v0.6.0

Spend a rate-limit reset credit from the TUI

Codex grants each account a small number of rate-limit reset credits. Spending one un-exhausts the account immediately, but until now the only way to do it was a hand-built curl.

/openai-reset puts it in the TUI: a preview listing every account with its usage, applicable credits, and whether it is eligible; an explicit confirmation naming what is about to be spent; and a result that reports what the server actually said.

The spend is irreversible, so the command is built to be honest rather than optimistic. A server no-op reports the exact reason instead of claiming success, an ambiguous outcome says it is ambiguous, and a retry reuses the original redemption identifiers rather than minting new ones — so retrying after a dropped connection cannot spend a second credit. Double-clicking confirm, or confirming from two windows, reuses one claim. If the account changed between preview and confirm, the command refuses.

Sticky pins follow the account, not the slot

In sticky-balanced routing, a session is pinned to keep its prompt cache warm. That pin recorded a slot — main, or a fallback's id — and which ChatGPT account occupies a slot can change: switching the main account does it, and so does re-logging a fallback, since re-adding the same account deliberately reuses its id.

When that happened, sessions stayed pinned to a slot now holding a different account, with a cold cache. The pin was protecting a cache that no longer existed while overriding the account you had chosen.

Pins now record the account identity as well, and a pin whose account changed is placed again against real quota. Pins written before this release are kept and stamped rather than dropped, so upgrading does not clear a single warm cache.

Also

• @opentui/core and @opentui/solid updated to 0.5.1.