The Agent Skills Directory

Command Execution: The utility script scripts/debug_workload.py performs command execution to interact with Google Cloud services. It uses subprocess.run() to execute gcloud and kubectl commands. This is a core part of the skill's diagnostic functionality. The script correctly uses list-based arguments for subprocess.run, which avoids shell execution and reduces the risk of command injection. It also employs shlex.quote for safe string handling when printing commands for the user.\n- Indirect Prompt Injection Surface: The skill processes logs and events from external GKE environments, which could theoretically contain malicious instructions. This presents a potential surface where malicious content in logs could influence agent behavior. \n
Ingestion points: Cloud Logging data and Kubernetes events are retrieved and processed in scripts/debug_workload.py. \n
Boundary markers: The script formats log output into structured strings, providing implicit separation for the agent. \n
Capability inventory: The skill can execute gcloud and kubectl commands through the provided script and agent instructions. \n
Sanitization: Input parameters for log queries are quoted to ensure they are treated as data by the logging backend. This is a standard consideration for any tool processing external telemetry data.

pathways-workload