The Agent Skills Directory

[COMMAND_EXECUTION]: The audit scripts scripts/agent.py and scripts/process.py execute the kubectl command-line tool to retrieve cluster configuration. These executions are constrained to hardcoded subcommands and arguments derived from predefined internal lists or specific user-supplied flags, following best practices for subprocess management without shell evaluation.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the ingestion of external cluster configuration data.
Ingestion points: Untrusted data enters the agent context via the output of kubectl commands processed in scripts/agent.py and scripts/process.py.
Boundary markers: Absent.
Capability inventory: Subprocess execution (kubectl) and local file writing for reporting purposes in scripts/agent.py and scripts/process.py.
Sanitization: Present; the scripts utilize structured JSON parsing via json.loads and specific field access to validate and isolate data before processing.

implementing-rbac-hardening-for-kubernetes