The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources, specifically Kubernetes cluster logs, resource descriptions (via describe), and command outputs. This data is potentially attacker-controlled if applications or services within the cluster are compromised, leading to a risk where the agent might follow instructions embedded in that data.
Ingestion points: Kubernetes resource outputs and logs retrieved via bun k8s-tool pods, bun k8s-tool logs, and bun k8s-tool describe in SKILL.md.
Boundary markers: There are no explicit delimiters or specific instructions for the agent to disregard natural language commands found within the retrieved Kubernetes data.
Capability inventory: The skill has the ability to execute shell commands via bun, interact with the Kubernetes API, and perform limited command execution within pods via the exec subcommand.
Sanitization: The skill documentation states that commands are parsed into arguments to avoid shell-controlled interpreters and that sensitive subcommands like cluster-info dump and secret reads are blocked.
[COMMAND_EXECUTION]: The skill is designed to facilitate the execution of commands on the host system using the bun runtime to interface with kubectl.
Evidence: Multiple examples in SKILL.md demonstrate executing bun k8s-tool with various flags, including a restricted exec command that allows running ls or redis-cli inside containers.
[EXTERNAL_DOWNLOADS]: The skill identifies a dependency on an external package and references a remote code repository.
Evidence: SKILL.md references the @blogic-cz/agent-tools dev dependency and links to its repository at https://github.com/blogic-cz/agent-tools for tool installation and updates.

k8s-tool