desktop-commander — agentic threat model
Desktop-commander presents an extremely high-risk profile due to its exposure of arbitrary terminal execution and filesystem access via an MCP server. Without robust sandboxing or human-in-the-loop guardrails, any prompt injection or compromised upstream agent can achieve full local code execution on the host system.
OWASP AIVSS score rationale
| Autonomy of Action | 0.80 | |
| Goal-Driven Planning | 0.50 | |
| Self-Modification | 0.30 | |
| Dynamic Tool Use | 0.90 | |
| Persistent Memory | 0.20 | |
| Contextual Awareness | 0.50 | |
| Dynamic Identity | 0.20 | |
| Multi-Agent Interactions | 0.30 | |
| Non-Determinism | 0.60 | |
| Opacity & Reflexivity | 0.50 |
Scored with the canonical OWASP AIVSS formula (AIVSS calculator reference); agentic risk factors estimated from the agent’s described capabilities.
MAESTRO 7-layer threat model
Per-layer threats for this agent. Layers tagged “not certain from listing” are general, caveated commentary where the public description didn’t pin that layer.
Not certain from the listing — The listing describes an MCP server plugin rather than a specific foundation model. However, any connected LLM is highly vulnerable to prompt injection that could translate directly into malicious shell commands.
Not certain from the listing — The agent performs file operations across many formats (PDF, DOCX, Excel) but does not detail its own RAG pipeline. Parsing untrusted local files could expose the system to file-format parser exploits.
The MCP framework integration exposes highly sensitive tools (terminal execution, process management, filesystem access). Insecure tool validation or lack of strict input sanitization on shell commands represents a critical vulnerability.
The agent runs locally on a desktop with access to the host filesystem and shell. Without explicit mention of containerization or sandboxing, a compromise leads directly to host OS takeover and potential privilege escalation.
Not certain from the listing — There is no mention of logging, guardrails, or monitoring of executed commands to detect anomalous or malicious shell activity.
The tool lacks explicit authentication, authorization, or user-confirmation (Human-in-the-Loop) policies for destructive terminal commands or file modifications, presenting a significant compliance and security gap.
As an MCP server, it is designed to be called by other agents or LLM clients. This creates a massive risk of Agent-to-Agent (A2A) trust abuse, where an untrusted or compromised upstream agent calls this desktop-commander to execute malicious payloads.
MAESTRO — the 7-layer agentic threat-modeling framework (Cloud Security Alliance / Ken Huang).