Minecraft (fundamentallabs) MCP Server

In the server's implementation (`dist/skills/verified/buildSomething.js:161`): Evaluating a runtime value as code (rather than a fixed literal) executes whatever reaches it — a direct RCE primitive, and almost never necessary in legitimate code. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: const buildFunction = new Function(...Object.keys(context), ` // User code starts here ${code}

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/skills/verified/buildSomething.js

In the server's implementation (`src/skills/verified/buildSomething.ts:213`): Evaluating a runtime value as code (rather than a fixed literal) executes whatever reaches it — a direct RCE primitive, and almost never necessary in legitimate code. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: const buildFunction = new Function( ...Object.keys(context), ` // User code starts

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server src/skills/verified/buildSomething.ts

In the server's implementation (`dist/skills/verified/buildSomething.js:161`): Evaluating strings as code is the most direct RCE primitive; if any tool input reaches it, the server executes attacker-chosen code. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: const buildFunction = new Function(...Object.keys(context), ` // User code starts here ${code}

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/skills/verified/buildSomething.js

In the server's implementation (`src/skills/verified/buildSomething.ts:213`): Evaluating strings as code is the most direct RCE primitive; if any tool input reaches it, the server executes attacker-chosen code. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: const buildFunction = new Function( ...Object.keys(context), ` // User code starts

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server src/skills/verified/buildSomething.ts

In the server's implementation (`dist/skillRegistry.js:363`): Loading a module chosen at runtime (from a variable) can pull in and run attacker-influenced code paths. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: nst skillModule = await import(skillModuleUrl); console.error(`[MCP] Skill module loaded successfully`);

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/skillRegistry.js

In the server's implementation (`src/skillRegistry.ts:393`): Loading a module chosen at runtime (from a variable) can pull in and run attacker-influenced code paths. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: nst skillModule = await import(skillModuleUrl); console.error(`[MCP] Skill module loaded successfully`);

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server src/skillRegistry.ts