Codex resume failed: cannot reopen a session

What you see

You close the client, or it crashes, and you cannot get the conversation back. Depending on the cause you may see:

• Failed to resume chat: paginated threads do not support initialTurnsPage; use turnsBackwardsCursor and itemsBackwardsCursor
• Error creating chat: paginated threads do not support initialTurnsPage; use turnsBackwardsCursor and itemsBackwardsCursor
• A store-level failure quoting a line beginning failed to read thread and ending with CRC (os error 23) — the middle is elided in the reporter’s own write-up, not literal text, so the string must not be published whole.
• A window that simply will not open, resume, or send messages.

What it usually means

Resume failure has at least three separable causes in the material, and the guide cannot rank them:

1. A client or protocol regression after an update. One reporter’s resume failure began right after updating the desktop client; the issue was closed as completed with no stated fix.
2. On-disk store corruption. One reporter’s opening, resuming and sending all failed with a failed to read thread … CRC (os error 23) store error, on the desktop app.
3. Provider or compaction items that cannot be loaded on resume. A title-only lead mentions an item type "compaction" is not supported error from third-party Responses providers; the body was not opened, so it is a lead, not established content.

The documented session command codex resume exists (“Continue a previous interactive session by ID or resume the most recent conversation”), but this research could not retrieve its flags.

Quick checks

1. Did the failure start right after a Codex client update? One reported resume failure followed an update.
2. Does the message name initialTurnsPage / paginated threads? That points at the pagination-regression shape.
3. Does the message mention failed to read thread / CRC? That points at store-level corruption.
4. Can you still open other sessions? If only one conversation is affected, the store for that session is the suspect.

Fixes, by branch

If it names paginated threads / initialTurnsPage

• Note the exact client version and that an update preceded it. One report was closed as completed with no stated resolution, so updating alone is not proven to fix it.
• Try codex resume with no flags; its flags were not retrievable, so do not add --last, --all, or a session-id you are unsure of.

If it names a store read failure (CRC / failed to read thread)

• Do not delete the session store. The reporter who hit this backed up ~/.codex/sessions and ~/.codex/archived_sessions before touching anything; no source recommends deletion.
• Move the affected store aside with a backup rather than deleting, so the action is reversible.

If resume fails after an update with no clear message

• Record the version before and after, and whether a different machine opens the same session. This is the branch with the thinnest evidence.

What not to do

• Do not delete ~/.codex/sessions, ~/.codex/archived_sessions, history.jsonl, or the SQLite state DB to recover a session. No source recommends it; back up and rename aside instead.
• Do not invent codex resume flags. Only the bare codex resume is documented in this research.
• Do not assume a resume failure and a context-window hang are the same problem; they have different strings and different fixes.

How to tell it worked

The conversation reopens and you can send a message. For the CRC/store shape, the session opens after you move the corrupted store aside and restore from backup if needed.

If it is still failing

If codex resume still cannot open the session and the store is intact, the cause is likely the pagination or provider shape, which this guide cannot rank. Include the client version, OS, the exact banner, and whether an update preceded it.

Known limits of this guidance

• Evidence level: reported, and thinner than for codex-session-stuck. It rests on a handful of issue reports: one closed as completed with no stated fix, one open store-level report, and two title-only leads whose bodies were not opened.
• The guide cannot rank the three causes; it only describes their shapes.
• No release note was read and nothing here establishes a fix. codex resume’s flags were not retrieved, so only the bare command is documented.
• The failed to read thread … CRC (os error 23) line is the reporter’s own elision; the string must not be published whole.