-
Notifications
You must be signed in to change notification settings - Fork 3
plat 392
| Coordination | Value |
|---|---|
| State | fixed and deployed to SparkQuill; shared provider 648234b
|
| Date | 2026-10-03 |
| Owner | coding-agent-bridge |
SparkQuill's existing parent conversation failed after the CLI confinement change.
Codex 0.160 exited with No saved session found with ID …, even though the
thread's rollout still existed in the installation account's home. The dead-pane
watchdog cancelled the turn, so the UI reported Response cancelled.
The session adopter assumed the ID immediately followed resume. Interactive
Codex puts profile, model and config options before the ID, so the adopter read
--profile and skipped migration. Only the thread selected by the trusted resume
handle should be adopted; no other account sessions may become visible.
A real short-conversation regression then exposed another startup issue: Codex
0.160 retains its old Resuming session banner in scrollback after rendering a
new ready header. The readiness check interpreted that historical banner as
current and waited until cancellation.
Shared provider 7441120 parses option values before the explicit resume ID;
648234b recognizes the newer ready header before an older resume banner.
Regression tests cover actual interactive argv, structured resume, missing IDs,
copying only the selected thread, and retaining the active-resume guard.
The full clisandbox package passed on the SparkQuill Linux host. A real Sol conversation completed, restarted into a different private home, and correctly recalled its previous-turn code (18.27 seconds across both turns). The Codex adapter package and lint passed locally. An unrelated installed-Codex macOS sandbox test fails on the unchanged provider baseline as well.
The affected existing rollout was copied into its private CLI home atomically,
without overwriting an existing private copy or changing its source. The builder
pins the shared fix. SparkQuill release sparkquill-7fc2ae97-20261003181010
runs provider 648234b; all three services, the shared provider/account APIs,
parent/child defaults, and the restored rollout passed live checks. The public
login page returns HTTP 200. Other deployments must update their provider build
to receive these shared fixes.
Auto-synced from docs/ on main. Edit there, not here.