service-template/LLM.md

LLM.md — next-agent handoff for hanzoai/template

This is the canonical skeleton every new Hanzo Go backend starts from. It is kept minimal on purpose: if a feature is not in every service, it does not belong here.

What the skeleton enforces

• config/config.go pulls env; nothing else. No flag soup. No YAML.
• cmd/appd/main.go calls claims.AssertGatewayUpstream() at boot. The daemon refuses to start without HANZO_GATEWAY_UPSTREAM=1.
• api/router.go is the one place middleware is wired. Top-to-bottom order: claims.Strip, claims.Inject, claims.RequireGateway for tenant routes. Public routes (/healthz, /readyz) mount separately, before the chain.
• store/store.go is the only caller of hanzoai/base/store. Handlers call app.Store.ForCtx(ctx) and get a *dbx.DB scoped to the caller's (org, user).
• models/example.go shows the orm.Typed[Example] pattern. Replace with your domain types — do not grow a domain-specific layer on top.

What NOT to add

• JWT middleware of any kind. The gateway is the sole verifier.
• A second cache library. luxfi/cache is it.
• nginx, caddy, or envoy configs. Ingress is hanzoai/ingress.
• Random SUMMARY.md / STATUS.md / NOTES.md. Only LLM.md.
• A Helm chart. Kustomize lives in hanzoai/platform.

Dependencies

Only three direct imports in go.mod:

• github.com/hanzoai/base — runtime + claims + store.
• github.com/hanzoai/orm — data layer.
• github.com/luxfi/cache — cache primitives.

Anything else is a review-blocking addition. Third-party libs go through hanzoai/base/tools/* or get their own module.