express-insight
について
このスキルは、複雑で複数領域にわたる洞察の伝達を構造化し、相互の関連性を損なうことなく、理解しやすく実践可能な形にします。適切な形式の選択、明確な出典明示による統合された理解の提示、建設的なフィードバックの招致という手順を提供します。開発者は、専門家、ジェネリスト、意思決定者に対して領域横断的な知見を効果的に共有するために、`integrate-gestalt`の後にこのスキルを使用すべきです。
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/express-insight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
Express Insight
Say multi-domain gestalt so it lands — keep links between domains, make synthesis open to audience, be honest about where integration shrinks or risks distortion. Express is final step of synoptic cycle: without it, integrated understanding stays private and unactionable. Challenge is language is linear but insights are not — this skill gives shapes for talking multi-dimensional understanding without shrinking to single dimension.
When Use
- After
integrate-gestalthas made cross-domain understanding that needs to be said - When find spans many domains and single-domain frame would lose key links
- When audience for insight differs from view that made it
- When integrated understanding feels clear inside but resists plain talk
- When pick depends on seeing how many domains touch, not just what each domain says alone
- When prior tries to talk cross-domain find met with confusion or domain-specific pushback
- When talking finds from synoptic-mind team session to stakeholders outside team
Inputs
- Required: Integrated insight (output of
integrate-gestaltor same cross-domain synthesis) - Required: Audience — who gets this insight (domain specialists, generalists, decision-makers, or mixed)
- Optional: Limits on form (e.g., "must fit in PR description", "needs to be verbal summary")
- Optional: Domains that were integrated (for clear credit)
- Optional: Past failed tries to talk this insight (what did not land)
Steps
Step 1: Assess Audience
Figure who gets this insight and what they need from it. Same gestalt said to three different audiences should take three different forms.
- Spot main audience:
- Domain specialists need their domain shown right — they will reject insight that oversimplifies their field, even if whole synthesis is right
- Generalists need big picture — links between domains matter more than details inside any single one
- Decision-makers need actionable implications with trade-offs — they want to know what to do, what it costs, what happens if they do nothing
- Mixed audiences need layered talk: lead with big picture, then give domain-specific depth specialists can check
- Check audience existing mind-model:
- What do they already know about each domain?
- What links between domains will be new to them?
- What assumptions might they hold that insight challenges?
- Spot trust need: how much backing does this audience need before they take cross-domain claim?
- Specialists trust insight that respects their domain rigor
- Generalists trust insight that makes complex stuff navigable without oversimplify
- Decision-makers trust insight that honest shows trade-offs rather than hides them
Got: Clear picture of who audience is, what they need, what will make insight credible to them. Audience check should shape every later step.
If fail: Audience unknown or too broad to pin? Default to mixed-audience way: big picture first, domain depth on demand. Talking to "everyone" is less useful than talking to someone specific, but it is better than guessing wrong.
Step 2: Choose Form
Pick express format that best serves audience and nature of insight. Form is not decoration — it picks what audience can see.
-
Check four main forms:
Form Structure Best for Narrative Story connecting domains — "when X happens in domain A, it creates Y in domain B, which means Z" Complex or novel insights where the audience needs to follow the reasoning path Diagram Spatial layout showing relationships — nodes are domain contributions, edges are connections Structural insights where the topology of relationships matters more than the sequence Comparison table Each domain's perspective on the same issue in parallel columns Analytical audiences who want to verify each domain's contribution independently Recommendation Actionable synthesis — "do X because domains A, B, and C converge on Y, with trade-off Z" Decision-makers who need to act, not just understand -
Match form to insight type:
- Insight is about causal chain across domains? Use narrative
- Insight is about structural links? Use diagram
- Insight is about convergence or divergence between domains? Use comparison table
- Insight is about what to do next? Use recommendation
-
Think mix forms: recommendation backed by comparison table, or narrative shown with diagram. But lead with one main form — cognitive load from many formats can hide rather than clear
-
Account for medium limits: verbal summary can't carry comparison table; commit msg can't carry narrative. If medium limits form, tune form rather than forcing content into wrong container
Got: Picked main form (and optional second form) with clear reason tied to audience and nature of insight.
If fail: No form feels right? Insight may not yet be fully integrated. Go back to integrate-gestalt — hard express often signals thin synthesis, not talk problem.
Step 3: Express the Gestalt
Say insight in picked form, clear note what it integrates, where it shrinks, and what it enables.
- State insight clear — one to three sentences that catch core understanding. This is gestalt itself, not backing evidence
- Name domains it integrates — clear list which domains fed this understanding. This is not credit for credit; it is credit for check. Each named domain is invite: "check this vs your expertise"
- Mark shrinks — every multi-domain insight shrinks. State where:
- Which domain-specific nuances were set aside?
- Which links treated as stronger or weaker than they might be?
