Stop the replaced session on /new for every harness

/new already released the old native session through the chat-close
release path, but two gaps let the replaced session stay alive, and the
client told the operator it could still be resumed in a terminal:

- Claude Code: ReleaseSession only closed a host this server was
  attached to. A host registered for the session but not attached (a
  failed reattach, or one left by an earlier server) kept running. It is
  now asked to close over its socket and killed as a fallback.
- Codex: ReleaseSession only interrupted an active turn. It now also
  sends thread/unsubscribe so the App Server unloads the thread; a later
  prompt reloads it through the existing "thread not found" resume path.
- The /new notice now says the previous session was stopped.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
2026-11-21 18:10:41 +00:00
parent cfebd3a05b
commit 01040a7b12
8 changed files with 183 additions and 15 deletions
+9 -5
View File
@@ -231,11 +231,15 @@ and with the same title. OpenCode makes a new `ses_` session, Codex a new thread
and Claude Code a new session UUID. The server then re-keys the
`operator_sessions` row in one transaction. The new row keeps the title,
directory, tag, backend, colour, star, creation time, model override and mode.
The old row is not deleted: it records `replaced_by`, is no longer listed, and
its native session can still be resumed in a terminal. Before the row is moved,
the server releases the old native runtime; this stops active/background work
without deleting the native transcript. If release fails, `/new` reports an
error and leaves the registry row on the old session.
The old row is not deleted: it records `replaced_by` and is no longer listed.
Before the row is moved, the server releases the old native runtime, the same
release chat close uses, so `/new` never leaves the replaced session running:
Claude Code's host and CLI exit (including a registered host this server is not
attached to), a Codex thread's turn is interrupted and the thread is
unsubscribed so the App Server unloads it, and OpenCode's turn and running
shells are stopped (the shared OpenCode service itself has no per-session
process). The native transcript is kept on disk. If release fails, `/new`
reports an error and leaves the registry row on the old session.
The row is re-keyed rather than mapped from a stable id because every other
layer is keyed by the native id: rooms, queues, transcripts, shells, exports and