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 fr

SKILL.md

GitHub

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

Input Required Format
Document text or excerpt Yes Pasted text, extracted notes, OCR
Document type Recommended Proposal, policy, memo, research, etc.
Audience Recommended Executive, manager, student, team
Desired length Optional 5 bullets, one page, detailed brief
Focus area Optional Risks, 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)

Check Question
Correctness Are all summarized claims traceable to the document text?
Completeness Are key points, actions, deadlines, risks, and review needs captured?
Context-fit Does the summary match the audience and intended decision?
Consequence What 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