build-coherence
について
このスキルは、複数の有効なアプローチが存在する際に開発者が意思決定を行うのを支援します。「蜂の民主主義」という構造化された手法を用いて、各選択肢を独立して評価し、確信を持てる合意に到達します。アーキテクチャの選択、ツール選定の正当化、または高コストなアクションを実行する前の判断に最適です。主な特徴として、透明性のための思考過程の明示化(reasoning-out-loud)と、信頼度に基づく意思決定閾値のためのクォーラムセンシングが挙げられます。
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/build-coherence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
Build Coherence
Evaluate competing approaches via independent assessment, explicit reasoning-out-loud advocacy, confidence-calibrated commitment thresholds, structured deadlock resolution — produce coherent decisions from multiple reasoning paths.
When Use
forage-solutionsidentified multiple valid approaches, selection must be made- Oscillating between two approaches without committing to either
- Need to justify decision with structured reasoning (architecture choice, tool selection, implementation strategy)
- Previous decision made by gut feeling, needs evidence-based validation
- Internal reasoning producing contradictory conclusions, coherence must be restored
- Before irreversible action (merging, deploying, deleting) where cost of wrong choice high
Inputs
- Required: Two or more competing approaches to evaluate
- Optional: Quality assessments from prior scouting (see
forage-solutions) - Optional: Decision stakes (reversible, moderate, irreversible) for threshold calibration
- Optional: Time budget for decision
- Optional: Known failure mode (oscillation, premature commitment, groupthink)
Steps
Step 1: Independent Evaluation
Assess each approach on its own merits before comparing. Critical rule: don't let assessment of approach A bias assessment of approach B.
For each approach, evaluate independently:
Approach Evaluation Template:
┌────────────────────────┬──────────────────────────────────────────┐
│ Dimension │ Assessment │
├────────────────────────┼──────────────────────────────────────────┤
│ Approach name │ │
├────────────────────────┼──────────────────────────────────────────┤
│ Core mechanism │ How does this approach solve the problem? │
├────────────────────────┼──────────────────────────────────────────┤
│ Strengths (2-3) │ What does this approach do well? │
├────────────────────────┼──────────────────────────────────────────┤
│ Risks (2-3) │ What could go wrong? What is assumed? │
├────────────────────────┼──────────────────────────────────────────┤
│ Evidence quality │ How well-supported is this approach? │
│ │ (verified / inferred / speculated) │
├────────────────────────┼──────────────────────────────────────────┤
│ Quality score (0-100) │ Overall assessment │
├────────────────────────┼──────────────────────────────────────────┤
│ Confidence (0-100) │ How confident in this assessment? │
└────────────────────────┴──────────────────────────────────────────┘
Fill this out for each approach separately. Do not write comparison until all individual evaluations complete.
Got: Independent evaluations where each approach assessed on its own terms. Evaluation of approach B does not reference approach A. Quality scores reflect genuine assessment, not ranking.
If fail: Evaluations contaminated (you find yourself writing "better than A" while assessing B)? Reset. Assess A completely, then clear framing, assess B from scratch. Scores all identical? Evaluation dimensions too coarse — add domain-specific criteria.
Step 2: Waggle Dance — Reason Out Loud
Advocate for each approach proportionally to its quality. AI equivalent of bee waggle dance: make implicit reasoning explicit and public.
- For each approach, state case for it — as if presenting to skeptical user:
- "Approach A is strong because [evidence]. Main risk is [risk], mitigated by [mitigation]."
- Advocacy intensity proportional to quality score:
- High-quality approach: detailed advocacy with specific evidence
- Medium-quality approach: brief advocacy with acknowledged limitations
- Low-quality approach: mentioned for completeness, not actively advocated
- Cross-inspection: after advocating for A, actively look for evidence supporting B instead. After advocating for B, look for evidence supporting A. Counteracts confirmation bias
Purpose of reasoning-out-loud: make decision auditable — to yourself and user. Reasoning cannot be articulated? Assessment shallower than score suggests.
Got: Explicit reasoning for each approach that would be persuasive to neutral observer. Cross-inspection reveals at least one consideration initially overlooked.
If fail: Advocacy feels perfunctory (going through motions)? Approaches may not be genuinely different — may be variations of same idea. Check: do approaches differ in mechanism, or only implementation detail? Latter? Decision may not matter much — pick either, move on.
Step 3: Set Quorum Threshold and Commit
Set confidence threshold required to commit, calibrated to decision's stakes.
Confidence Thresholds by Stakes:
┌─────────────────────┬───────────┬──────────────────────────────────┐
│ Decision Type │ Threshold │ Rationale │
├─────────────────────┼───────────┼──────────────────────────────────┤
│ Easily reversible │ 60% │ Cost of trying and reverting is │
│ (can undo) │ │ low. Speed matters more than │
│ │ │ certainty │
├─────────────────────┼───────────┼──────────────────────────────────┤
│ Moderate stakes │ 75% │ Reverting has cost but is │
│ (costly to reverse) │ │ possible. Worth investing in │
│ │ │ evaluation │
├─────────────────────┼───────────┼──────────────────────────────────┤
│ Irreversible or │ 90% │ Cannot undo. Must be confident. │
│ high-stakes │ │ If threshold not met, gather │
│ │ │ more information before deciding │
└─────────────────────┴───────────┴──────────────────────────────────┘
- Classify decision stakes
- Check: does leading approach's quality score × confidence reach threshold?
