RubyCop
RubyCop scans Ruby code and tells you whether it's safe or not. We use it at CodeSchool to check user-submitted code before we eval it on the server.
RubyCop is made up of two pieces: NodeBuilder and Policy. NodeBuilder is responsible for parsing Ruby code (using Ripper) and building an AST (Abstract Syntax Tree). Policy then scans the AST and tells you whether the code is safe or not.
Requirements
RubyCop requires Ruby 1.9, though it should work under 1.8 if you include the "ripper" gem.
Usage
Here's a quick example of building the AST, and evaluating it with the Policy:
>> require "ruby_cop"
=> true
>> policy = RubyCop::Policy.new
>> ast = RubyCop::NodeBuilder.build("x = 1 + 2")
>> ast.accept(policy)
=> true
And if you pass in some unsafe code:
>> ast = RubyCop::NodeBuilder.build("x = `ls -la`")
>> ast.accept(policy)
=> false
Drawbacks
Ruby is a very dynamic language, so this kind of static analysis will only get you so far. RubyCop blocks obvious things like backticks and unsafe Kernel methods (#eval, #exec, #fork, etc.), but has to err on the side of safety in other places. For instance, #send is considered unsafe.