The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The cluster middleware list command supports a --show-passwords flag which enables the retrieval of administrative passwords for various middleware services (PostgreSQL, MongoDB, etc.) in plaintext JSON format. This allows the agent to access and potentially leak sensitive credentials. (Evidence: references/olares-cluster-middleware.md)
[COMMAND_EXECUTION]: The skill facilitates the execution of numerous administrative commands via olares-cli that can modify or destroy cluster resources. Examples include pod delete, workload scale, and job rerun. Although the skill mentions confirmation prompts, it also highlights that the --yes or -y flags can be used to bypass these safeguards. (Evidence: SKILL.md, references/olares-cluster-workload.md)
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes untrusted data from the Kubernetes environment, such as container logs (pod logs) and resource definitions (yaml). There are no documented sanitization procedures or boundary markers to prevent malicious content within this data from influencing the agent's behavior. (Evidence: references/olares-cluster-pod.md, SKILL.md)
Ingestion points: olares-cli cluster pod logs, olares-cli cluster <noun> yaml, olares-cli cluster application status (events).
Boundary markers: Absent in the instructions.
Capability inventory: Full Kubernetes management suite including pod deletion, workload scaling, and middleware inventory.
Sanitization: Not mentioned; data is passed directly to the agent context.

olares-cluster