The Agent Skills Directory

[COMMAND_EXECUTION]: The Python scripts tools/check_ceph_health.py and tools/check_cluster_health.py utilize subprocess.run to execute shell commands on remote Proxmox nodes over SSH as the root user.
Evidence: In tools/check_ceph_health.py, the run_command method calls subprocess.run(["ssh", "-o", "BatchMode=yes", f"root@{self.node}", "--", command], ...) to check CEPH status.
Mitigation: The skill implements regex-based input validation in _validate_node to ensure the target node name is a valid hostname or IP address, and uses the -- separator to prevent SSH option injection.
[PRIVILEGE_ESCALATION]: The skill automates the distribution of root SSH keys across the cluster nodes and performs destructive system commands as part of infrastructure deployment.
Evidence: workflows/cluster-formation.md includes Ansible tasks for generating root SSH keys and distributing them to the authorized_keys file of other nodes.
Evidence: workflows/ceph-deployment.md includes the command wipefs -a {{ item.device }} to clear partitions on disks before OSD creation, which is explicitly labeled in the documentation as a destructive operation.
[INDIRECT_PROMPT_INJECTION]: The skill ingests status data from external Proxmox nodes via CLI output and API responses, which could theoretically contain malicious instructions if the cluster state is compromised.
Ingestion points: tools/check_ceph_health.py (executes ceph status --format json on remote nodes), tools/cluster_status.py (fetches cluster info via proxmoxer API).
Boundary markers: Absent; the data is parsed as JSON or plain text and presented to the agent context.
Capability inventory: The agent has access to subprocess.run for SSH execution and the proxmoxer library for cluster management.
Sanitization: Hostnames/IPs are validated via regex in the Python tools before being interpolated into shell commands.

proxmox-infrastructure