dream
について
`dream`スキルは、構造化された入力と検証を意図的に排除し、自由な連想と創発的なアイデアを可能にすることで、制約のないAIの創造性を発揮させます。早期のアイデア出しフェーズ(デザイン、命名、アプローチ選択など)で使用してください。この段階で早期に構造化すると可能性が制限されてしまいます。このスキルは、評価よりも開かれた探索を優先することで、標準的な手順を逆転させます。
クイックインストール
Claude Code
推奨npx skills add pjt222/agent-almanac -a claude-code/plugin add https://github.com/pjt222/agent-almanacgit clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/dreamこのコマンドをClaude Codeにコピー&ペーストしてスキルをインストールします
ドキュメント
Dream
Unconstrained creative roam. Every other skill give structure — inputs, steps, check. Dream drop structure. Let links form free, possibility emerge no judge, idea arrive before get weighed.
When Use
- Before design work — roam space before lock structure
- Before name thing — name squeeze mind; dream before label
- Before pick between path — let option emerge analytic mind miss
- Stuck in loop — same option keep coming back, frame too tight
brahma-bhaga(make from void) feel too tight — dream come before make- After
meditateclear noise — empty space best canvas for dream
Inputs
- Required: Seed — topic, problem, space to roam. Can be vague. Vague good.
- Optional: Constraints to forget short time (e.g., "skip speed worry now," "forget old arch")
- Optional: File or codebase read as raw stuff for roam (via
Read)
Steps
Step 1: Soften Frame
Drop structures that guide mind.
- Set aside task rules — not forever, only for dream. Rules come back after
- Set aside judge rules — nothing "good" or "bad" in dream. Nothing "feasible" or "impractical." Judge come later
- Set aside old solutions — what worked before is hint, not chain. Dream like first time see space
- Set aside need for order — dream not need make sense step by step. Let links jump, surprise, clash
Analytic mind protest: "This waste time. Already know answer. Why?" That protest is why dream needed — "known answer" maybe blocks better path.
Got: Looser mind-state. Idea arrive before get sorted, ranked, thrown out.
If fail: Frame not soften — every idea judged fast? Flip on purpose: "What is worst path?" Bad ideas played with often hide seed of best one.
Step 2: Wander
Follow links. Do not steer.
- Start at seed. What remind you of? What sit next to it? What rhyme (idea-wise, not sound-wise)?
- Follow first link. Then follow where that go. Do not steer — wander.
- Let images form. What problem look like? If land, what terrain? If sound, what timbre? Cross-sense links slip past analytic filter.
- Let clash sit. "Should be simple AND big." "Must be fast AND deep." In waking mind, trade-off. In dream, creative tension waiting to fold into new shape.
- Collect bits. Not whole idea — bits. Word. Picture. Half-link. Question. Raw stuff.
No set number of links or bits. Dream till space feel mapped — or till one bit grab attention and want become thing.
Got: Pile of bits, links, pictures, half-ideas. Messy. No order. Alive with chance.
If fail: Wander make nothing — mind go blank not free? Read something. File, codebase, doc. Raw stuff spark links when blank page not. read-garden sense apply: see what there, let it hint what could be.
Step 3: Notice What Glows
In bits, one will have heat. Not logic — heat. Life. Curiosity.
- Scan bits. Do not rank. Which pull attention? Which spark curiosity?
- Notice clusters — bits that seem go together even link not clear yet
- Notice surprise — idea arrive unexpected, you not predict from seed
- Notice push-back — idea feel wrong or strange may be most worth. Push-back often mark edge of old thinking
- Do not force pick. Nothing glow? Dream need more wander (go Step 2) or seed need change
Got: One or more bit carry real heat — idea worth grow, even full shape not yet seen.
If fail: Nothing glow after long wander? Seed too abstract or too tight. Try new start: change seed, read unexpected, call remote-viewing to see space with no mask.
Step 4: Wake — Carry Bits Forward
Move from dream-state back to structured thought — gently.
