adaptic
について
アドパティックは、瞑想、観察、統合を含む5段階のシノプティック・サイクルを調整するマスタースキルであり、3つ以上の領域にわたって統合されたパノラマ的総合を生成します。個々の領域の深さよりも領域間の相互作用が重要である場合、逐次的分析が不十分に感じられる場合、または複数のステークホルダーに影響を与える主要なアーキテクチャ上の決定を行う前に使用してください。これは、順次的な妥協ではなく、首尾一貫した全体的な理解を生み出します。
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/adaptic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
Adaptic
5-step synoptic cycle → panoramic synthesis across domains. Sequential = compromise ("little of each"). Synoptic = integration → unified understanding holds all domains at once → emergent whole.
Use When
- Problem spans 3+ domains, interactions > depth in any one
- Polymath sequential tried → synthesis feels like compromise
- Existing approaches = "little of each" not unified vision
- Before major architectural decisions, multi-stakeholder
- Domain experts disagree → resolution lives between
Use NOT When
- Single-domain → use domain agent direct
- Well-understood trade-offs → polymath sequential enough
- Self-care/wellness → tending team
- Speed > depth → full cycle needs sustained attention
In
- Required: Problem requiring multi-domain synthesis
- Optional: Explicit domain list (default: auto-detect)
- Optional: Depth —
light,standard,deep(default:standard) - Optional: Expression form —
narrative,diagram,table,recommendation(default:auto)
Config
settings:
depth: standard # light (skip meditate), standard, deep (extended perceive)
domains: auto # auto-detect or explicit list
expression_form: auto # narrative, diagram, table, recommendation
Do
Step 1: Clear — Empty Workspace
Run meditate → clear prior ctx, assumptions, single-domain bias.
- Full meditate proc: prepare, anchor, observe distractions, close
- Domain bias = tendency to frame via recently-active domain
- Clear premature solutions arrived pre-full-picture
depth: light→ brief ctx-clearing pause
→ Workspace empty. No domain priority. No solution pre-selected. Neutral receptive → hold multiple perspectives.
If err: Domain keeps asserting → name bias: "I frame this as primarily [domain]." Naming loosens. Clearing fails → genuinely single-domain → reconsider.
Step 2: Open — Panoramic Mode
Run expand-awareness → narrow → wide-field perception.
- Inventory all domains → no pre-filter/rank
- Per domain: core concerns, constraints, values — no eval
- Soften focus: hold all simultaneously vs cycling
- Resist "start solving" → opening field only
- Domains in inputs → starting set, open to more
→ Panoramic field open. All domains simultaneous. Full landscape sensed. Spacious not overwhelming.
If err: List incomplete → "What missing would change picture?" Simultaneous → sequential scan → slow down. >7 domains → cluster related.
Step 3: Perceive — Cross-Domain Patterns
While maintaining panoramic, run observe + awareness → notice patterns, tensions, resonances across domains.
- Hold Step 2 field open → no narrow
observe→ what present: patterns across domains? tensions? resonances?awareness→ what not seen: ignored domains? blind spots? surface assumptions?- Record cross-domain no interpret:
- Tensions: domains pull opposite
- Resonances: domains reinforce/echo
- Gaps: no domain addresses, whole reveals
- Surprises: domain unexpected contribution
depth: deep→ cycle multiple times → subtler patterns
Critical: perceive across all simultaneously, not each in turn. Sequential loses cross-domain patterns = entire point.
→ Rich cross-domain obs — tensions, resonances, gaps, surprises. Span boundaries not live within. Noticed something invisible from any single domain.
If err: All within single ("in domain A, I notice X") → field collapsed → Step 2. No cross-domain → problem not synoptic → genuinely decomposable. Overwhelming → prioritize tensions (integration happens there).
Step 4: Integrate — Emergent Whole
Run integrate-gestalt → synthesize cross-domain obs → unified understanding.
- Map Step 3 tensions → don't resolve prematurely → hold as creative constraints
- Find figure: unified understanding when all held together? Not compromise/avg → new pattern includes+transcends individual
- Test whole: honors each domain's core concerns? Resolves tensions or papers over?
- Name insight one clear statement → unstatable simply = incomplete
- Verify emergent: reachable sequentially? Yes → synoptic added no value → sequential suffices
→ Single integrated understanding holding all simultaneously. Feels like discovery not construction — emerged from whole. Each domain honored, tensions resolved not compromised.
If err: "Little of each" not unified → gestalt not formed → Step 3, find avoided tensions — integration happens through tension. No gestalt → decompose: 2-3 strongest-tension domains first, then expand.
Step 5: Express — Communicate
Run express-insight → communicate synthesis.
- Assess audience: what domains familiar? framing makes accessible?
- Expression form (or input):
- Narrative: audience needs parts→whole journey
- Diagram: structural relationships
- Table: systematic comparison
- Recommendation: actionable decision
- Express w/ transparency: which domains contributed, where tensions resolved, emergent insight beyond any single
- Invite challenge: which aspects strongest, which most speculative
→ Clear expression accessible to audience. Shows work → audience sees domain contributions → whole. Form matches audience needs.
If err: Feels like list not integrated → insight lost → Step 4 one-statement summary, build outward from center. Wrong framing → "Who needs this and what decision does it inform?"
Check
- Step 1 (Clear) ran → ctx + domain bias released
- Step 2 (Open) produced panoramic 3+ domains
- Step 3 (Perceive) cross-domain patterns (not within-domain)
- Step 4 (Integrate) single emergent transcends individual
- Step 5 (Express) form appropriate to audience
- Output unreachable by sequential single-domain
- Each domain's core concerns honored
- Tensions resolved through integration, not compromise
Traps
- Sequential masquerading as simultaneous: Cycling domains + stapling results ≠ synoptic. Test: cross-domain interactions produced new, or just concatenation?
- Premature integration: Jump synthesis pre-panoramic field open. Steps 2+3 build foundation → rushing = shallow.
- Compromise instead of emergence: Avg ("50% security, 50% usability") = compromise. True integration finds frame where both fully met, or honestly names irreducible trade-off.
- Overuse single-domain: Not every problem panoramic. Single domain → synoptic adds overhead no value. "Use NOT When" exists.
- Losing insight in expression: Step 4 gestalt → Step 5 fragments back to domain list. Keep integrated insight center; domain details supporting evidence.
- Domain inflation: Artificially expand count → justify synoptic. 3 genuinely relevant > 7 where 4 peripheral.
→
meditate— Step 1; clears ctx + neutral stateexpand-awareness— Step 2; narrow → panoramicobserve— Step 3; what present across fieldawareness— Step 3; what not seen, blind spotsintegrate-gestalt— Step 4; emergent whole from cross-domainexpress-insight— Step 5; communicate integrated understanding
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。
