Demo, all content is generated
Question

Cursor codebase answers only seem to know about 10 of my files

Solved · 331 views · asked by noah_s · edited

When I ask "where do we handle refunds" the agent searches and says it can't find anything, but there's literally a refunds.ts. The indexing status in settings has been stuck around 12% for days.

The repo isn't huge in terms of code (maybe 150 files), but it has a data/ folder with exports from our old system.

What I’ve tried

Restarted Cursor, logged out and in, waited overnight. Still 12%.

Comment
What's in data/ and how big is it? pawel_z · edited

3 answers

Marked as helpful by the asker
pawel_z · edited

How big is data/? Indexing chokes on large JSON/CSV exports and generated stuff, and it tends to get stuck on them rather than skip them.

# .cursorignore
data/
venv/
.venv/
*.csv
coverage/

Then Command Palette → resync the index. In Settings → Indexing & Docs you can view which files are included, check that it dropped to roughly your 150.

Note .cursorignore also blocks the agent from reading those files at all. If you still want to @-mention them sometimes, put them in .cursorindexingignore instead, which only keeps them out of the index.

Comment
data/ is 1.8 GB. yeah. noah_s · edited
Indexed in 2 minutes after ignoring it. Refunds found immediately. noah_s · edited
elif_y · edited

Also if there's a committed Python virtualenv in there, that's another 20k files. Happens a lot when the AI created the venv inside the repo and it got pushed.

Comment
femi_o · edited

While it's broken you can still tell the agent to grep: 'search the repo for refund with grep'. The text search tools don't depend on the index, only the semantic search does.

Comment