The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is entirely centered around executing the olares-cli binary to manage Kubernetes resources. While this allows for significant system mutation (scaling workloads, deleting pods, rerunning jobs), the instructions explicitly mandate human confirmation for all destructive actions, even when automation flags like --yes are available.
[REMOTE_CODE_EXECUTION]: The cluster exec functionality (documented in references/olares-cluster-exec.md) allows for running arbitrary commands inside containers via K8s WebSockets. The skill includes a 'Permission Gate' that restricts exec access based on the user's identity and namespace ownership (verified against the /capi/app/detail endpoint), and specifically instructs that interactive TTY modes are for humans only while agents should use bounded one-shot executions.
[DATA_EXPOSURE]: The cluster middleware reference (references/olares-cluster-middleware.md) highlights a security feature where administrative passwords for databases and other middleware are redacted by default in both table and JSON outputs. A specific flag, --show-passwords, is required for exposure, and the instructions warn agents against using this flag in contexts where output might be logged.
[PROMPT_INJECTION]: The skill implements clear boundary instructions, such as 'The identity-vs-server-decides principle,' which ensures the agent treats server-side 403 errors as authoritative and does not attempt to bypass security filters or broaden its scope without explicit user approval.

olares-cluster