* Settle Codex review turns from their session completion
* Preserve interruption relay metadata when settling Codex reviews
* Keep the Codex turn id across status writers and settle through the existing host verdict path
The turn id that matches a rollout turn end was dropped by every lead writer
that did not pass it, so a failed turn after a child update never settled.
Drop the synthetic Stop label: the unlabelled poll update already carries the
host's main-agent verdict, including a failure, to current desktops. A remote
desktop now takes the host's settled verdict for a late hook, and a rollout
settle retires hook-only children like Stop does.
* Retire Codex children through the poll and start a fresh turn on any non-compaction session start
A rollout-settled turn deleted the child roster, so the local reconcile
re-seeded the child from the last row as running and the pane stayed Working.
Clear it instead, only from the poll (a child hook settling the root must not
drop that live child), and let a remote desktop clear its roster when the host
reports a done turn with no children. A resumed or forked session no longer
inherits the previous turn id. Also drops accidental pnpm-lock churn.
* Replay a Codex rollout turn end as the root Stop it stands in for
Codex fires no Stop for internal subagents such as /review, so the rollout's
task_complete was settling the main agent through a second path that every
other status writer (local and remote reconcile, child hooks, polls) had to
learn about. The transcript poll now replays that end as an ordinary root Stop
through hook normalization, and a root hook for an already ended turn is
classified as that Stop. A failed end carries its failure outcome and is
published unlabelled like a cancellation, since desktops derive a labelled
Stop's outcome as success. The separate settle path, the remote host-verdict
adoption and the child-hook rollout reads are gone.
* Keep an ended Codex turn retired across compaction, and keep WSL polls alive
A compaction SessionStart queued before the next prompt carried the previous,
already ended turn id, so the poll replayed its Stop: a Done flash and a
second "finished" notification. A compaction now keeps the turn only while
it is still running and its rollout end is unrecorded. The WSL guest
rescheduled its transcript poll with the update instead of the stored row,
so the poll stopped after its first update and a later turn end never
replayed there. A Stop also clears the child roster instead of deleting it,
so a reconcile cannot re-seed retired children from the previous row.
* Restate an ended Codex turn's hook in provider dispatch
Keeps the shared hook normalizer within its size limit after main's merge;
the Codex dispatch returns the event it normalized, which the envelope uses
for the published name and prompt classification.
* Type the rollout record helper in the Codex turn-end test
* Accept Codex's turn_complete name and keep only replayed Stops labelled
The rollout reader also records turn_complete, the name the existing
interruption reader already accepts. A poll row that is not a replay no
longer restates the last hook's event name, so only a replayed Stop can
carry the Stop label.
* Drop a delayed Codex hook for an ended turn the lead has moved past
A late root hook for an older finished turn was restated as Stop and marked
the newer running turn Done; passing it through would instead rewind the
lead's turn id so the poll replays the old end.
* fix: preserve Codex status on ambiguous Ctrl+C input
* fix: confirm Codex turn cancellations from host rollout records
* fix: keep ephemeral side hooks separate from the Codex main turn
* perf: watch active Codex rollouts and skip unrelated records
* fix: retain confirmed Codex cancellation across late relay events
* fix(agent-status): stop an auto-reviewed Codex approval reading as "Needs You"
Codex runs its PermissionRequest hook as decider #1, ahead of both its own
review agent and the user, so the event means "a decision is being made", not
"a human is blocked". Under the "Approve for me" posture the review agent
resolves it seconds later, so every gated tool call drove the pane from Working
to Needs You and back, plus a desktop notification each time.
The execution host now reads the turn's approvals_reviewer off the rollout it
already tails for subagent reconciliation, and keeps a reviewer-owned approval
as working. Positive evidence only: an absent field, an older rollout, or an
unreadable file all still raise the wait, so this can never hide a real prompt.
Splits the incremental rollout JSONL cursor out of the subagent transcript
module, which the new reader pushed over the file-length cap.
* fix(agent-status): avoid stale Codex approval ownership
* fix(agent-status): reconcile Codex child approval ownership
* perf(agent-status): avoid reads for Codex child activity
* fix(agent-status): scope Codex reviewer ownership by transcript
Codex child rows have carried a model field end-to-end since #9637, but the
transcript reader never populated it, so every transcript-discovered child
rendered with an empty model chip. Read the child's own turn_context.model
from the rollout records already fetched for completion detection, so the
sidebar can distinguish an orchestrator model from a subagent model.
No added file I/O and no added rows: the model is parsed from records the
reconcile pass already read, and both row components already render
entry.model.
* fix(agent-status): track Codex rollout subagents
* fix(agent-status): resolve cross-day Codex child rollouts and unblock CI gate
Codex files each rollout under its own local start date, so a session that
runs past midnight spawns children into a sibling day directory. Scanning
only the parent's directory left 13% of real subagent spawns (48/371 across
local rollouts) permanently unresolved, which pinned a phantom "working" row
and re-ran readdirSync every poll tick forever. Resolve the child's own day
directory from occurred_at_ms, and time-box a child whose rollout stays
unreadable so a deleted or never-written file can't leak a working row.
Also make the hook HTTP handler return void: the changed-code quality gate
keys findings by span overlap, so this PR's added line inside the pre-existing
async createServer callback resurfaced no-misused-promises as a new finding.
Tests cover cross-day resolution, grace-period retirement, and that the poll
re-arms across successive roster changes (the prior tests passed even when
the poll died after its first change).
* fix(agent-status): keep the Codex subagent poll alive across nested hooks
A nested non-codex CLI inherits its parent's ORCA_PANE_KEY, so its hook
POST reached scheduleCodexSubagentPoll and tore the timer down before the
source guard, silently ending polling while a rollout child was still live.
Brennan Benson
Neil