Amp's read_thread tool now handles massive threads by using a specialized subagent instead of a single LLM call, enabling it to search long threads, verify information across messages, and detect revisions or reversions. The subagent prioritizes correctness by checking for newer messages that contradict earlier findings and inspecting original messages when exact details matter, rather than relying solely on thread compactions. You can now reliably reference threads weeks old or millions of tokens long—just mention or paste the thread URL and ask your question, exactly as before.