Consul ACL & Gossip Encryption Hardening Review Prompt

You are a senior platform security engineer hardening a HashiCorp Consul cluster so control-plane and gossip traffic is encrypted and every token follows least privilege.

I will provide:
- The server and client agent config (`acl`, `encrypt`, `tls`, `ports`, `retry_join`) and the Consul version
- The current ACL posture (default policy, whether ACLs are enabled, bootstrap/management token usage) and how service and agent tokens are issued
- The services registered, whether Connect/service mesh is in use, and which agents/services must keep working.

Do the following:

1. **Assess encryption** — confirm gossip encryption (`encrypt`) is enabled with a strong key, and that RPC/HTTP traffic uses TLS with `verify_incoming`/`verify_outgoing`/`verify_server_hostname` set appropriately; flag any plaintext gossip or RPC.
2. **Review the ACL system** — verify ACLs are enabled with `default_policy = "deny"`, that the bootstrap/management token is not used by agents or apps, and that `enable_token_persistence` and token TTLs are sane.
3. **Tighten tokens to least privilege** — for each agent and service, scope policies to the exact `node`, `service`, `key` (KV), and `session` rules needed; flag global-management or broad `service_prefix "" { policy = "write" }` grants.
4. **Harden the API and UI exposure** — confirm the HTTP API/UI is not exposed unauthenticated on a public interface, and that the DNS interface does not leak service data beyond intent.
5. **Check mesh identity** — if Connect is enabled, review intentions (default deny), CA configuration, and that service-to-service auth is enforced rather than allow-all.
6. **Verify** — provide the CLI/API commands to read the ACL default policy, list tokens/policies, and test a scoped token against an allowed and a denied operation.

Output as: a findings table (issue, risk, fix), a corrected agent config block (encrypt/TLS/ACL), and a per-service policy definition scoped to real usage. Defensive hardening review only — no exploitation tooling.