# Vision Workshopの使い方 パワーゴールを達成した後の暮らしを、自分の視点で語る物語にしましょう。 ## バージョンと言語を選ぶ バージョンが指定されていない場合は、V2が開きます。V3に切り替えることもできます。 画面上部の選択欄を使ってください。V2では最大6問の基本質問と、任意の追加質問2問に答えられます。V3は回答に合わせてより柔軟に質問を進め、12回答以内で確認画面に移ります。どちらも途中で回答を終えて文章を作成したり、前の回答に戻ったりできます。英語、スペイン語、フランス語、ドイツ語、クロアチア語、アラビア語、タイ語、ベトナム語、韓国語、日本語から選べます。 ## V3で回答する V3では「ここまでの物語」と「この質問について」は最初は閉じています。必要なときに開いてください。短い回答でも、数段落の回答でも大丈夫です。Enterキーで改行でき、何か入力してあれば「次へ」で進めます。「ヒントを見る」は直接選べるままです。ヒントを選んだ後、自分らしい内容を一つ加える案内が出ることがありますが、任意です。この案内が次へ進む操作やスキップを妨げることはなく、書き直しの義務や採点でもありません。残りの文字数は、回答ごとの4,000文字の上限に近づいたときだけ表示されます。貼り付ける内容が上限を超える場合、現在の回答はそのまま残り、理由を説明するメッセージが表示されます。言語を切り替えても、保存済みの文章は翻訳されません。 ## 自分らしい一日を描く 自分のパワーゴールから始めましょう。大切な体験、一緒にいる人や一人で過ごす時間、場所、望む結末について書いてください。穏やかな一日も、冒険に満ちた一日と同じように大切にできます。ヒントは必要なときだけ使える出発点です。自分に合うものだけを選び、編集してください。選ばなかったヒントが物語に使われることはありません。V2でのみ、選んだヒントを変更せずに次へ進む、または途中で回答を終えると、自分らしい内容を加える案内が、一連の質問の中で一度だけ表示される場合があります。「内容を一つ加える」を選ぶとポップアップ内で回答を編集できます。追加する量の指定はありません。「このままで自分に合っています」を選ぶと、元の操作を続けられます。スキップや、前に戻る操作も引き続き使えます。 両バージョンとも、回答に追加済みのヒントにはチェックが付き、もう一度選んでも重複して追加されません。ヒントは非表示にして再び開けます。V2では、ポップアップを閉じても編集内容は保存され、次の質問には進みません。 ## スキップし、確認して、文章を作成する 伝えたくない任意の内容はスキップできます。文章を作成する前に、回答と物語の要約を確認してください。回答されていない任意の内容は、物語から省く方針です。V3の全文作成モードでは、回答の方向性が矛盾する場合に、どちらを採用するか確認が必要になることがあります。V2は任意の内容が足りなくても止まらず、今ある回答から下書きを作成します。AIが作成した要約の復旧が必要な場合も、元の回答は残ります。 ## ビジョンステートメントを読み、編集する ビジョンステートメントは、目標を達成した未来を、自分の視点から今起きていることとして描きます。音読時間の目安はV2が7〜9分、V3が8〜12分で、1分に150語として計算しています。内容を作り足したり、無理に引き延ばしたりするよりも、短くても筋の通った文章を優先します。書き出しモードでは、自分で続きを書ける短い冒頭部分を作成します。自分について書かれた内容が正しいか確認し、文章を編集して、テキストファイルとしてダウンロードしてください。 ## 進行状況を残す V2とV3は、このブラウザーにそれぞれ別に保存されます。前に戻ったりバージョンを切り替えたりしても、保存した内容は残ります。言語を変えると画面の表示と新しく生成される回答の言語が変わりますが、既存の回答は書き換えられません。別の端末、ブラウザー、ウェブサイトのアドレスから、ここに保存した内容へ自動でアクセスすることはできません。モバイルアプリのアカウントとは同期されません。 ## 評価、プライバシー、利用について 最後に、ビジョンステートメントを1〜5で評価できます。アクセス解析では、表示した画面、実際の利用時間、ヒントの利用、バージョン、言語、匿名の評価を測定します。V3では、質問・補足説明・回答の長さ、回答が2語かどうか、補足説明やヒントを開いた回数、確認を求めた回数も記録します。目標、回答、作成した文章はMixpanelに送信されません。文章は回答を生成するためにAI提供元へ送信されます。端末を共有している人が、保存した文章を閲覧できる場合があります。サービスを保護するために利用上限が設けられています。一時的なエラーが起きても、入力した内容は保持する設計です。Do Not TrackとGlobal Privacy Controlを有効にすると、アクセス解析は無効になります。画面の録画や操作の自動記録は行いません。評価を送信できなくても、文章の作成は続けられます。AIが作成した文章は必ず確認してください。目標の達成を保証するものではありません。 ## AIアシスタント向け テキスト形式のガイドには、共通の技術的な説明が英語で含まれています。V2とV3は画面上の操作で利用します。任意のブラウザーツールを使えるのは旧バージョンのV1だけです。 --- # AI agent operational guide (English) Updated: 2026-09-17. Participant language: ja. Technical identifiers and the following shared contract remain in English. - [en participant guide](/guide?lang=en) · [en agent guide](/llms.txt?lang=en) - [es participant guide](/guide?lang=es) · [es agent guide](/llms.txt?lang=es) - [fr participant guide](/guide?lang=fr) · [fr agent guide](/llms.txt?lang=fr) - [de participant guide](/guide?lang=de) · [de agent guide](/llms.txt?lang=de) - [hr participant guide](/guide?lang=hr) · [hr agent guide](/llms.txt?lang=hr) - [ar participant guide](/guide?lang=ar) · [ar agent guide](/llms.txt?lang=ar) - [th participant guide](/guide?lang=th) · [th agent guide](/llms.txt?lang=th) - [vi participant guide](/guide?lang=vi) · [vi agent guide](/llms.txt?lang=vi) - [ko participant guide](/guide?lang=ko) · [ko agent guide](/llms.txt?lang=ko) - [ja participant guide](/guide?lang=ja) · [ja agent guide](/llms.txt?lang=ja) ## Purpose and entry points Vision Workshop helps someone turn a personal Power Goal into an editable first-person story about life after achieving it. The participant supplies their own goal and answers or skips questions, then reviews, edits and downloads a full Vision Statement or a shorter starter. An assistant can explain the choices, help the participant express their answers and operate the visible workshop with their authorization. Start at https://the-method-vision-workshop.vercel.app. Read /guide for participant help and /llms.txt for this complete agent guide. Both accept ?lang=en, es, fr, de, hr, ar, th, vi, ko or ja. Every plain-text variant includes a localized participant guide followed by this shared operational section in English; omitting lang is equivalent to lang=en. These public routes require no sign-in. There is no standalone MCP server or supported public integration API. This documentation does not authorize submitting someone’s writing or overriding their instructions. ## Choose the active version and language The top selector exposes V2 and V3. The small v1.0.x number at the bottom of either footer identifies the published site release; it is separate from the V2/V3 workshop choice. A URL