本文へ移動
cccskills
無料GitHub で公開

codexkit-document-summarizer

Summarize document-heavy inputs such as memos, policies, PDFs, research notes, proposals, and multi-document packets. Use when users need key points, obligations, questions, and action implications from documents.

インストール方法を見る

含まれるファイル(2)

  • SKILL.md3.7 KB
  • agents/openai.yaml180 B

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。

Document Summarizer

When to Use

  • Summarizing long documents that are not formal business reports.
  • Extracting key points from policies, proposals, PDFs, research notes, or document packets.
  • Preparing a quick readout before deciding whether to read the full document.
  • Comparing multiple documents for themes, contradictions, or action items.

Procedure

Step 1 - Identify Document Type

Classify the source: policy, proposal, memo, research, contract-like document, transcript, academic paper, operating document, or mixed packet.

Step 2 - Extract Core Content

Capture:

  • purpose
  • main claims or findings
  • dates, parties, obligations, or deadlines
  • decisions requested
  • open questions
  • risks or caveats

Step 3 - Separate Summary From Interpretation

Keep source-backed summary separate from analysis or recommendation. Mark anything inferred.

Step 4 - Produce Audience-Specific Output

Adapt format for executive, manager, student, legal/compliance reviewer, or team member.

Step 5 - Add Reading Guidance

State which sections deserve human review and why.

Inputs

InputRequiredFormat
Document text or excerptYesPasted text, extracted notes, OCR
Document typeRecommendedProposal, policy, memo, research, etc.
AudienceRecommendedExecutive, manager, student, team
Desired lengthOptional5 bullets, one page, detailed brief
Focus areaOptionalRisks, obligations, actions, decisions

Output

## Document Summary - [Document Name]

### One-Line Takeaway
[Core point]

### Key Points
- [Point backed by source]

### Actions / Decisions
| Item | Owner | Deadline | Source Note |
|------|-------|----------|-------------|

### Risks / Caveats
- [Risk]

### Needs Human Review
- [Section or issue]

Quality Criteria

  • The summary stays faithful to the document and does not invent missing facts.
  • Important dates, obligations, numbers, and decisions are preserved.
  • Source-backed summary is separated from interpretation.
  • The output states what was not reviewed or could not be verified.
  • High-risk documents are flagged for expert review.

Verification (4C)

CheckQuestion
CorrectnessAre all summarized claims traceable to the document text?
CompletenessAre key points, actions, deadlines, risks, and review needs captured?
Context-fitDoes the summary match the audience and intended decision?
ConsequenceWhat could be misread if the user relies only on the summary?

Edge Cases

  • OCR or extraction errors - Flag low confidence and avoid definitive claims.
  • Legal, financial, or compliance content - Summarize only and recommend expert review before acting.
  • Multiple documents conflict - Show a contradiction table instead of choosing a winner.
  • Very long packet - Summarize by section and list skipped or unread portions.

Examples

Prompt: "Summarize this vendor proposal for an operations manager. Focus on deliverables, pricing assumptions, risks, and questions for the vendor."

Good pattern: "The proposal assumes customer data export is available by June 1. This is a dependency, not a confirmed commitment."

Definition of Done

  • The user can decide whether to read the full document.
  • Important actions and risks are visible.
  • Uncertainty and review needs are explicit.

Changelog

  • v1.0.0 - Initial release

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

Design rigorous A/B test plans with hypothesis, sample size calculation, Minimum Detectable Effect (MDE), randomization strategy, and decision rules. Includes guardrail metrics and rollout playbook. Use when planning product experiments, conversion optimization, or data-driven feature decisions.

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。

hoavdc/CodexKit252026年10月8日 更新

Review REST and GraphQL API designs for consistency, usability, and best practices. Covers naming conventions, versioning strategy, error format, pagination, authentication patterns, and breaking change detection. Use when reviewing API specs, designing new APIs, or auditing existing endpoints.

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。

hoavdc/CodexKit252026年10月8日 更新

Write Architecture Decision Records (ADRs) following the Michael Nygard format. Captures context, options considered, decision rationale, and consequences. Use when making technology choices, framework selections, or any architectural decision that future developers need to understand.

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。

hoavdc/CodexKit252026年10月8日 更新

Assess organizational readiness for financial audits (internal or external). Map assertions to account balances, check evidence completeness, score readiness using a Red/Amber/Green framework, and generate a remediation timeline. Aligned with SOX, IFRS, and GAAP audit standards. Use before scheduled audits or when preparing for first-time 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。

hoavdc/CodexKit252026年10月8日 更新

Design safe recurring Codex automations with clear prompts, outputs, schedules, and gating 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。

hoavdc/CodexKit252026年10月8日 更新

Refine Product Backlog Items to meet INVEST criteria. Write User Stories with Acceptance Criteria in Given/When/Then format, estimate with Story Points, and flag dependencies. Use before sprint planning when backlog items need grooming. Do not use to prioritize the backlog — that is the Product Owner's decision.

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。

hoavdc/CodexKit252026年10月8日 更新

hoavdc のスキルをすべて見る

このスキルの問題を報告する