The ten-point recoverability check
- 01
You know the active CODEX_HOME
You have confirmed the native Windows path instead of assuming the default.
- 02
You have checked for separate SQLite state
Custom CODEX_SQLITE_HOME or sqlite_home settings are documented and included in the plan where required.
- 03
Your repositories have their own protection
Committed code is pushed and important uncommitted work has a deliberate backup path.
- 04
Credentials are not casually copied
auth.json and other token material are excluded or strongly encrypted with restricted access.
- 05
The destination survives loss of the PC
At least one recovery copy is physically or operationally separate from the Windows drive.
- 06
The backup is complete and tamper-evident
You can detect missing, changed, or partially written data before recovery.
- 07
Useful history is retained
An accidental deletion can be recovered from a point before the incident, not only from the latest state.
- 08
Recovery uses an empty safe folder first
The first recovery action does not overwrite the live Codex home.
- 09
You have tested the process
A recent backup has been reconstructed and inspected, not merely copied.
- 10
The scope is honest
Native Windows, WSL, cloud-only state, repositories, and other agents are not being silently treated as interchangeable.
How to interpret the result
- 8-10 checks: the plan is credible, but keep testing after important changes
- 5-7 checks: recoverability has meaningful gaps that should be closed
- 0-4 checks: a local failure could still remove important Codex continuity
Run the checklist after change
Repeat the review after moving PCs, changing CODEX_HOME, enabling a separate SQLite state location, changing backup destinations, or adding important skills and workflows. A backup plan can become stale even when the last copy succeeded.
What Calmbu currently claims
Calmbu shows the current supported scope, safer exclusions, encrypted protection, recovery status, and safe-folder recovery. Native Windows Codex is supported in the current test build; the public Windows download remains gated until release requirements pass.
See the recovery model How Calmbu verifies its claimsCommon questions
How often should I test Codex recovery?
Test after setup, after changing storage locations, and periodically after important Codex or Windows changes. A successful copy does not prove the recovery process still works.
Is OneDrive enough by itself?
A synced location can help with device loss, but sync is not the same as versioned, integrity-checked recovery. Deletions and corruption can also sync unless useful history is retained.
What is the most important check?
Prove a recent backup can be reconstructed in a separate folder. That test exposes more real gaps than a green copy-completed message.
Official references used for Codex facts
OpenAI documentation establishes Codex behaviour. Calmbu-specific statements describe the current product contract and are not presented as OpenAI guidance.