Archived local sessions hidden and fail to resume after switching auth/provider

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

• #19318
• #19088
• #19345

Powered by Codex Action

Adding the recovery script I used to validate the workaround. It restores archived local sessions by copying archived rollout JSONL files back into the normal sessions/YYYY/MM/DD/ layout and updating the matching state_5.sqlite rows.

It creates timestamped backups first, supports DRY_RUN=1, leaves archived_sessions untouched, and uses $HOME/CODEX_HOME instead of machine-specific paths.

#!/usr/bin/env bash
set -euo pipefail

CODEX_HOME_DIR="${CODEX_HOME:-${HOME}/.codex}"
STATE_DB="${CODEX_HOME_DIR}/state_5.sqlite"
SESSION_INDEX="${CODEX_HOME_DIR}/session_index.jsonl"
ARCHIVED_DIR="${CODEX_HOME_DIR}/archived_sessions"
DEFAULT_ARCHIVED_DIR="${HOME}/.codex/archived_sessions"
SESSIONS_DIR="${CODEX_HOME_DIR}/sessions"
BACKUP_ROOT="${HOME}/codex-recovered-sessions/all-archived-chats"
BACKUP_DIR="${BACKUP_ROOT}/backup-$(date +%Y%m%dT%H%M%S)"
DRY_RUN="${DRY_RUN:-0}"

if [[ ! -f "${STATE_DB}" ]]; then
  echo "Missing Codex state DB: ${STATE_DB}" >&2
  exit 1
fi

if [[ ! -d "${ARCHIVED_DIR}" ]]; then
  echo "Missing archived sessions directory: ${ARCHIVED_DIR}" >&2
  exit 1
fi

mkdir -p "${BACKUP_DIR}"

ARCHIVED_COUNT="$(sqlite3 "${STATE_DB}" "SELECT count(*) FROM threads WHERE archived = 1;")"

if [[ "${ARCHIVED_COUNT}" == "0" ]]; then
  echo "No archived Codex chats found."
  exit 0
fi

echo "Archived chats to restore: ${ARCHIVED_COUNT}"
echo "Backup directory: ${BACKUP_DIR}"

sqlite3 "${STATE_DB}" ".backup '${BACKUP_DIR}/state_5.sqlite.before'"

if [[ -f "${SESSION_INDEX}" ]]; then
  cp -p "${SESSION_INDEX}" "${BACKUP_DIR}/session_index.jsonl.before"
fi

sqlite3 -header -csv "${STATE_DB}" \
  "SELECT id, rollout_path, archived_at, title, cwd, model_provider, model, cli_version FROM threads WHERE archived = 1 ORDER BY created_at;" \
  > "${BACKUP_DIR}/archived_threads_before.csv"

if [[ "${DRY_RUN}" != "1" ]]; then
  probe_path="${SESSIONS_DIR}/.codex-restore-write-test-$$"

  if ! ( : > "${probe_path}" ) 2>/dev/null; then
    echo "Cannot write to ${SESSIONS_DIR}. Run this script from a normal local terminal, outside the Codex sandbox." >&2
    exit 1
  fi

  rm -f "${probe_path}"

  if ! sqlite3 "${STATE_DB}" "BEGIN IMMEDIATE; ROLLBACK;" >/dev/null 2>&1; then
    echo "Cannot write to ${STATE_DB}. Run this script from a normal local terminal, outside the Codex sandbox." >&2
    exit 1
  fi
fi

restore_one() {
  local thread_id="$1"
  local rollout_path="$2"
  local source_path="${rollout_path}"

  if [[ ! -f "${source_path}" ]]; then
    source_path="$(find "${ARCHIVED_DIR}" -maxdepth 1 -name "*-${thread_id}.jsonl" -print -quit)"
  fi

  if [[ -z "${source_path}" && "${ARCHIVED_DIR}" != "${DEFAULT_ARCHIVED_DIR}" && -d "${DEFAULT_ARCHIVED_DIR}" ]]; then
    source_path="$(find "${DEFAULT_ARCHIVED_DIR}" -maxdepth 1 -name "*-${thread_id}.jsonl" -print -quit)"
  fi

  if [[ -z "${source_path}" || ! -f "${source_path}" ]]; then
    echo "missing source: ${thread_id}" >&2
    return 2
  fi

  local filename
  filename="$(basename "${source_path}")"

  if [[ ! "${filename}" =~ ^rollout-([0-9]{4})-([0-9]{2})-([0-9]{2})T ]]; then
    echo "bad rollout filename: ${source_path}" >&2
    return 2
  fi

  local year="${BASH_REMATCH[1]}"
  local month="${BASH_REMATCH[2]}"
  local day="${BASH_REMATCH[3]}"
  local active_dir="${SESSIONS_DIR}/${year}/${month}/${day}"
  local active_path="${active_dir}/${filename}"

  if [[ "${DRY_RUN}" == "1" ]]; then
    echo "dry-run: ${thread_id} -> ${active_path}"
    return 0
  fi

  mkdir -p "${active_dir}"

  if [[ -e "${active_path}" ]]; then
    if ! cmp -s "${source_path}" "${active_path}"; then
      echo "destination conflict: ${active_path}" >&2
      return 2
    fi
  else
    cp -p "${source_path}" "${active_path}"
  fi

  local changed
  changed="$(
    sqlite3 "${STATE_DB}" <<SQL
.timeout 10000
UPDATE threads
SET rollout_path = '${active_path}',
    archived = 0,
    archived_at = NULL
WHERE id = '${thread_id}' AND archived = 1;
SELECT changes();
SQL
  )"

  if [[ "${changed}" != "1" ]]; then
    echo "thread row not updated (${changed} rows): ${thread_id}" >&2
    return 2
  fi

  echo "restored: ${thread_id}"
  return 0
}

restored=0
failed=0

while IFS=$'\t' read -r thread_id rollout_path; do
  if restore_one "${thread_id}" "${rollout_path}"; then
    restored=$((restored + 1))
  else
    failed=$((failed + 1))
  fi
done < <(sqlite3 -separator $'\t' "${STATE_DB}" \
  "SELECT id, rollout_path FROM threads WHERE archived = 1 ORDER BY created_at;")

if [[ "${DRY_RUN}" == "1" ]]; then
  echo
  echo "Dry run complete. Planned restores: ${restored}; failed mappings: ${failed}."
  exit 0
fi

sqlite3 "${STATE_DB}" "PRAGMA wal_checkpoint(FULL);" >/dev/null

remaining="$(sqlite3 "${STATE_DB}" "SELECT count(*) FROM threads WHERE archived = 1;")"
integrity="$(sqlite3 "${STATE_DB}" "PRAGMA integrity_check;")"

echo
echo "Restored: ${restored}"
echo "Failed: ${failed}"
echo "Archived remaining: ${remaining}"
echo "State DB integrity: ${integrity}"
echo "Backup saved at: ${BACKUP_DIR}"

if [[ "${failed}" != "0" || "${remaining}" != "0" || "${integrity}" != "ok" ]]; then
  exit 1
fi

echo
echo "Verify with:"
echo "  codex resume --no-alt-screen 019d0250-8664-7713-bfde-19a0e1110b3f"