auditing-context-window-pressure

Claude Code session slowdowns and prompt-cache misses often stem from context-window pressure. As a 200k context window fills, token costs rise and response latency increases, a condition this audit diagnoses by checking cache hit rates and measured context usage.

Reduce context window bloat by applying triage steps from a session audit. The most effective approach prioritizes moving large tool results to files, re-dispatching subagents with summary-only outputs, and using /compact to reclaim performance without discarding useful context.

Prompt-cache drift is frequently caused by mid-session CLAUDE.md edits or dynamic content leaking into cached prompts. You can fix cache drift by diagnosing these hit rate drops and enforcing CLAUDE.md hygiene to stabilize the cached prompt prefix.

Use /compact to compress existing context into a smaller summary when you need to retain session continuity. Use /clear as a last resort only when the session is too degraded and you need a completely fresh context window for new agent work.

Yes, you can run a pre-flight session pressure audit before dispatching heavy subagent tasks. This checks if token-efficient mode is needed and estimates available space in the 200k context window to prevent mid-task cache misses and slowdowns.