agentscope-ai / agentscope-ai/agentscope-java
[Feature]: Pod securityContext / hardening customizer for Fabric8KubernetesPodRuntime
- 主要語言
- Java
- 星號
- 5.6k
- 分支
- 1.3k
- 平均合併
- 4 天 12 小時
- 30 天內合併 PR
- 77
描述
### Summary
`Fabric8KubernetesPodRuntime.createPod()` provides no way to set a pod/container `securityContext`. For untrusted code-execution sandboxes this blocks standard pod hardening.
### Details
`createPod()` builds `PodSpecBuilder` + `ContainerBuilder` with restartPolicy, image, command/args, resource requests, bind-mounts, nodeSelector and serviceAccount — but there is no hook, `Consumer`, or option to set any of:
- `runAsUser` / `runAsNonRoot` / `fsGroup`
- `readOnlyRootFilesystem`
- `capabilities.drop` (e.g. `ALL`)
- `allowPrivilegeEscalation`
- `seccompProfile` (e.g. `RuntimeDefault`)
- `automountServiceAccountToken: false`
`KubernetesSandboxClientOptions` has no security-context fields either (namespace, image, containerName, workspaceRoot, serviceAccount, nodeSelector, podLabels, cpu/memoryRequest). #1713 (env injection) adds an env map at the same insertion points but does not touch `securityContext`.
### Request
Add a customizer so callers can inject a hardened security posture, e.g.:
- `Consumer` + `Consumer` hooks on the runtime, **or**
- explicit security fields on `KubernetesSandboxClientOptions` (runAsNonRoot, runAsUser, readOnlyRootFilesystem, dropCapabilities, seccompProfile, automountServiceAccountToken, ...).
Same `createPod()` insertion points as #1713. Happy to contribute the PR if the API shape is agreed.
貢獻指南
評估
這個 Issue 還沒有評估資料。