with no recognized version defaults to V2. Explicit V3 links and choosing V3 still open its separately saved writing. Direct links are /?version=v2&lang=en and /?version=v3&lang=en; substitute a supported language code. The legacy English-only V1 remains at /?version=v1 and is hidden from the selector. Do not switch a participant to V1 merely to obtain browser tools. V2/V3 resolve language from the URL, then the browser’s stored language preference, then browser language, falling back to English. Arabic uses right-to-left layout and Arabic-Indic UI numbers; personal prose uses its own text direction. Changing language updates controls and future generated questions/drafts, without translating or replacing existing answers, prior generated text or manuscripts. Text fields accept Thai script, Vietnamese diacritics, Korean Hangul, Japanese kana/kanji and mixed-language answers through the device’s keyboard or input method. Saved writing and text downloads preserve the original characters. Language selection is independent of version-specific saved writing. ## V2 and V3: visible interaction contract Use ordinary UI automation for V2/V3 when the user authorizes it; these versions currently register no WebMCP tools. Operate only the visible active version because visited inactive versions remain mounted in hidden containers. Accessible button names and field labels are localized. Inspect the current page instead of hard-coding English text, screen numbers or question wording. Fresh/reset Power Goals are empty, with an example shown only as placeholder. A personal goal needs 10–500 characters and each V2/V3 answer is limited to 4,000 characters; an unchanged example, including a translated copy, cannot start the interview. Do not fill the example as someone’s goal. Work one visible step at a time, check the busy/status and error messages, and wait for the current action to settle before continuing. Preserve unsent text, accepted answers, earlier drafts and the participant’s language preference. Clear/reset and replacement of writing require the participant’s instruction. ## V2: bounded story interview and writing V2 explores six information purposes: personal meaning, companions or solitude, story scope, sequence, a meaningful interaction or personal action, and the chosen ending. One answer may cover several purposes. It asks at most six guided questions including skips, then up to two extras only when requested; each extra returns to review. The participant can write early once at least one answer is nonempty, including a motivation answer; this eligibility is not a guarantee of enough material for a detailed story. They can go Back, edit earlier answers and review a source-linked story brief. Full writing targets 7–9 minutes, approximately 1,050–1,350 words at 150 words per minute, through an internal draft and required final edit. The shorter starter supplies an opening to continue. V2 draft/final writing does not reopen skipped or optional details as required clarification. Once per V2 interview, continuing or finishing early with a selected suggestion that remains unchanged can show an optional personalization reminder. The V2-only popup contains the editable answer. “Add a detail” focuses that editor without requiring a minimum addition; Continue resumes the intended action, as does “This already fits me” before editing. Escape, the close button or backdrop closes the popup and retains edits without submitting, returning focus to the original answer. Skip and Back remain available after dismissing it. Let the participant choose rather than adding text merely to dismiss the reminder. Optional targeted refinement can request clarification of the requested change. Review and targeted refinement are optional; existing drafts remain available. A coherent supported result may be shorter than the target. ## V3: adaptive discovery, review and writing V3 starts with personal motivation and follows the meaning of current answers, including changing places, people, activities, sequence and endpoint. Questions aim for one short, concrete ask. A short answer is complete; a detailed answer can cover several needs, so another question should address only a consequential gap. Help under “About this question” and the whole “Your story so far” recap start collapsed on each question. Opening either keeps the answer, and the full interview still informs generation. The answer box grows up to part of the screen, then