Long silent stretches during a multi-batch investigation; the harness had to prompt for status three times

**What happened**
During the investigation the harness injected "The user hasn't heard from you in a while" three times. Between batches the model wrote one line and then ran batches that took minutes each, with no cost estimate or checkpoint offered.

**Expected**
Shorter batches, a running account of what was read and what it cost, and an early stop point for the user.

**What the user said**
"you destroyed my plans for this session"

**Repro**
Give a broad multi-repository investigation task; observe update frequency.

**Environment:** Claude Code CLI 2.1.270 (Claude Code), macOS, auto mode, model claude-fable-5-1, session of 2026-09-14.

Reproduce the reported behavior in Claude Code CLI 2.1.270 with a broad multi-repository investigation and observe the update frequency between batches. Compare the session against the requested shorter batches, running cost and progress account, and an early user stop point; done means the harness no longer needs to prompt for status during long investigations.

Written by the indexing model from the issue text.