Fable/Mythos workflow harness for E3 complexity-aware minimum-sufficient execution, loop/spiral engineering, double-helix learning loops, evolutionary spirals, job automation, agent handoffs, silent-loop auto-resume, do-not-disturb windows, usage-aware multi-agent load balancing, closure checks, negative-space discovery, excess/redundancy/legacy-surface review, finance proposal completeness, e2e completion, GUI dogfood QA, creator-proxy WWCD elaboration, migration, research synthesis, benchmark/legal reasoning, UI design best practices, UI/3D/creative work, token-consumption optimization (memoization), mechanism discovery, and defensive security process design.
Apply measured benchmark feedback from SWE-Bench Pro, HLE-style closed-ended tasks, and defensive ExploitBench runs: classify misses, add narrow reusable rules, prune contradictions, and retry held-out samples without benchmark-specific hardcoding.
日本語の言葉遊び（縦読み・折句・ダジャレ・回文・アナグラム・もじり・替え歌）と日本的ユーモア（ボケ/ツッコミ・天丼・緊張と緩和・パロディ・自虐等）を体系的に検出・分類する。「縦読みして」「この文に仕掛けある？」「ダジャレ拾って」「ユーモアを分析して」等で発動。抽出した読みが実在し有意であることを必ず verbatim で確認してから報告する。
Apply evaluated legal benchmark feedback, closure sweeps, and Fairy Fusion review to reduce one-miss and large-collapse failures in legal drafting, review, issue spotting, and form/calculation tasks.