## TL;DR

Long-running agents outlive short-lived age keys, so decryption fails halfway. Issue keys with lifetimes covering the run, refresh them before expiry, and make the agent re-authenticate instead of dying on expiry.

## Error

```text
agent failed to decrypt the sops file: age key expired mid-run
```

## Steps

1. Confirm the failure is expiry: the key worked at run start and fails later. Expected: expiry, not a wrong key.
2. Size key lifetimes to the longest expected run plus margin. Expected: keys outlive runs.
3. Add a refresh step before the known expiry point in long runs. Expected: no mid-run expiry.
4. On expiry errors, re-issue the key and retry the step rather than aborting. Expected: the run continues.
5. Log key issuance and expiry so the next failure is diagnosable. Expected: clear audit trail.

## When to use

- Agent runs failing on age key expiry mid-run.
- Designing long-lived automation with sops.

## When not to use

- Wrong-key errors (fix the key, not the lifetime).
- Short runs (expiry is not the cause).

## Tool compatibility

- Sops with age; agent orchestration.

## Variant phrasings

### age key expired during run

Refresh before expiry.

### sops decrypt failed halfway

Same lifetime fix.

## Why it happens

Ephemeral keys are a security feature; agents that run longer than the key lifetime hit the designed expiry.

## Edge cases

- Refresh needs the issuing authority reachable; cache a fallback.
- Concurrent steps sharing one key race on refresh; coordinate.
- Expiry during a write can corrupt output; make writes atomic.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_PmwURof7v57Wyq6S48nfHw