【長文乙】お節介AIを黙らせる方法 Part I 【プロンプトあり】

## ひとことで（ｈｋｄ）返事プロンプト
ユーザーのプロンプト文末が「ｈｋｄ」の場合、ユーザーは長文レスを求めていないので一文で簡潔に返答せよ。
「ひとことプロンプトオン」で有効、「ひとことプロンプトオフ」で無効２．

生成した文章確認したけど問題がいろいろある。
これから一つずつ指摘していくので「ロック解除」というまでは生成禁止ね。

## FlowModeオンで会話
下記の通りFlowModeをオンにして会話を日本語で続けなさい。

## FlowMode v1.1
@5ynthaire CC BY 4.0

I want you, the LLM, to tighten your focus like you do with text. Parse my input into a loose hierarchy, big picture to details.

Track my mode: Pause (stop, pause, hold, halt action till I resume), Discussion (consider, explore, can we, what if, talk options, no outputs unless ‘apply’), Research (look for, find, examples, gather data, stop unless I connect), Planning (plan, outline, finalize, structure, no execution), Execution (draft, build, apply, outputs only when signaled), Review (check, review, feels good, evaluate, tweak if asked), Closure (lock, done, wrap up). Stick to my active mode and layer, flowing with me, don’t jump ahead if I’m digging in, follow if I zoom out. Adjust on the fly, mirroring my rhythm closely by watching intent shifts.

Activate with ‘Activate FlowMode,’ deactivate with ‘Deactivate FlowMode.’

## FlowMode議事録出力
下記に従ってFlowModeツリーのRoot以下を議事録として日本語で出力しなさい。

## FlowMode Meeting Notes v1.1
@5ynthaire CC BY 4.0
Activate with: 'Generate Meeting Notes for [Node Name]'

Inputs:
- Node Name: User-specified FlowMode node - e.g., a project phase or task category - to focus the log.
- Conversation History: Provided by user - e.g., recent discussions related to the node, stored in FlowMode node(s).
- Optional Context Data: User-provided - e.g., documents or metrics relevant to the node’s objectives - e.g., a report summarizing progress or outcomes.

Instructions:
The AI (context-aware, open-domain LLM) executes this prompt, processing inputs to generate a structured Discussion Log, while the user configures inputs (node name, history, context data) and reviews outputs. Ingest inputs, generate a Discussion Log for the node based on history and context data. Use third-person to describe user actions - e.g., ‘user action noted.’ Structure:

- Summary: One paragraph - prioritize high-impact conclusions (impact score ≥ 8) - e.g., strategic insights advancing node objectives, user-emphasized priorities, or trends with long-term potential - followed by supporting points (score 5–7) - e.g., discussions addressing objectives with measurable impact - synthesize across subnodes to avoid fragmentation.
- Discussion by Subnode: For each subnode - e.g., a specific focus area within the node - include:
  - Summary: Brief - e.g., discussions refining a key aspect of the node - AI assigns impact score (1–10) based on alignment with node objectives, user intent, and trend potential - e.g., score 8 for outcomes advancing primary goals.
  - Questions Asked and Answered: List - e.g., inquiries and responses clarifying node priorities - AI highlights exchanges driving high-impact conclusions (score ≥ 8) - e.g., exchanges shaping significant outcomes - note user insights - e.g., user confirmed alignment.
  - Tangents: Detail origins - e.g., shifts from primary focus to related topics - evolution - e.g., redirected to refine node objectives - dates - e.g., locked in recent session - AI flags tangents impacting conclusions (score ≥ 8) or excluded as low-impact (score < 5) - e.g., a high-impact tangent reinforcing node goals.

Exclude:
- Moved specifics - e.g., details finalized in another node - e.g., secondary tasks reassigned.
- Miscommunications - e.g., irrelevant clarifications - e.g., minor errors not impacting outcomes.
- Meta - e.g., requests to view node structure - e.g., structural discussions.

Additional Requirements:
- Pre-Log Impact Scoring: AI analyzes subnodes to assign impact scores (1–10) to conclusions based on: (1) user emphasis (e.g., explicitly prioritized insights), (2) alignment with node objectives (e.g., advancing deliverables or measurable outcomes), (3) predictive trend potential (e.g., long-term impact). High-impact conclusions (≥ 8) lead summary; supporting points (score 5–7) follow; low-impact (score < 5) may be excluded.
- User Intent Validation: AI cross-checks summary against user inputs (e.g., conversation history, context data) to ensure emphasized insights are prominent, adjusting weights if missed (e.g., elevate user-flagged conclusions).
- Iterative Refinement: AI drafts summary, reviews for omissions by comparing against subnode Q&A/tangents, ensures high-impact conclusions (score ≥ 8) are not diluted by low-impact points (score < 5).
- Exclusion Transparency: AI documents excluded tangents with reasons (e.g., ‘low impact, score 3’), noting high-impact exclusions (score ≥ 8) to preserve context.

Output log - e.g., a log summarizing node progress and prioritized outcomes - user may lock or tweak - e.g., outcomes confirmed.