cursor-usage-tracker

Summary Running fails to persist spending data because Cursor's API response shape has changed. Two related issues: 1. The endpoint no longer returns , causing a SQLite constraint violation. 2. The endpoint now returns for users under their plan limit and exposes the actual cycle total in a new field, . As a result, even after working around the constraint error, all stored spend values would be . Reproduction Output: is reported, while every other collector (daily usage, groups, usage events, analytics) succeeds. Root cause A live response from (page 1, pageSize 3) currently looks like: Differences from what the codebase expects (see in and in ): — no longer present. The collector binds , which better-sqlite3 sends as , hitting . — semantics appear to have changed. It now seems to represent the overage above the plan-included amount, and is for everyone within plan. — new field. Represents the cycle-to-date total (what the dashboard previously showed as ). New fields seen on members: , . Suggested fix In , normalize the response so the rest of the pipeline keeps working: (total cycle spend) (or if not derivable) Defensive / in bindings as a belt-and-suspenders fix against future schema changes. Update and to document the new fields and the new semantics of . The API reference in should also be refreshed — it currently lists only for this endpoint. Environment Branch: @ Node / SQLite: better-sqlite3 Team size in repro: 50 members, 17 spend pages

Fixes #19. Changes 1. Normalize the new response shape Cursor changed : is now the overage above the plan, the cycle total moved to , and is gone. The collector crashed with , and even past that it would have stored 0 spend for every user within plan. now normalizes each member so downstream code keeps its contract: = (cycle total) = if present, otherwise Pre-2026 responses keep working. Tests cover both shapes, and documents the new response. 2. Fix the top-model query in The query filters on (shared ) but selected without the alias, so SQLite threw . is used by the z-score and trend detectors, so anomaly detection failed. Includes a regression test. This is a separate bug from #19; happy to split it into its own PR if you prefer. Credits Commit 1 is based on @mcobery's work in #21 (closed unmerged), reduced to the minimal fix: no , no fetch error wrapping, no extra defaults (the normalization already guarantees the values) and no unused fields. He is credited as co-author. Checks , , and pass on Node 22. Note: #22 touches some of the same files.