The second seat platform. Same record, same exits.
Codex sessions run through the proxy on the workspace's own OpenAI key, with a provider entry sleds init writes and uninstall removes. Bars and credits are read from the seat; a wall on Codex gets the same menu, and a hand-off from a Claude seat can land on a Codex one through a buddy, metered to the right workspace.
$ sleds pickup nn75kg2hyr4r Handed off by jake@ via SLEDS. Pick this up and carry it through. Where it stands: migration 40 min in · tests not yet run · worth high … Seat: openai · gpt-5.6 · 71% of the weekly bar left · metered to workspace acme Estimate: $3–9 · 25–60 min · 69% · recorded est-2f7a working…
Reads
OpenAI bars and credits, the reset.
Model, tokens, cost, the wall, a snapshot at the limit.
The workspace's OpenAI key; batch tier for overnight.
Writes
A sleds provider in Codex's config; removed by uninstall.
The same six, in the session.
Rows keyed by seat and platform, never one word for three rails.
In Codex CLI, in its own words.
Everything sleds writes into Codex CLI is something Codex CLI already knows how to show. If you leave, it keeps the rows.
Next. OpenAI walls with defer and overnight on the batch tier; the second reader for the seat's usage page.