The Agent Skills Directory

[COMMAND_EXECUTION]: The inspect_agent_sandbox.py script utilizes subprocess.run to interact with kubectl. The script is designed with security in mind, implementing a 'redacting inspector' that specifically filters out sensitive metadata (labels and annotations) and avoids reading Kubernetes Secret resources to prevent accidental data exposure in logs or evidence trails.
[REMOTE_CODE_EXECUTION]: The skill facilitates the execution of code within remote ACS Agent Sandboxes via the E2B SDK. As the core purpose is managing these sandboxes, the instructions include significant warnings to treat all sandbox workloads as untrusted and provide mechanisms to verify isolation and network boundaries.
[EXTERNAL_DOWNLOADS]: The skill documentation guides the user to install specific, pinned versions of the E2B SDK (e2b-code-interpreter==2.7.0 and e2b==2.24.0) from standard registries. It also mentions a private protocol requiring a custom patch package (kruise_agents.patch_e2b), which is consistent with the Alibaba Cloud OpenKruise project scope.
[CREDENTIALS_UNSAFE]: The skill demonstrates high security awareness regarding credentials. It mandates that API keys and cloud tokens be managed via environment variables or secret managers, explicitly forbidding their inclusion in logs, command history, or evidence files. It also provides specific guidance on avoiding kubectl commands that might inadvertently reveal secrets in container arguments.

alicloud-acs-agent-sandbox