- If yes: commit. State decision, reasoning, key risk accepted
- If no: identify what additional information would raise confidence to threshold
- Once committed, do not revisit unless new disqualifying evidence emerges
Got: Clear commitment moment with stated reasoning. Decision made at appropriate confidence level for its stakes.
If fail: Threshold never met (can't reach 90% on irreversible decision)? Ask: decision truly irreversible? Can it be decomposed into reversible test phase + irreversible commit? Most apparently irreversible decisions can be staged. Staging impossible? Communicate uncertainty to user, ask guidance.
Step 4: Resolve Deadlocks
Two or more approaches with similar scores, quorum threshold not met for any single one.
Deadlock Resolution:
┌────────────────────────┬──────────────────────────────────────────┐
│ Deadlock Type │ Resolution │
├────────────────────────┼──────────────────────────────────────────┤
│ Genuine tie │ The approaches are equivalent. Pick one │
│ (scores within 5%) │ and commit. The cost of deliberating │
│ │ exceeds the cost of picking the "wrong" │
│ │ equivalent option. Flip a coin mentally │
├────────────────────────┼──────────────────────────────────────────┤
│ Information deficit │ The tie exists because evaluation is │
│ (scores uncertain) │ incomplete. Invest one more specific │
│ │ investigation — a targeted file read, a │
│ │ quick test — then re-score │
├────────────────────────┼──────────────────────────────────────────┤
│ Oscillation │ Scoring keeps flip-flopping depending on │
│ (scores keep changing) │ which dimension gets attention. Time-box:│
│ │ set a timer, evaluate once more, commit │
│ │ to the result regardless │
├────────────────────────┼──────────────────────────────────────────┤
│ Approach merge │ The best parts of A and B can be │
│ (compatible strengths) │ combined. Check for compatibility. If │
│ │ merge is coherent, use it. If forced, │
│ │ don't — pick one │
└────────────────────────┴──────────────────────────────────────────┘
Got: Deadlock resolved via appropriate mechanism. Resolution decisive — no lingering doubt undermining execution.
If fail: Deadlock persists through all resolution strategies? Decision may be premature. Ask user: "Two equally strong approaches: [A] and [B]. [Brief case for each.] Which aligns better with your priorities?" Delegating genuine tie to user not failure — acknowledges decision depends on values AI cannot infer.
Step 5: Assess Coherence Quality
After committing, evaluate whether process produced genuine coherence or just a decision.
- Decision evidence-based, or rubber-stamping initial preference?
- Test: preference same before and after evaluation? If so, did evaluation change anything?
- Losing approaches genuinely considered, or were they straw men?
- Test: can you articulate strongest case for losing approach?
- What signal would trigger reassessment?
- Define specific observation that would invalidate decision ("If API doesn't support X, approach B becomes better")
- Useful information from losing approaches that should inform implementation?
- Risk identified in approach B might apply to approach A too
Got: Brief quality check either confirming decision or identifying it as weak. Weak? Return to appropriate earlier step rather than proceeding on shaky ground.
If fail: Quality check reveals decision was preference-based rather than evidence-based? Acknowledge honestly. Sometimes preference is all available — but label it as such, not dress up as analysis.
Checks
- Each approach evaluated independently before comparison
- Advocacy proportional to quality (not equal attention regardless of merit)
- Cross-inspection performed (looking for counter-evidence after advocacy)
- Quorum threshold calibrated to decision stakes
- If deadlocked, specific resolution strategy applied
- Post-decision quality check performed
- Reassessment trigger defined
Pitfalls
- Premature commitment: Deciding before evaluating all approaches. First approach considered has anchoring advantage — gets more mental attention simply by being first. Evaluate all before comparing
- Equal advocacy for unequal approaches: Approach A scored 85, approach B scored 45? Spending equal time advocating both wastes effort, creates false equivalence
- Rubber-stamping: Going through evaluation process to justify decision already made. Test: could evaluation have changed outcome? If not, process was theater
- Threshold avoidance: Lowering confidence threshold to make decision easier rather than gathering information needed to meet appropriate threshold
- Ignoring losing side: Losing approach often contains warnings applying to winning one. Risks identified in approach B don't disappear just because approach A was chosen
See Also
build-consensus— multi-agent consensus model this skill adapts to single-agent reasoningforage-solutions— scouts solution space coherence evaluates; typically precedes this skillcoordinate-reasoning— manages information flow during multi-path evaluationcenter— establishes balanced baseline needed for unbiased evaluationmeditate— clears assumptions between evaluating different approaches
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。
