Components of an LLM agent's memory

Name the common components of an LLM agent's memory (for example the context window, a short-term scratchpad or conversation buffer, long-term stores of facts and past episodes, and learned procedures or skills). For each one, explain what it holds, how information gets written into it, how it is read back at the right moment, and how the components work together during a task.

A good answer also covers what to forget or summarize, how retrieval is triggered, and how stale or wrong memories are corrected.