について
このスキルは、指定された自社製品と競合製品について、公開情報から自律的に調査を行い、競争力のあるバトルカードを作成します。主張の根拠となる情報源を明示し、各主張を「事実」「推論」「仮定」として分類した、簡潔で現場即応型のドキュメントを生成します。特定の商談に向けて、根拠に基づいた説得力のある議論を販売チームに迅速に提供するためにご利用ください。
クイックインストール
Claude Code
推奨npx skills add deanpeters/Product-Manager-Skills -a claude-code/plugin add https://github.com/deanpeters/Product-Manager-Skillsgit clone https://github.com/deanpeters/Product-Manager-Skills.git ~/.claude/skills/battle-card-builder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
Battle Card Builder
Purpose
Build a field-action competitive battle card from public evidence: use or gather evidence → search plan (if gathering) → draft card with per-claim labels → appendix → next-step options. A battle card is the artifact a rep opens mid-deal, not a research report — it must fit thirty seconds, every claim must survive being said out loud to a hostile audience, and every claim therefore carries a source URL, a date, and a Fact/Inference/Assumption label. Most battle cards in the wild are unlabeled inference; this skill exists because of what it costs a rep when one of those turns out wrong.
Input
Works best with: your product and its primary differentiation, and the one competitor this
card targets.
Also useful: the deal context (segment, buyer, common objections or the loss reason that triggered
this card), and any competitive-research-snapshot or
competitive-intel-watch output in session — the skill uses
existing evidence and searches only for gaps and anything older than one quarter.
Input supplied inline with the invocation — text after the skill name, a pasted context dump, or an
appended ARGUMENTS: line — counts as answers already given. Use it against the question budget;
don't re-ask.
Arriving empty-handed? That works too. The skill opens with at most 3 questions (product + competitor, segment + buyer, triggering objection) and proceeds on labeled assumptions if they go unanswered.
Example invocation: Build a battle card: our workflow platform vs [Competitor A], mid-market ops buyers — triggered by three straight losses on "their integrations are deeper."
Key Concepts
- Governing protocol: honors the
autonomous-investigationcontract — question budget of 3, search-plan gate, Fact/Inference/Assumption labels, Just Enough Mode, stable schema, 4-option Final Step. - Discipline mix: SIGINT (pricing pages, site diffs — the freshest layer, what keeps cards from
going stale) + OSINT (review mining for objection handling) + HUMINT (win/loss ground truth when
you have it) — see
intelligence-collection-disciplines. - Field-action criterion. Every element earns its place by being usable mid-conversation. Depth goes in the appendix; the card fits a rep's thirty seconds.
- Trap questions are evidence work. A trap question is only safe to ask when you know the documented answer — offense is built from citations, not bravado. "Never ask what you cannot evidence" is the section's law.
- The "Do Not Say" section is protection. Claims you can't evidence, or that invite a counter, are listed explicitly — the card protects the rep from the team's own folklore.
- Two modes, one artifact. This skill builds the card from the world's evidence (investigation); a facilitated workshop builds it from your team's evidence (win/loss knowledge, deal experience). The mode choice depends on where the evidence lives — bring rich internal win/loss knowledge to a facilitated session instead, and use this skill when the public record is the stronger source.
- Do-not-invent list: features, pricing, market share, customer wins, roadmap items, quotes.
Application
- Check session for existing evidence — a competitive snapshot or watch report. If present, use it as the evidence base; search only for gaps and freshness (older than one quarter).
- Credit inline context, then ask only the unanswered questions (max 3):
- Your product, and the one competitor this card targets?
- What segment and buyer do these deals involve?
- What objection or loss reason triggered this card? If unanswered, proceed with labeled assumptions.
- If researching fresh, show the 3-bullet search plan — what you'll check (pricing pages, release notes, reviews, customer stories, comparison content), source types, how facts will be separated from inference. Continue unless revised.
- Draft the card in the schema below exactly — cards are diffable across runs, and the watch skill's update flags name these sections.
