: Darwin arm64 24.1.0
Doesn’t block usage of Cursor… but does incentivize me to try other tools that may offer more transparency about context usage.
Related feature requests:
- Usage-based models need to have a counter on the cursor app directly, or a dashboard on the https://www.cursor.com/settings
- Transparency in Token Usage, Search Results, and Context Handling
- Cursor need to share Context window size for all models
- Suggest Adding a Notification When Context Tokens Approach LLM Limit