The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the send_direct_cli_command and apply_config_commands tools to execute shell-level commands on Lantronix SLC/EMG OOB devices. While this is the intended primary purpose for diagnostics and remediation, it represents a high-privilege interaction with live network infrastructure. The skill mitigates this risk by mandating explicit operator confirmation for all mutating or destructive commands.
[EXTERNAL_DOWNLOADS]: The documentation provides instructions to clone and install MCP server components from external GitHub repositories belonging to the hardware vendor (github.com/Lantronix/percepxion-mcp-server and github.com/Lantronix/slc-mcp-server). These are legitimate vendor-owned resources for the described management tasks.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data ingested from external network devices via get_cli_command_output, get_device_syslogs, and query_device_access_log.
Ingestion points: Device CLI output, system logs, and access logs retrieved via MCP tools in SKILL.md.
Boundary markers: None explicitly defined in the skill instructions for segregating tool output from agent instructions.
Capability inventory: The agent has the ability to write configurations (update_device_config), execute shell commands (send_direct_cli_command), and perform reboots or firmware updates.
Sanitization: The skill relies on platform-level (Percepxion) Role-Based Access Control (RBAC) and server-side CLI filtering rather than agent-side text sanitization.
[PRIVILEGE_ESCALATION]: The platform supports configuration via environment variables such as PERCEPXION_CLI_YOLO and SLC_CLI_YOLO which are designed to bypass built-in command filtering and safety denials. The skill explicitly warns against using these flags in production environments.

percepxion-oob