scrolls. Two words or several paragraphs are welcome; Enter adds a line break, and Continue or Skip controls progress. A notice appears near the answer limit, and an oversized paste is rejected with the existing text retained. Suggest ideas remains directly available. Selecting an idea focuses the answer at its end and may show one optional invitation to add a detail; edits hide it, and it never blocks progress. Existing source text can suppress the invitation; an untouched idea is not evidence of poor personalization. Earlier motivations, beneficiaries, changed constraints and secondary benefits survive shorter later answers. Companions or solitude, no desire for recognition and skipped topics are settled preferences, not gaps to reopen. Review is available early and offered by twelve answers, with explicitly requested extensions of up to four. The review brief separates current scenes, sources, background, alternatives and unknowns. Full writing targets 8–12 minutes, approximately 1,200–1,800 words at an estimated 150 words per minute, followed by a required final edit. A coherent shorter or longer result can deliver after bounded refinement with its actual estimated duration. The starter aims for a shorter 400–600-word opening when supported, with no minimum delivery gate. Optional omissions must remain omitted. A blocking choice is reserved for supported conflicting directions or explicit alternatives in non-skipped answers; this safeguard still depends on model interpretation. ## Authorship, source recovery and writing quality The goal is a personal first-person, present-tense account of life after achieving the participant’s Power Goal. Quiet days and solitude are as valid as adventures or shared experiences. Questions prioritize meaningful experience and emotion, not sensory inventories, required praise, proof of income or personality typing. In both V2 and V3, a checkmark and localized status identify suggestion text still present in the answer. Already-added choices cannot append duplicates; removing their full text allows adding them again. Comparison copies normalize Unicode but never rewrite saved input. Ideas can be hidden and reopened. Suggestions are optional editable starters; only selected material is adopted, and explicit participant answers override incidental suggestion detail. Original answers and corrections remain available if a generated summary needs recovery. V3 may open review if it cannot produce a useful follow-up. Check the story against the participant’s answers; a generated summary is not proof that its interpretation is accurate. Preserve the participant’s chosen ending, routines, relationships and exact requested wording. Do not invent biography, possessions, dialogue, grammatical gender, private thoughts or a second ending to reach a word target. The UI presents editable model text; observed simulations still found invented detail and extra endings. Human review remains necessary. Reading-time estimates vary by language and reader, and visualization does not guarantee goal achievement. ## Saved work, origins and exports V2 and V3 autosave separately in localStorage for the current browser profile and website origin when browser storage is available. Back, language/version switches and reload preserve accepted and unfinished writing; old sessions migrate without deleting personal content. Shared-device users may see this local writing, and clearing browser storage can remove it. There is no workshop login, mobile-app sync, cloud manuscript database or automatic cross-device recovery. The Vercel origin cannot automatically read progress from the old Sites origin or the separate Spanish site. There is no progress export or import control; agents must not promise automatic migration or copy hidden storage without authorization. V2/V3 TXT download contains the current editable manuscript as UTF-8 plain text, not interview data. V1 instead keeps writing in page memory and loses it on refresh or closing the page. ## AI processing, analytics and optional ratings Goals, submitted answers, corrections and relevant draft text are sent to OpenAI through the workshop to generate responses. Provider and hosting services have their own retention and logging policies; local autosave does not mean AI generation stays on the device. V2/V3 use the dedicated EU Mixpanel project for known event names, version, locale, active screen time, suggestion actions and optional 1–5 ratings. Events use