orchestration: restored Codex terminal reuse times out before task injection (recurrence of #1) #45

Observed recurrence of the timeout symptom in closed issue #1; shared root cause is UNKNOWN. Related open #29 reports terminal_handle_stale, a different observed error; this entry reports restored-terminal reuse returning timeout.

orca --version retained output: 1.4.238-local.1792814572196.b76579683cfd.

Expected: a retained Codex terminal accepts its next Task, or reports a precise unusable-session identity/recovery disposition. Actual: three retained starts failed at agent_readiness with lastError=timeout, no accepted task brief and empty residualResources. Fresh Codex replacements accepted input on the same Tasks without resetting budgets.

Exact diagnostic command preserved in the accepted retrospective:

orca orchestration worker-show --dispatch ctx_d48ae9f7d386 --json

It reports failed readiness, dispatched 2026-10-24 09:27:13, completed 2026-10-24T09:28:13.056Z, startOptions.timeoutMs=60000, and later observation status=identity_changed, exactWorker=false. The later observation does not establish the cause of the earlier timeout.

Literal original launch argv: UNKNOWN (not preserved). Root cause: UNKNOWN. This is observed evidence, not an invented executable reproduction. Durable start options for the diagnostic Dispatch retain the existing terminal, existing worktree, terminal/user_default mode, timeout 60000, setup not_applicable/existing_worktree, and null agent/model/effort selections. Sanitized excerpts below redact only private placement:

{"worktree":"path:<REDACTED-private-worktree>","terminal":"term_2f2cb540-f79f-4e77-8f8d-50db2ff14f96","agent":null,"timeoutMs":60000,"setup":"not_applicable","setupSource":"existing_worktree"}

Three retained failure → fresh accepted-input receipt pairs, each pair on the same Task:

Each failure has state=failed, stage=agent_readiness, failedStage=agent_readiness, lastError=timeout, terminal action reused and residualResources=[]. Each fresh receipt has terminal action created, state=ready, stage=input_accepted, turnStart=observed, effective Codex/gpt-6.1-sol/medium and timeout60000. Successful replacement is not a root-cause fix claim, and missing contact was not treated as process death.

Accepted local retrospective: .orca/2026-10-23-todo-issues/orchestrate/retrospective.md, Orca issues / “Restored Codex terminal reuse times out before task injection (repeat; cause unproven)”. Earlier evidence: .orca/2026-10-05-open-issues-482-486/orchestrate/retrospective.md. Retained diagnostic/version receipts: tmp/retrospective/readiness-show.json and version.json. Private checkout paths are redacted; no raw private files are attached.