The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple powerful kubectl commands, including kubectl exec, which allows running arbitrary shell commands inside target containers (e.g., df, du, stat, cat, mount).
[COMMAND_EXECUTION]: It executes local shell scripts located at deployment/collect-logs.sh and {baseDir}/scripts/diagnose.sh to gather system-wide diagnostic data.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process potentially untrusted data from Kubernetes sources which can lead to the agent following instructions embedded in logs or events.
Ingestion points: Commands like kubectl logs and kubectl get events in SKILL.md bring external data into the agent's context.
Boundary markers: None identified; there are no specific instructions for the agent to ignore or delimit embedded commands within logs.
Capability inventory: The skill has broad capabilities including container-level execution (kubectl exec) and local script execution, which increases the impact if an injection occurs.
Sanitization: There is no evidence of log filtering or sanitization to prevent the agent from acting on instructions found in pod output.

debug-buttercup