本文へ移動
cccskills
無料GitHub で公開

pitch-deck-reviewer

Reviews pitch decks and investor presentations. Reads slide content, evaluates narrative flow, problem/solution clarity, market sizing, competitive positioning, financial projections, team credibility, and ask clarity. Generates a scored pitch-review.md with slide-by-slide feedback, overall score, top improvements, investor objection predictions, and comparisons to successful decks. Use when reviewing fundraising materials, investor decks, or pitch presentations.

インストール方法を見る

含まれるファイル(4)

  • SKILL.md3.9 KB
  • references/evaluation-framework.md13.3 KB
  • references/objections-and-benchmarks.md2.5 KB
  • references/output-template.md4.8 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

Pitch Deck Reviewer

Review a pitch deck with VC-grade rigor and produce a scored, actionable pitch-review.md.

Contents

  • references/evaluation-framework.md -- The 8 scored dimensions, weights, red flags, scoring guides, design notes, and overall-score calculation.
  • references/objections-and-benchmarks.md -- Investor objection categories and the canonical successful-deck benchmarks.
  • references/output-template.md -- The exact pitch-review.md structure to generate.

Workflow

  1. Ingest the deck. Determine file type and location, then read every slide. Supported inputs: PDF (Read directly; for 10+ pages read in batches via the pages parameter), PowerPoint .pptx (invoke the pptx skill or Read), Google Slides links (WebFetch if public), images PNG/JPG (Read to inspect visually), markdown/text descriptions, and directories of slide images (Glob to enumerate, then Read each). For .pen files, state that they cannot be read directly.
  2. Build a slide inventory. Catalog each slide by number and primary content before any analysis. Identify company name, stage, and industry. Flag any deck over 20 slides as a concern itself; most successful decks run 10-15.
  3. Score the 8 dimensions. Apply references/evaluation-framework.md. Score each 1-10 with specific evidence from the deck, then compute the weighted overall score and assign the grade label.
  4. Write slide-by-slide feedback. Cover every slide with content summary, purpose, effectiveness, strengths, weaknesses, and recommendations.
  5. Predict investor objections. Generate at least 5 using the categories in references/objections-and-benchmarks.md, each with severity, trigger, suggested verbal response, and deck fix.
  6. Compare to benchmark decks. Reference at least 2 relevant decks from references/objections-and-benchmarks.md; name where this deck is stronger, weaker, and what to borrow.
  7. Rank the top 5 improvements by impact and list any missing slides or sections for this stage and industry.
  8. Assemble pitch-review.md per references/output-template.md. Write the executive summary last. Save it beside the source deck (or the working directory). Show the overall-score math.

Calibration

  • Verify any market-size, revenue, or growth claim with WebSearch before accepting it. Never fabricate metrics, comparisons, or statistics.
  • Reserve 9-10 on a dimension for work that sets a new standard. Give an overall below 3.0 only when the deck has almost no usable content.
  • Adjust for stage: weight pre-product decks toward problem/solution clarity, team, and market; relax traction and financials expectations and note the calibration.
  • Score absent major sections (e.g., missing financials or team) 1-2 and flag them in the executive summary and the missing-sections list. Review works in progress rather than refusing.
  • Evaluate intentionally unconventional decks (memo-style, demo-first, single-page) on their own terms while noting the divergence from convention.
  • For multiple versions, review the latest and note which changes improved the deck and which regressed.

Voice

  • Be direct and specific. Replace vague praise with concrete observations.
  • Pair every critique with a concrete fix.
  • Use investor framing ("This raises a question about...") over verdicts ("This is bad").
  • Acknowledge what works before identifying gaps. Define any jargon used.
  • Write for the founder, to help them improve.

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

Audit websites for accessibility issues and WCAG compliance. Use when checking accessibility, fixing a11y issues, or ensuring WCAG compliance.

日本語の概要は準備中です。原文の説明を表示しています。

OneWave-AI/claude-skills3392026年10月2日 更新

Deploy a 2-layer parallel agent hierarchy for large, parallelizable work — big refactors, multi-file migrations, codebase-wide audits, bulk generation. A top-tier commander (Fable or Opus) orchestrates the swarms; the user picks a power level (Max Power / Heavy / Balanced / Economy) that sets the Opus/Sonnet/Haiku model mix per layer. Layer 1 is 3-50+ specialist agents, each with its own full context window; Layer 2 is 2+ sub-agents per member. Includes git safety, tiered sizing, a pre-deploy gate, phantom-completion checks, and multi-wave follow-up.

日本語の概要は準備中です。原文の説明を表示しています。

OneWave-AI/claude-skills3392026年10月2日 更新

Use to test whether an AI agent, prompt, or skill actually works before or after it ships — build a small eval set, score runs, and catch regressions when the prompt or model changes. Trigger when the user asks whether a prompt is good, wants to compare models or versions, sees inconsistent agent output, or is about to put an agent in front of customers.

日本語の概要は準備中です。原文の説明を表示しています。

OneWave-AI/claude-skills3392026年10月2日 更新

Deploys swarms of sub-agents for massive parallel data processing tasks. Unlike agent-army (which is for code changes), this is for DATA tasks -- processing 1000 documents, analyzing datasets, bulk content generation. Configurable swarm size, task distribution, result aggregation, progress tracking, and error recovery.

日本語の概要は準備中です。原文の説明を表示しています。

OneWave-AI/claude-skills3392026年10月2日 更新

Designs and deploys custom agent teams for specific business workflows. Interactive discovery of business processes, then generates complete team configurations with specialized agent roles, tool access, communication protocols, and handoff rules.

日本語の概要は準備中です。原文の説明を表示しています。

OneWave-AI/claude-skills3392026年10月2日 更新

Agent-to-Agent (A2A) communication protocol. Connect two or more Claude agents that pass messages, share context, delegate tasks, and collaborate. Implements structured handoffs, shared memory, and multi-agent conversations.

日本語の概要は準備中です。原文の説明を表示しています。

OneWave-AI/claude-skills3392026年10月2日 更新

OneWave-AI のスキルをすべて見る

このスキルの問題を報告する