The Agent Skills Directory

[COMMAND_EXECUTION]: The management script scripts/provider-ctl.py uses subprocess.run to execute tailscale ssh commands, which in turn run pct exec and Docker commands on a remote host. While it uses shlex.quote for argument safety, it provides a path for the agent to execute administrative operations on infrastructure.
[PRIVILEGE_ESCALATION]: The scripts/provider-ctl.py script defaults to connecting via SSH as the root user (root@foxtrot), providing high-privilege access to the target host as the default configuration.
[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to process and apply machine classes and cluster templates. This creates an attack surface where malicious instructions embedded in these YAML templates could influence agent behavior.
Ingestion points: YAML definitions for MachineClass and Cluster resources processed by the agent.
Boundary markers: None identified in the provided templates to delimit untrusted data from instructions.
Capability inventory: Execution of management scripts (provider-ctl.py), cluster management (omnictl), and Kubernetes operations (kubectl, helm).
Sanitization: The Python management script employs shlex.quote to sanitize arguments before shell execution.

omni-talos