- What would specialist in domain X want to add or qualify?
- State emergent value — what does this insight enable that single-domain analysis does not?
- What pick becomes possible that was not before?
- What risk becomes visible that was hidden inside single domains?
- What chance shows at the crossing that no single domain owns?
- Keep multi-domain texture — resist pull to flatten insight into one domain language. If insight integrates engineering and UX, use both vocabs. If it links research and ops, keep both framings. Texture is the insight
- Sequence for grasp — even though insight is non-linear, talk is sequential. Pick entry point that gives audience best foothold: start with domain they know best, then bridge out into unfamiliar domains. First sentence picks whether audience leans in or tunes out
Got: Said insight audience can grasp, check vs own expertise, and act on. Shrinks are visible, not hidden. Emergent value clear.
If fail: Express feels like list of domain contributions rather than integrated whole? Gestalt got broken during talk. Step back and re-express: start from what combination shows, not from what each domain says apart. Synthesis is message, not parts.
Step 4: Invite Challenge
State strongest reason insight might be wrong. Integrated insights can feel more certain than they are since they blend many inputs — convergence gives sense of validity that may be unearned. This step is not disclaimer slapped on for polite; it is structural part that makes insight usable.
- Spot weakest link — which domain link in insight is least well-backed? Where does integration lean on analogy rather than evidence?
- Name assumption at risk — what would need to be true for insight to hold, and how sure are you that it is true?
- State counter-insight — if someone with equal access to all same domains reached different conclusion, what would their strongest argument be?
- Frame challenge as valuable — make it clear that challenge strengthens insight. "Strongest objection I see is..." signals confidence and openness at once
- Say what would change your mind — name evidence or argument that would revise or collapse insight. This makes insight falsifiable, not just persuasive
Got: Honest statement of uncertainty that raises rather than drops audience trust. Insight is now challengeable — and so improvable.
If fail: No weakness can be spotted? That itself is warn sign. All cross-domain insights need translation between frames, and translation always loses something. If loss invisible, it has not been found yet, not avoided. Look harder at domain edges — that is where hidden assumptions live. Common hide spots: shared metaphors that work differently in each domain, stat correlations assumed causal across domain edges, and analogies that hold structurally but not quantitatively.
Validation
- Audience was clear named and their needs shaped express
- Form was picked by insight type and audience, not habit or convenience
- Insight stated as coherent whole, not broken into per-domain summaries
- Contributing domains named for check, not just credit
- Shrinks stated clear — what was set aside and what was approximated
- Emergent value said — what integration enables that parts do not
- Multi-domain vocab kept rather than flattened into one domain language
- Entry point picked for audience existing knowledge — starts where they are, bridges to where insight goes
- Strongest reason insight might be wrong was stated
- Insight is falsifiable — specific evidence or arguments that would revise it were named
- Domain specialist reading their domain contribution would know it as right, not caricatured
Pitfalls
- Domain-by-domain report: Showing each domain contribution one by one is not expressing insight — it is showing raw stuff. Insight is what shows up from combo. Lead with synthesis, then back with domain detail if needed
- False certainty from convergence: When three domains all seem to point same way, it feels like strong evidence. But if those domains share hidden assumptions or data sources, convergence is less independent than it looks. Always check whether domains truly independent
- Flatten to audience domain: When talking to specialist, urge is to translate whole insight into their language. This makes it open but destroys multi-domain nature. Keep texture — unfamiliar vocab is not noise, it is signal
- Skip challenge step: Leaving out "here is why I might be wrong" feels like it makes insight stronger. It does not. It makes insight less trustworthy and less improvable. Epistemic honesty is feature, not weakness
- Insight inflation: Claim synthesis shows more than it does. Cross-domain watch is not auto-breakthrough. Be precise about scope: "this applies to X in context Y" is more valuable than "this changes everything"
- Premature express: Express before gestalt fully formed gives half-insights that sound integrated but fall under scrutiny. If express keeps stalling, problem is upstream in
integrate-gestalt, not here - Hide behind complex: Use multi-domain vocab to sound fancy rather than to keep real texture. If simpler frame catches same insight without losing links, use simpler frame. Complex should be needed, not performed
See Also
integrate-gestalt— makes insight this skill expresses; express-insight is talk phase of synoptic cycleargumentation— builds logic case for claim; express-insight says a seeing. Argumentation says "here is why X is true"; express-insight says "here is what becomes visible when you look at A, B, and C together"teach— passes known, set knowledge; express-insight sends emergent understanding just formed. Teaching transmits; expressing showsshine— channels real presence into talk; express-insight can use that radiance to carry multi-domain seeing without losing warmth or honestyexpand-awareness— widens seeing field that makes integration possible; express-insight closes cycle by talking what that widened field showedadaptic— meta-skill composing full synoptic cycle; express-insight is fifth and final step in clear-open-see-integrate-express sequence
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。