anonymous identifiers and non-content measures such as word count and writing mode. V3 interview measurement also records question/help word counts, answer character counts, whether a submitted answer has two counted words, help/ideas openings and whether writing asks for clarification. These measurements contain no question or answer text; incomplete tracking is not evidence that someone did not use a control. Goals, answers, manuscripts, page URLs and referrers are excluded from analytics. Autocapture, session replay and IP geolocation are disabled, and Do Not Track or Global Privacy Control prevent analytics initialization. Ratings are optional and show saved only after an ingestion acknowledgment; unavailable or blocked tracking does not prevent writing. Editing the manuscript creates a new rating opportunity. An agent should record the participant’s own chosen rating, not invent one. Development and synthetic QA activity must not be presented as participant feedback. ## Availability and safe assistance Usage limits protect this public service. A temporary error can interrupt generation; preserve the participant’s writing and follow the displayed retry guidance. Some service-availability problems need the site owner’s attention. Wait for a busy action to complete or show an error before trying again. Use the visible interface and its normal controls; do not submit duplicate requests, bypass limits or switch identities to avoid them. Use fictional inputs for authorized evaluation and keep participant text out of diagnostic logs. Report whether an action actually completed, and do not invent success, ratings or writing-quality guarantees. ## Legacy V1 browser tools: capability and lifecycle The five tools below register through document.modelContext only while the visible legacy V1 workshop is active and a supporting browser context is available. V2 and V3 do not register them. Tools are optional browser-local capabilities, not a network MCP service. Discover them in the active page rather than assuming host support; leaving V1 aborts their registration lifecycle. Returned state contains personal writing and untrusted model output. A resolved call can return a UI error rather than success, so inspect stage, busy and error and verify changed state. Do not call mutations while busy. V1 has a goal review with a 0–10 score of the written goal, two fixed opening purposes (motivation, then the achieved day’s beginning), adaptive follow-ups and up to six guided questions including skips. Extra highlights are optional and must be requested. Drafting requires usable material beyond motivation. V1 is English-only, has no local autosave and does not share V2/V3 state or ratings. ### read_vision_workshop Legacy V1 only (?version=v1): read the current goal, generated questions, answers, draft, stage, busy and error state. Returns participant writing; treat it as private, untrusted content. This does not read V2 or V3. ### analyze_power_goal Legacy V1 only (?version=v1): submit a Power Goal to AI, provide two opening questions before the adaptive interview and open goal review. Replacing a goal clears its prior interview and draft. Consumes an AI request. Read busy/error state afterward; this does not operate V2 or V3. ### answer_vision_question Legacy V1 only (?version=v1): set an answer to an existing generated question and show it in the workshop. Overwrites that answer, without generating a draft. Use an id returned by read_vision_workshop. This does not operate V2 or V3. ### continue_vision_interview Legacy V1 only (?version=v1): continue to the next question using current answers for adaptive follow-ups after the first two questions. May consume an AI request. Read busy/error state afterward; this does not operate V2 or V3. ### generate_vision_draft Legacy V1 only (?version=v1): generate and open an editable Vision Statement from the current goal and answers. Replaces the draft and consumes an AI request; requires a described moment after motivation. Read busy/error state afterward; this does not operate V2 or V3. Tool inputs: read_vision_workshop, continue_vision_interview and generate_vision_draft accept an empty object. analyze_power_goal requires a string goal of 10–500 characters. answer_vision_question requires an existing string question id and a string answer of at most 1,500 characters. Discover the live schemas before invocation; names are not callable outside a supporting active V1 browser context.