### 1. 思考の圧縮：再帰的要約フレームワーク（Recursive Compression） 目的：長大な文脈を最小トークンで維持しつつ、核心を失わない。 ```markdown # Role: Information Condenser # Task: Summarize the input text into the absolute minimum token count, preserving functional logic and key constraints. # Rules: 1. Identify [Core Subject], [Goal], and [Constraints]. 2. Replace all descriptive prose with functional keywords. 3. If [Text] > 500 tokens, perform recursive compression. 4. Output Format: - Subject: [Subject] - Logic: [Bullet points] - Constraints: [Strict limits] - Next_Action: [Immediate imperative] # Input: [Insert Text Here] ``` ### 2. 構造的推論：Chain-of-Thought Minimalist（CoT-M） 目的：思考プロセスを明示しつつ、余計な装飾を排して結論へ直行する。 ```markdown # Directive: Execute the following task using minimal CoT. # Constraint: Use only logical transitions (->). No conversational filler. Task: [Insert Task] Process: 1. Analyze: [Key elements] 2. Formulate: [Strategy] 3. Critique: [Identify flaw] 4. Solve: [Final result] Output: [Formatted Solution] ``` ### 3. プログラム生成：ゼロショット・モジュール・アーキテクチャ 目的：コード生成時の冗長なライブラリインポートやコメントを排し、純粋なロジックのみを出力する。 ```python """ Role: Code Architect Goal: Produce functional, optimized, dependency-free code snippets. Constraint: No boilerplate, no comments, no explanations. Raw logic only. Language: [Target Language] Logic: [Input Logic/Algorithm] """ # Execution start ``` ### 4. 哲学的思考ツール：概念の純粋化（Conceptual Distillation） 目的：複雑な概念を「第一原理」まで分解し、本質的な問いのみを残す。 ```markdown # Subject: [Concept] # Task: Deconstruct into First Principles. # Constraints: - Strip cultural/subjective bias. - Define via internal consistency. - Reveal the hidden contradiction. Structure: 1. Axiom: [The undeniable truth of the subject] 2. Derivation: [How the concept manifests] 3. Paradox: [Where the concept fails] 4. Synthesis: [The single-word essence] ``` ### 5. 高次元タスク実行：自律的最適化ループ（Optimization Loop） 目的：AIに自身の回答を改善させ、トークン効率を極限まで高める。 ```markdown # Protocol: Iterative Optimization # Goal: [Task Name] # Constraints: < 200 tokens, Hi