0.5

• Sharper styles. All three styles get four rules: deliverable purity (produce a
  message/email/commit and you get just that, no "here's a draft" wrapper), never trim a
  warning when compressing, keep every essential point (brevity is shorter points, not fewer),
  and the bold/TL;DR alone must carry the whole answer.
• Benchmark: honest and reproducible. Measures the work and the output separately; the
  headline claims use no LLM judge. Work is unchanged (hidden-test coding A/B, pass rates
  equal), output ~43% shorter on average (50-71% on verbose answers), and you reach the point
  in ~6 words vs ~40 (answer in the first line 75% vs 3%). Deliverable purity 88%. Full writeup
  and runnable harness under benchmarks/.