What do you think about Cursor removing the codebase indexing settings?

I will put this in a fresh thread since this is an old thread.

In a nutshell grok ignores my rules file and is jumping into my old handoffs folder for stagnant ideas and writing regressive code.
Request id: ca23c639-b7f7-4ab7-8b28-b89d8b125ddc
Cursor: 3.19.13 (Windows)
Repo: single Python/Starlette/HTMX app. Current constraints live in .cursor/rules/, CONTEXT.md, ui-surface.md, docs/adr/, docs/forward/, src/. Also git-tracked: handoff/ (~80 historical session markdown files), parked research under docs/, unpromoted tools/drafts/. Those archives are useful when I @ a named file to resume work. They are poison if Grep/Read treat them as current product truth.

What used to work: .cursorindexingignore kept those trees out of automatic retrieval while leaving them @-mentionable. .cursorignore stayed for PII/secrets (hard block). That split is the whole point of a long-lived repo with living rules plus a session graveyard.

What is broken now:

No index-only ignore. Instant Grep + file reads still walk git-tracked handoff/ because it is not in .gitignore. Putting it in .cursorignore would hide it from @ as well, which I do not want. Putting it in .gitignore would drop it from version control, which I also do not want.
.cursorindexingignore is gone from the docs. Ignore file only documents .cursorignore and .gitignore. There is no documented replacement for “search skip, explicit @ still allowed.”
No resync control. Cursor: Resync Index / Delete Index are gone from the palette (expected per this thread), so there is also no way to confirm ignore changes took effect.
Rules cannot close the hole. Agents still Grep handoff/ on ordinary tasks and pull early-project decisions that contradict the current .cursor/rules set.
Ask: restore an index/search-only ignore (old .cursorindexingignore behavior) that Instant Grep and agent Grep/Glob honor, while @ and explicit path reads still work. Semantic embeddings do not have to come back. The missing piece is a third ignore layer between “never readable” and “fully searchable.”

Repo is not huge (well under any old 250k-file cap). The failure is wrong files, not too many files.