fix(#56): 补齐 k8s 部署产物 —— 编排器镜像装 kubectl、RBAC 加 secrets、Deployment 接拉起 env + workload-identity label

让 k8s 拉起链路在仓内"可部署即可用"(此前线上实测全缺):

- Dockerfile.orchestrator:装 `kubectl`(pin AKS 1.34;linux/amd64)。launcher 靠 shell
  `kubectl apply/delete` 建/删 agent Pod + 每-swarm key Secret,镜像无 kubectl 则 k8s 后端必失败。
- k8s/rbac/orchestrator-role.yaml:加 `secrets`(create/delete/list/get)——否则建不了模型 key
  Secret、agent keyless。并注明 Role 命名空间须与 AGENT_POD_NAMESPACE 一致。
- k8s/orchestrator-deployment.yaml:
  - pod 模板加 `azure.workload.identity/use: "true"`(AKS webhook 注入 token,配合已注解的
    SA + UAMI 读 heicode-vault)。
  - 接入拉起 env:`AGENT_LAUNCH_BACKEND=kubernetes`、`AGENT_POD_IMAGE=heicode.azurecr.io/swarm-agent:latest`、
    `AGENT_POD_NAMESPACE=swarm-system`(同 RBAC ns)、`ORCHESTRATOR_PUBLIC_URL=ws://orchestrator-service...:8000`、
    `AGENT_OPENAI_API_BASE=https://code.heicode.cc/v1`、`SECRET_RESOLVER=azkv`。
  - image 指向 ACR(`heicode.azurecr.io/swarm-orchestrator`,tag 部署时 pin)。

校验:两个 manifest YAML 解析通过(label/env/secrets 均在);test-agent-launcher / test-swarm-guard /
test-contract-freeze 全绿。无明文密钥。

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Dockerfile.orchestrator

@@ -1,5 +1,16 @@
 FROM python:3.11-slim
 
+# kubectl — the kubernetes launch backend (agent_swarm#16/#56) shells out to `kubectl apply/delete`
+# to create/teardown agent Pods + per-swarm key Secrets. Without it the k8s backend fails (0 agents).
+# Pinned to the cluster minor (AKS 1.34) per kubectl skew policy. (linux/amd64 — AKS default node arch.)
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends curl ca-certificates \
+    && KUBECTL_VERSION="$(curl -fsSL https://dl.k8s.io/release/stable-1.34.txt)" \
+    && curl -fsSL "https://dl.k8s.io/release/${KUBECTL_VERSION}/bin/linux/amd64/kubectl" -o /usr/local/bin/kubectl \
+    && chmod +x /usr/local/bin/kubectl \
+    && kubectl version --client=true 2>/dev/null \
+    && apt-get clean && rm -rf /var/lib/apt/lists/*
+
 WORKDIR /app
 
 # Install dependencies

k8s/orchestrator-deployment.yaml

@@ -30,11 +30,15 @@ spec:
     metadata:
       labels:
         app: orchestrator
+        # Required for the AKS workload-identity webhook to inject AZURE_FEDERATED_TOKEN_FILE etc.
+        # into the pod, so the orchestrator can read the per-user model key from heicode-vault via
+        # its UAMI (SA swarm-orchestrator is annotated with azure.workload.identity/client-id). #56
+        azure.workload.identity/use: "true"
     spec:
       serviceAccountName: swarm-orchestrator
       containers:
         - name: orchestrator
-          image: swarm-orchestrator:latest
+          image: heicode.azurecr.io/swarm-orchestrator:latest  # deploy pins the tag (e.g. :<gitsha>)
           imagePullPolicy: IfNotPresent
           ports:
             - containerPort: 8000
@@ -66,6 +70,26 @@ spec:
             # (also set REDIS_CLUSTER=1 for an OSSCluster endpoint).
             - name: LOG_LEVEL
               value: "INFO"
+            # ── Swarm agent launch (agent_swarm#16/#56): make the runtime actually spawn agents ──
+            # Without AGENT_LAUNCH_BACKEND the launcher is a no-op (0 agents → run hangs).
+            - name: AGENT_LAUNCH_BACKEND
+              value: "kubernetes"
+            - name: AGENT_POD_IMAGE
+              value: "heicode.azurecr.io/swarm-agent:latest"   # build: az acr build -r heicode -t swarm-agent:latest -f Dockerfile.agent .
+            # Agent Pods + per-swarm key Secrets are created here — MUST match the RBAC Role's ns
+            # (k8s/rbac/orchestrator-role.yaml is in swarm-system).
+            - name: AGENT_POD_NAMESPACE
+              value: "swarm-system"
+            # In-cluster Service DNS the launched agents connect back to (Service: orchestrator-service).
+            - name: ORCHESTRATOR_PUBLIC_URL
+              value: "ws://orchestrator-service.swarm-system.svc.cluster.local:8000"
+            # HM model gateway (OpenAI-compatible). code.heicode.cc is the official address (#56).
+            - name: AGENT_OPENAI_API_BASE
+              value: "https://code.heicode.cc/v1"
+            # Resolve per-user model key from heicode-vault via the pod's workload identity (#56).
+            # (Redundant when AZURE_FEDERATED_TOKEN_FILE is injected, but explicit is clearer.)
+            - name: SECRET_RESOLVER
+              value: "azkv"
           resources:
             requests:
               memory: "256Mi"

k8s/rbac/orchestrator-role.yaml

@@ -6,12 +6,23 @@ metadata:
   labels:
     app: heicode-swarm
     component: orchestrator
+# NOTE: a Role is namespaced — these permissions only apply in `swarm-system`. The k8s launch
+# backend creates agent Pods + per-swarm key Secrets in AGENT_POD_NAMESPACE, so that env MUST equal
+# this namespace (set AGENT_POD_NAMESPACE=swarm-system), or a matching Role+RoleBinding must exist in
+# the agent-pod namespace. (agent_swarm#56)
 rules:
   # Pod management permissions
   - apiGroups: [""]
     resources: ["pods"]
     verbs: ["create", "delete", "list", "watch", "get"]
 
+  # Per-swarm model-key Secret: the launcher applies a Secret (build_secret_manifest) referenced by
+  # the agent Pod via secretKeyRef, and deletes it on teardown (`kubectl delete pod,secret`).
+  # Without `secrets` create/delete the key Secret can't be made → agents start keyless. (#56)
+  - apiGroups: [""]
+    resources: ["secrets"]
+    verbs: ["create", "delete", "list", "get"]
+
   # Pod log access
   - apiGroups: [""]
     resources: ["pods/log"]