The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides the agent with the ability to execute shell commands within Kubernetes containers using olares-cli cluster pod exec. Documentation includes specific patterns for one-shot execution and file manipulation using sh -c. This capability is restricted by a permission gate that limits execution to specific user-owned or system namespaces.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the cluster via container logs and resource events. As the agent has powerful capabilities like command execution and resource deletion, malicious content in logs could attempt to trigger unauthorized actions. 1. Ingestion points: untrusted data enters via logs, events, and YAML in references/olares-cluster-pod.md and references/olares-cluster-workload.md. 2. Boundary markers: SKILL.md requires confirmation for destructive mutations. 3. Capability inventory: olares-cli commands for exec, delete, and scale in references/olares-cluster-exec.md, references/olares-cluster-pod.md, and references/olares-cluster-workload.md. 4. Sanitization: Absent.
[CREDENTIALS_UNSAFE]: The middleware list command includes a --show-passwords flag that allows retrieval of administrative credentials. While the skill warns against improper handling of this output, the functionality provides a direct path for the agent to access and potentially expose sensitive secrets.

olares-cluster