progressive-summarization

Progressive Summarization

Apply Tiago Forte's Progressive Summarization technique to transform raw captures into scannable, layered notes. This skill takes raw input and produces distilled output ready for a Second Brain.

The Four Layers

Layer 1: Original Excerpts

The raw text the user provides -- quotes, highlights, passages from source material. This layer already exists when the user pastes content.

Layer 2: Bold the Main Points

Read through the excerpts and bold the key sentences, phrases, and keywords that capture the core ideas. Target roughly 10-20% of the original text.

What to bold:

• Sentences that capture a complete, standalone insight
• Surprising claims or counterintuitive findings
• Specific data points, statistics, or examples
• Actionable advice or concrete recommendations
• Definitions of key concepts

What NOT to bold:

• Context-setting sentences ("In this article, we explore...")
• Redundant explanations of the same point
• Transitional phrases
• General statements everyone already knows

Layer 3: Highlight the Best of the Bold

Look ONLY at the bolded text and ==highlight== (or mark with a different format) the absolute best, most insightful sentences. This creates an ultra-scannable summary. Target roughly 10-20% of the bolded text.

Selection criteria:

• If the user could only read 3 sentences, which would give the most value?
• Which bolded passages would be useful across multiple projects?
• Which statements are most surprising or counterintuitive?

Layer 4: Executive Summary

Write a brief summary in the user's own voice at the very top using bullet points. This is the "TL;DR" -- 2-5 bullet points that capture the essence.

Executive summary rules:

• Write in first person or neutral instructional voice
• Use the user's vocabulary, not academic jargon
• Each bullet should be a complete, actionable insight
• Include the "so what" -- why this matters

Process

Step 1: Receive Input

Accept any of these input types:

• Raw article or blog post text
• Meeting transcript or voice memo transcription
• Book highlights or Kindle exports
• Research paper excerpts
• Brain dump or freeform notes
• Multiple related notes for synthesis

Ask for the input if not provided:

"Paste the text to distill. This can be an article, transcript, highlights, or any raw notes. I'll apply Progressive Summarization layers to make it scannable."

Step 2: Ask for Context

Before distilling, ask one question:

"What project or goal is this related to? Knowing this helps me prioritize the most relevant points. (If it's general reference, just say so.)"

This context shapes which points get bolded and highlighted -- the same article distills differently for different purposes.

Step 3: Apply Layers

Apply all requested layers in a single output. Default to applying Layers 2-4 unless the user requests specific layers.

Step 4: Format Output

Ask the user's preferred output format if not already known:

Format	Output Style
Markdown (default)	**bold** for L2, ==highlight== for L3, bullet summary for L4
Notion	Bold for L2, highlight color for L3, callout block for L4
Obsidian	**bold** for L2, ==highlight== for L3, admonition for L4
Plain text	[KEY] prefix for L2, [!!!] prefix for L3, summary block for L4

Output Template

## Executive Summary (Layer 4)
- [Insight 1 -- actionable takeaway]
- [Insight 2 -- actionable takeaway]
- [Insight 3 -- actionable takeaway]

---

## Distilled Note

[Original text with **Layer 2 bolding** applied throughout
and ==Layer 3 highlights== on the most critical passages]

Multi-Note Synthesis

When the user provides multiple notes on the same topic:

1. Read all notes and identify common themes
2. Produce a single synthesized note that:
   • Merges overlapping insights (don't repeat the same point from different sources)
   • Preserves unique perspectives from each source
   • Notes areas of agreement and contradiction
   • Attributes key claims to their sources
3. Apply Layers 2-4 to the synthesized result
4. Add a "Sources" section at the bottom

Key Principles

• Distill opportunistically. Don't batch-summarize everything upfront. Add layers when the user is actually reviewing a note for a specific purpose.
• Context matters. The same article distills differently depending on the active project.
• Preserve the original. Never delete source text -- only add formatting layers on top.
• Favor the user's own words at Layer 4. Their phrasing is faster to process.
• AI augments, not replaces. The act of distilling is itself valuable for learning. Explain what was prioritized and why so the user builds their own judgment.

Reference Files

• references/layer-examples.md -- Concrete before/after examples of all 4 layers applied to articles and meeting transcripts, plus format variations for Obsidian/Notion/plain text
• ../para-organize/references/code_framework.md -- Full CODE framework including distillation principles