New warning before resuming a long, idle conversation that would use much of the 5-hour limit

What

When you resume a conversation that has been inactive for a while and is large, Claude Code can now show a dialog before continuing. It reads "This conversation has been inactive for" a duration "and is" a number of "tokens long.", then "Resuming it will use about" a percentage "of your 5-hour usage limit.", or "all" when the estimate is 100% or more.

• The choices are "Resume" and "Start a new conversation". The dialog can also be dismissed. Starting a new conversation runs /clear.
• The estimate uses the conversation's context tokens, the model and your rate-limit tier, with per-tier fiveHourCredits and per-model creditsPerMillionTokens tables from the remote config tengu_amber_tally. Both tables need an entry for your tier and model.
• The built-in default of tengu_amber_tally is empty and does not pass validation, so the dialog cannot appear until the tables are supplied remotely. Nothing has been read about this config yet.
• It does not appear when the estimate is below minPercent (which falls back to 5), when the resumeReturnDismissed setting is set, or when the model is unchanged and the prompt cache still looks warm. There is one more failure check the members did not resolve.
• It runs after the existing stale-resume dialog that offers /compact, only when that dialog does not apply or returns "skipped_no_summary". That earlier step is now asynchronous.

Why

Resuming a large conversation after a long break sends its whole context again, which can take a large share of a subscriber's 5-hour usage limit without warning. This gives a chance to start fresh instead.