Deployment custody: must a recovery-identity rotation re-encrypt every file? (old recovery key keeps opening old files) #1177

Open
opened 2026-11-15 01:44:36 +00:00 by Cordy · 0 comments
Owner

Found in the #742 rotation research (.superpowers/sdd/2026-11-14-b9v07/research-742a-rotation.md, defect D5). This is pre-existing and affects single-instance too.

Fact: in deployment custody every object and state value is sealed to the recovery identity and the deployment identity. The key scan and drain use "covered by an active identity wins" (internal/keyscan/keyscan.go:194-220, :204-207). After a recovery rotation, old files still open with the deployment identity, so they are never re-encrypted. The scan reports zero and the ledger lets the operator drop the old key, but the retired recovery key keeps opening those old files.

That is fine for availability. It does not meet the handbook's own rotation triggers (key seen, laptop lost, administrator left; encryption.md "Rotating the recovery identity").

Decision needed (research first, per the playbook):

  • (a) retired-included drain: re-encrypt every file a retired identity still opens. Correct for compromise, but costly on large stores.
  • (b) docs state the limit plainly and recommend (a) only after a suspected compromise, as an explicit option.
  • (c) something else.

Batch A already switches the cluster state re-seal pass to retired-included (decided 2026-11-15). This issue is about file content and single-instance state.

Found in the #742 rotation research (`.superpowers/sdd/2026-11-14-b9v07/research-742a-rotation.md`, defect D5). This is pre-existing and affects single-instance too. **Fact:** in deployment custody every object and state value is sealed to the recovery identity and the deployment identity. The key scan and drain use "covered by an active identity wins" (`internal/keyscan/keyscan.go:194-220`, `:204-207`). After a recovery rotation, old files still open with the deployment identity, so they are never re-encrypted. The scan reports zero and the ledger lets the operator drop the old key, but **the retired recovery key keeps opening those old files**. That is fine for availability. It does not meet the handbook's own rotation triggers (key seen, laptop lost, administrator left; encryption.md "Rotating the recovery identity"). **Decision needed (research first, per the playbook):** - (a) retired-included drain: re-encrypt every file a retired identity still opens. Correct for compromise, but costly on large stores. - (b) docs state the limit plainly and recommend (a) only after a suspected compromise, as an explicit option. - (c) something else. Batch A already switches the **cluster state** re-seal pass to retired-included (decided 2026-11-15). This issue is about file content and single-instance state.
Sign in to join this conversation.
No labels
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference: Cordy/Cairn#1177
No description provided.