// 20 tips tagged "context-window"
The /btw command lets you ask quick side questions that are answered immediately but never added to the conversation history.
Writing AGENTS.md as plain text in CLAUDE.md has no effect — use @AGENTS.md to actually load the file into context.
Rewind bad turns instead of correcting them, so Claude never processes the wrong output and you save tokens on every subsequent message.
Pass custom instructions to /compact so Claude preserves test output, code changes, and other context that matters to your workflow.
CLI tools like gh, aws, and gcloud don't add persistent tool definitions to your context, saving tokens compared to equivalent MCP servers.
Trigger automatic tool search earlier so idle MCP tool definitions stop consuming your context window and tokens.
Use --fork-session to branch a conversation into a new session, keeping the original intact while exploring alternatives.
Use /rename to give any session a meaningful name mid-conversation — it updates the terminal title immediately and makes the session easy to find in the resume picker later.
The --resume flag lets you pick up any past session by name or UUID, so you can defer work and return to it days later without re-explaining context.
Writing a plan.md before Claude starts building gives you a persistent checkpoint you can hand to a fresh session to continue exactly where you left off.
When running multiple Claude sessions in parallel, a shared plan.md acts as the source of truth — each agent reads it to understand what's in progress, what's done, and what to pick up next.
Don't wait until your context window is nearly full -- run /compact at around 60% while Claude still has full visibility over the session.