Security audit
Security checks for vulnerabilities and agentic risk
This is a coherent local FreeCAD automation skill, with a real but manageable caution around launching an unpinned CADPilot package through uvx.
Install this only if you want an agent to control a local FreeCAD session. Keep FreeCAD RPC bound to 127.0.0.1, verify that `uvx` and `cadpilot` resolve to sources you trust, and consider pinning a known-good CADPilot version before use.
The README instructs launching an MCP server via uvx cadpilot without pinning a specific version or artifact digest. This creates a supply-chain risk: future releases or a compromised package could change behavior and would be executed locally by users or agents with the skill's privileges.
Referencing uvx cadpilot again without version pinning repeats the same risk: the tool resolved at runtime may not be the one originally reviewed by the skill author. In an agent-tooling context, this is more sensitive because the MCP server mediates actions against a local FreeCAD instance and could be abused if a malicious update is pulled.
Although mentioned as an older setup, uvx freecad-mcp is still an unpinned executable package reference. Users copying legacy examples could fetch and execute an unexpected or compromised version, introducing the same supply-chain execution risk.
The usage instructions tell the agent to register uvx cadpilot if needed, again without pinning a version. This is especially relevant because it encourages on-demand execution in a local automation flow, increasing exposure to package-replacement or malicious-update attacks.
The skill depends on invoking uvx/cadpilot without pinning an exact package version or immutable source, which introduces a supply-chain risk: future installs or executions may resolve to a different package build than the author tested. Because this skill is specifically designed to bridge into local FreeCAD and then expose advertised CAD tools, an unexpected package update or malicious upstream release could execute unintended local code in a trusted desktop context.
No suspicious patterns detected.
