Imagine trying to run a code snippet you found online, only to discover it's missing crucial parts, making it impossible to execute. This is a common problem with shared code examples, as well as when working with large codebases where dependencies can be intricate and difficult to reproduce. Now, there's a new solution that leverages the power of Large Language Models (LLMs) to tackle this challenge: SelfPiCo (Self-Guided Partial Code Executor). SelfPiCo dynamically fills in missing code elements to make partial code runnable. Unlike previous approaches that relied on pre-defined dummy values, SelfPiCo utilizes LLMs to understand the context of the code and predict appropriate missing values in an interactive loop. If an injected value causes an error during execution, SelfPiCo refines its prediction based on the error message, acting like a skilled developer debugging their code. This iterative process, known as "LLM-in-the-loop," helps SelfPiCo learn and improve its predictions. A supplemental module, the complementary type predictor, handles scenarios where the value prediction process stalls. This module predicts the type of missing elements and injects standard dummy values, ensuring continued execution. In tests using open-source project functions and Stack Overflow code snippets, SelfPiCo significantly outperformed existing methods, demonstrating its potential for dynamic analysis tasks like runtime type error detection. By making snippets of code executable, SelfPiCo has promising applications in areas like automated software testing and code snippet comprehension. The framework successfully detects 18 type errors from GitHub projects and 33 type errors from Stack Overflow posts, highlighting its real-world impact. While SelfPiCo excels in many situations, it faces challenges when dealing with complex code requiring deep contextual understanding. Further development could refine SelfPiCo's handling of these intricate scenarios and unlock its full potential in the world of code analysis and execution.