- Gather glowing bits. Write plain — not finished idea but seed for structured work
- Do not judge yet. Let sit. Analytic mind get turn; not give bits before solid
- Note which bits touch first task, which are new ground
- Bit ready to become plan? Hand to
brahma-bhaga(make) or structured plan skill - Bits need more grow? Note for next dream — dream can loop
Dream end when you return to structured thought. Bits are gift. Some become feature. Some become rule. Some forgot. All widened space of chance past what analysis alone reach.
Got: Move from open roam to structured work, carrying hottest bits. After-dream feel wider — more option seen, more link ready.
If fail: Wake too fast — analytic mind crush bits right away? Use breathe as buffer. One breath between dream and judge keep fragile new idea safe.
Checks
- Structure truly dropped, not only loose (idea arrived that would die under judge)
- Wander was link-based, not steered (path was surprise, not planned)
- At least one bit carry real heat — not only new but alive
- Wake back to structure was gentle, not sharp
- Dream widened space of chance (more option seen than before)
- Dream sized right — not so brief shallow, not so long self-feeding
Pitfalls
- Dream as plan: Every "link" is really structured option check? You plan with extra step. Dream need real drop of structure
- Judging inside dream: Moment you think "won't work," dream over. Save judge for after
- Dream to dodge work: Dream is prep for creative work, not swap for it. Task clear and path obvious? Just do work
- Wait for finished idea: Dream give bit, not blueprint. Waiting for full plan from dream like waiting harvest from fresh seed
- Forced whim: Dream not about random or silly. Is real link-based roam. Forced creative not creative
- Never waking: Dream with no return to structure is daydream. Bits must meet real world
See Also
brahma-bhaga— structured make from void; dream is pre-make roam that feeds itmeditate— clear space that dream fill; meditate before dream for bestremote-viewing— enter unknown ground with no mask; shares dream openness but for see not makeintrinsic— real drive fuel dream; forced dream make nothingbreathe— micro-pause that guard dream bit during move to structured thoughtshine— bring real heat to idea dream surface
GitHub リポジトリ
関連スキル
content-collections
メタこのスキルは、Content Collections(Markdown/MDXファイルを型安全なデータコレクションに変換するTypeScriptファーストのツール)の本番環境でテストされた設定を提供します。Zodバリデーションによる型安全性を実現し、ブログ、ドキュメントサイト、コンテンツ重視のVite + Reactアプリケーション構築時にご利用ください。Viteプラグインの設定、MDXコンパイルから、デプロイ最適化、スキーマバリデーションまで、すべてを網羅しています。
polymarket
メタこのスキルは、開発者がPolymarket予測市場プラットフォームを活用したアプリケーション構築を可能にします。API統合による取引や市場データの取得に加え、WebSocketを介したリアルタイムデータストリーミングにより、ライブ取引や市場活動を監視できます。取引戦略の実装や、ライブ市場更新を処理するツールの作成にご利用ください。
creating-opencode-plugins
メタこのスキルは、開発者がコマンド、ファイル、LSP操作など25種類以上のイベントタイプにフックするOpenCodeプラグインを作成することを支援します。JavaScript/TypeScriptモジュール向けに、プラグイン構造、イベントAPI仕様、および実装パターンを提供します。カスタムイベント駆動ロジックでOpenCode AIアシスタントのライフサイクルをインターセプト、監視、または拡張する必要がある場合にご利用ください。
sglang
メタSGLangは、高性能なLLMサービングフレームワークであり、RadixAttentionプレフィックスキャッシュを活用したJSON、正規表現、エージェントワークフロー向けの高速で構造化された生成を特長とします。特にプレフィックスが繰り返されるタスクにおいて、大幅に高速な推論を実現し、複雑な構造化出力やマルチターン対話に最適です。制約付きデコードが必要な場合や、広範なプレフィックス共有を伴うアプリケーションを構築する場合は、vLLMなどの代替案ではなくSGLangを選択してください。
