Claude Code Field Guide
05 / 35

The Context Window

Claude's short-term memory for the current conversation — everything it can “see” right now, and it has a hard limit.

Still accurate
Currency check: Concept and “context rot” symptom are both still accurate; the tools for managing it have gotten more sophisticated (see Compacting Context).

Every message, every file Claude Code has read, and every response it has generated in the current session lives in its context window. It's finite, and when a conversation runs long or pulls in a lot of file content, older information starts getting crowded out. The video calls the resulting symptom “context rot”: Claude Code seems to forget earlier instructions or contradicts itself, similar in feel to hallucination in a regular chatbot.

The fix is still the same: keep sessions focused on one task, start a fresh session (/clear) when you switch topics, and use /compact to summarize and free up space mid-session. Most interfaces show a visual indicator of how full the context window is (a colored bar or percentage) so you can see this coming before quality drops.

Catching context rot before it bites

Mid-way through a long refactor, Claude Code starts re-suggesting an approach you already rejected. Recognizing this as context rot rather than a reasoning failure, you run /compact instead of arguing with it further.

One task, one session

Instead of using a single marathon session for the whole day's work, a user starts a new session per distinct task (backend fix, then frontend styling, then docs), keeping each context window focused and outputs sharper.