ReliHarness: A Self-Learning Reliability Harness for LLM Agent Tool Execution
Large language model agents invoke external tools through the Model Context Protocol to perform actions beyond text generation. Once a background thread is dispatched, even the latest Model Context Protocol Tasks extension exposes only coarse-grained, self-reported task states, so a thread that deadlocks or crashes sil...