diff --git a/.forgejo/workflows/deploy.yaml b/.forgejo/workflows/deploy.yaml
index c4efb6d..a001ae7 100644
--- a/.forgejo/workflows/deploy.yaml
+++ b/.forgejo/workflows/deploy.yaml
@@ -9,9 +9,12 @@ env:
   REGISTRY: pipe.gohorsejobs.com
   IMAGE_NAMESPACE: bohessefm
   DOCKER_HOST: unix:///run/docker.sock
+  # ATIVA O BUILDKIT: Necessário para usar --mount (cache) no Dockerfile
+  DOCKER_BUILDKIT: 1
 
 jobs:
   build-and-push:
+    # Ajustado para usar as labels que aparecem no seu Runner ID 50
     runs-on: [self-hosted, linux-amd64]
     defaults:
       run:
@@ -28,17 +31,16 @@ jobs:
 
       - name: Build & Push Backend
         run: |
-          # Usamos --quiet no push para não inundar o Forgejo com logs de progresso (evita Erro 500)
+          # Build usando BuildKit (suporta o cache do seu Dockerfile)
           docker build -t ${{ env.REGISTRY }}/${{ env.IMAGE_NAMESPACE }}/gohorsejobs:${{ github.sha }} \
             -t ${{ env.REGISTRY }}/${{ env.IMAGE_NAMESPACE }}/gohorsejobs:latest ./backend
           
+          # Push --quiet evita o flood de logs que derruba o Forgejo
           docker push --quiet ${{ env.REGISTRY }}/${{ env.IMAGE_NAMESPACE }}/gohorsejobs:${{ github.sha }}
           docker push --quiet ${{ env.REGISTRY }}/${{ env.IMAGE_NAMESPACE }}/gohorsejobs:latest
 
       - name: Build & Push Backoffice
         run: |
-          # O build do Backoffice (NestJS) é o que mais trava. 
-          # Reduzimos o log e focamos no push silencioso.
           docker build -t ${{ env.REGISTRY }}/${{ env.IMAGE_NAMESPACE }}/backoffice:${{ github.sha }} \
             -t ${{ env.REGISTRY }}/${{ env.IMAGE_NAMESPACE }}/backoffice:latest ./backoffice
           
@@ -50,7 +52,6 @@ jobs:
     runs-on: [self-hosted, linux-amd64]
     steps:
       - name: Install Deploy Tools
-        # Removido nodejs daqui (não parece necessário para rodar kubectl)
         run: apk add --no-cache curl kubectl
 
       - name: Checkout code
@@ -61,12 +62,10 @@ jobs:
           kubectl create namespace gohorsejobsdev --dry-run=client -o yaml | kubectl apply -f -
           kubectl apply -f k8s/dev/ -n gohorsejobsdev
 
-          # Rollout progressivo: garante que o K8s não derrube tudo de uma vez
           kubectl -n gohorsejobsdev set image deployment/gohorse-backend-dev \
             backend=${{ env.REGISTRY }}/${{ env.IMAGE_NAMESPACE }}/gohorsejobs:${{ github.sha }}
           
           kubectl -n gohorsejobsdev set image deployment/gohorse-backoffice-dev \
             backoffice=${{ env.REGISTRY }}/${{ env.IMAGE_NAMESPACE }}/backoffice:${{ github.sha }}
           
-          # Verifica se o deploy deu certo (se travar aqui, o problema é no K3s, não no Forgejo)
-          kubectl -n gohorsejobsdev rollout status deployment/gohorse-backend-dev --timeout=90s
\ No newline at end of file
+          kubectl -n gohorsejobsdev rollout status deployment/gohorse-backend-dev --timeout=120s
\ No newline at end of file