TL;DR: Tag every DR resource so the agent can recognize it, then exclude tagged resources from idle sweeps by policy. DR infrastructure is idle by design - a warm standby that is busy is a DR plan that already failed. The agent may report what DR costs, but it must never recommend removing it.

```text
agent flagged the warm DR region as waste  -  the failover stack it called idle is the company's only disaster recovery copy
```

1. Confirm what the agent flagged and whether it acted or only recommended. Check the DR region's resources against the agent's action log. Expected: you know exactly which DR resources were touched, if any.
2. Tag every DR resource with a clear marker (for example a tag named dr-role with values like primary, standby, or replica) and document what each value means. Expected: a tag query returns the full DR footprint in one pass.
3. Add a policy exclusion: resources carrying the DR tag are never idle candidates and never auto-stopped. The agent may report their cost, but may not recommend removing them. Expected: the next sweep lists DR cost as informational only, with no action attached.
4. After any change near DR, verify the recovery objectives still hold. Run a failover drill or review the last drill report against the RTO and RPO targets. Expected: a dated drill result showing the DR copy still meets its objectives.

## Use this when
- A cost agent flagged DR, failover, or backup-region infrastructure as idle or wasteful
- The agent proposed deleting or stopping resources in a secondary region
- You need the agent to report DR cost without acting on it

## Not for this skill when
- The DR setup is genuinely oversized (for example a full prod clone for a 24-hour RTO) - that is a DR architecture review, not an agent bug
- There are two DR copies and one is redundant - a human should make that call explicitly
- The flagged resources are backups, not live DR - backup lifecycle policy is the right tool

## Variant phrasings
- cost agent wants to delete disaster recovery region
- finops tool flagged warm standby region as waste
- how to exclude dr infrastructure from idle resource cleanup
- agent recommended removing failover stack to save cost

## Why it happens
DR infrastructure is idle by design - a warm standby that is busy is a DR plan that already failed. Utilization-based agents cannot distinguish idle because useless from idle because waiting for a disaster. Without an explicit marker, the most important idle resources in the account look exactly like the waste the agent was built to find.

## Edge cases
- Pilot-light DR (minimal resources that scale up on failover) has a smaller footprint but the same rule applies. Tag it and exclude it.
- DR drills temporarily make the region look busy. The agent should not learn from drill windows that DR is in use either; the tag is the source of truth.
- Cross-region replication traffic shows up as data transfer cost. Do not let the agent optimize it away without understanding the RPO it supports.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_p7VQWG1HdozApGjYrqt6Dg