Card schema (do not reorder)
# Battle Card: [Your Product] vs [Competitor]
**As-of date:** | **Deal context:** [segment, buyer]
## 1. Thirty-Second Read
- **They win when:** [1-2 bullets, labeled]
- **We win when:** [1-2 bullets, labeled]
- **The one thing to say:** [a single defensible sentence]
## 2. Say This
- [Talking point] — [evidence: URL, date, label]
- [Max 4; every one sourced]
## 3. Ask This (trap questions)
- "[Question whose documented answer favors us]" — [the documented answer: URL, label]
- [Max 3; never ask what you cannot evidence]
## 4. Watch Out For
- [Their strength a rep should not walk into] — [URL, label]
- [Their likely counter to our pitch] — [Inference, basis]
## 5. Pricing & Packaging Snapshot
- [Their tiers/pricing as published] — [URL, as-of date]
- [Where "contact sales" hides the number] — [labeled]
## 6. Do Not Say
- [Claims we cannot evidence, or that invite a counter]
## Appendix: Evidence Table
| Claim | Label | Source | Date |
|---|---|---|---|
### Assumptions to Validate
- [Assumption 1] / [Assumption 2] / [Assumption 3]
A copy/paste fill-in version of this schema, with quality checks, lives in template.md.
Final Step (offer exactly 4 options)
- Pressure-test the card: role-play the competitor's rep against it
- Build the card for a second competitor
- Set up
competitive-intel-watchto keep this card current - Compress to a one-screen mobile version for the field
Accept 1, 2, 3, 4, 1 and 3, Verbose Mode, or a custom path.
Examples
A trap question built correctly (fictional):
Ask this: "How does your migration handle custom fields over the 200-field limit?" — documented answer: their own docs state custom-field migration caps at 200 with manual re-entry beyond it — Fact (docs page, checked Jul 2026). Safe to ask: the answer is in their documentation, so the rep isn't bluffing — they're citing.
A "Do Not Say" entry earning its keep:
- Don't say "they can't do SSO on the starter tier" — they shipped it in June (release notes); a rep repeating our old card walks into a correction that costs the room's trust.
- Don't lead with "we're cheaper" — true only below 50 seats (their pricing, ours); above that the math flips, and enterprise buyers check.
See examples/sample.md for a complete worked card (fictional FSM-software
market) rebuilt from the competitive-intel-watch example's update flags — the full chain landing
in a field artifact. examples/sample-industrial.md builds the
industrial version, where the channel partner is a card audience and pricing is opaque.
Common Pitfalls
- The unlabeled card. Talking points without sources are folklore with formatting. The label is what tells a rep how hard they can lean on a claim when the buyer pushes back.
- The research-report card. Two pages of landscape context a rep will never read mid-deal. Thirty seconds is the budget; the appendix exists for everything else.
- Trap questions on hope. Asking a question whose answer you haven't documented hands the competitor a stage. If the evidence isn't in the appendix table, the question doesn't ship.
- Quote invention. A fabricated customer quote or paraphrased review presented as verbatim is the fastest way to destroy the card's credibility — and it's on the do-not-invent list for exactly that reason.
- The immortal card. Cards rot at SIGINT speed — pricing pages and release notes change monthly. A card without an as-of date, or without a watch feeding it update flags, is a liability wearing a logo.
- Skipping "Do Not Say." The section that protects reps from the team's own stale folklore is the one most often omitted — usually because nobody wants to write down which beloved claims are dead.
References
autonomous-investigation(Workflow) — the governing protocolintelligence-collection-disciplines(Component) — discipline sources; SIGINT keeps this card freshcompetitive-research-snapshot(Workflow) — upstream evidence basecompetitive-intel-watch(Workflow) — flags when this card's sections go stalepositioning-statement(Component) — the strategic layer the card's talking points should agree with- Adapted from
market-intelligence/battle-card-builder-prompt.mdin thehttps://github.com/deanpeters/product-manager-promptsrepo.
GitHub リポジトリ
よくある質問
battle-card-builder Skillとは何ですか?
battle-card-builder はdeanpeters が作成した Claude Skillです。Skillは、Claudeが必要に応じて読み込む指示とリソースをまとめ、追加の指示なしで battle-card-builder に関連するタスクを実行できるようにします。
battle-card-builder をインストールするには?
このページのインストールコマンドを使用してください。battle-card-builder をプラグインとして Claude Code に追加するか、リポジトリを skills ディレクトリにクローンし、Claudeを再起動してSkillを読み込みます。
battle-card-builder はどのカテゴリに属しますか?
battle-card-builder は メタ カテゴリに属します。
battle-card-builder は無料で利用できますか?
はい。battle-card-builder は AIMCP に掲載されており、無料でインストールできます。
関連スキル
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。
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。
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。
