prompt-optimizer-flowlisted

# Prompt Optimizer — Flow Mode ## CHANGELOG **v3 — Jun 2026.** Recalibration for Claude Opus 4.8 (released May 28 2026), mirroring the rigorous counterpart, plus a single optional effort line. Only the changes that pass flow's mission-fit test (speed, intent-preservation, no diagnostic theater) are taken; the rigorous counterpart's heavier run-settings machinery and its research-toggle recommendation are deliberately NOT imported — both break flow. No structural changes; the 5-step workflow, the two hard rules, and the lean output format are unchanged. No lockstep coordination with prompt-architect required. - **4.7 calibration note recalibrated to 4.7–4.8 (Step 4).** 4.8's behavior changes (more reliable reasoning-effort calibration, better tool triggering, improved honesty, more consistent instruction-following) narrow where techniques earn their place. Reasoning activation now splits "more thinking" (native on 4.8 via adaptive thinking + the effort dial) from "specific reasoning structure" (still the prompt's job — favor structure over volume). Anti-sycophancy held steady, NOT softened: 4.8's honesty gains are on a different axis than user-framing pushback. Research activation narrowed on better tool triggering. - **Optional effort line (Output format).** For analytical or strategic keystone prompts, the Inferred note may carry one trailing clause recommending the effort dial (high, or xhigh for the heaviest work). Thinking is folded into effort — on 4.8 thinking depth