ai-conversation-extractor - SKILL.md Agent Skill

name: ai-conversation-extractor description: Extract valuable insights, decisions, patterns, and ideas from AI conversation exports (ChatGPT, Claude, Claude Code) into Obsidian-compatible daily notes and atomic notes. Use when user wants to process conversation exports, build a second brain from AI chats, create daily notes from past sessions, or extract knowledge worth preserving from AI interactions.

AI Conversation Extractor

A second brain is not a storage system. It is a thinking system.

The goal is not to archive conversations—it's to distill them into knowledge that makes you think better, see patterns you missed, and strengthen your ideas over time.

Core Philosophy

"When you revisit something in 6 months, you should not be confronted with a wall of text, but with the signal inside the noise."

Every extraction should pass this test: Would future-you, who has forgotten the context, find this immediately useful?

Configuration

Set these environment variables to customize paths (or accept the defaults):

Variable	Default	Description
`OBSIDIAN_VAULT_PATH`	`$HOME/Documents/Obsidian Vault`	Path to your Obsidian vault
`CLAUDE_EXPORT_PATH`	`<plugin>/data/exports/claude/conversations.json`	Claude.ai export file
`CHATGPT_EXPORT_PATH`	`<plugin>/data/exports/chatgpt/conversations.json`	ChatGPT export file

Output Structure:

${OBSIDIAN_VAULT_PATH}/
└── Claude Sessions/
    ├── Daily Notes/           # Day-by-day summaries
    ├── Insights/              # Atomic insight notes
    ├── Decisions/             # Decision records with rationale
    ├── Patterns/              # Reusable techniques
    ├── Questions/             # Open loops to revisit
    ├── Models/                # Mental frameworks
    └── Topics/                # Deep dives on subjects

The Extraction Categories

Category	What to Look For	Note Type
Insights	Aha moments, shifted understanding, surprising connections	`Insights/`
Decisions	Choices made with rationale, rejected alternatives	`Decisions/`
Patterns	Techniques that worked, reusable approaches	`Patterns/`
Mental Models	Frameworks for thinking, conceptual models	`Models/`
Questions	Unanswered queries worth returning to	`Questions/`
Vocabulary	New terms, precise definitions, distinctions	Inline in relevant notes
Narrative	What happened, evolution of thinking	Daily note

The Three Tests

Before extracting anything, apply these tests:

1. The Resonance Test

Did this make you pause, reconsider, or connect to something you care about?

Not everything in a conversation resonates
Some days have nothing worth extracting—that's fine
Trust your gut: if it doesn't feel significant, it probably isn't

2. The Atomicity Test

Can you easily give it a clear, simple title?

If you struggle to name it, you haven't found the essence yet
One idea per note—but capture the entirety of that idea
The note should be understandable at a glance

3. The Production Test

Does this connect to something you're building, thinking about, or might use?

Organize knowledge around what you're producing, not just consuming
Ask: "How might future-me use this?"
Ideas that connect to active projects/interests are more valuable

Distillation, Not Capture

The note should BE the signal, not a pointer to the original.

Capture	Distillation
"Interesting discussion about X"	The actual insight, in your own words
Link to conversation	Self-contained understanding
Raw quotes	Paraphrased essence
"See context in chat"	Context embedded in note

When you extract, transform:

Read the valuable section
Close the source
Write what you understood in your own words
If you can't write it without looking, you didn't understand it

What NOT to Extract

Troubleshooting that failed - Unless the failure itself is the lesson
Process chatter - "Let me search...", "I'll check that..."
Easily re-derived info - Facts you can Google in 10 seconds
Full transcripts - That's archiving, not thinking
Things that don't resonate - If you're forcing it, skip it

Processing Workflow

Step 1: Load and Scan

Read through the day's conversations quickly. Don't extract yet—just notice what stands out.

Step 2: Apply the Three Tests

For anything that caught your attention:

Resonance: Does it matter?
Atomicity: Can you name it simply?
Production: Does it connect to something you're building?

If it passes all three -> extract. If not -> let it go.

Step 3: Distill into Notes

For each extraction:

Write in your own words (not copy-paste)
Make it self-contained (no context needed)
Give it a clear, simple title
Add relevant tags and links

Step 4: Create Daily Note

Even if no atomic notes were extracted:

Summarize what was discussed
Note the themes (even without deep insights)
Link to any extracted notes
Capture open questions or threads

Note Formats

Daily Note

---
title: Claude Session - YYYY-MM-DD
date: YYYY-MM-DD
tags:
  - claude-session
  - YYYY-MM
  - [project-tags]
type: daily-note
---

# Claude Session - YYYY-MM-DD

## Summary
[2-3 sentences: What was the day about? What themes emerged?]

## Extractions
- [[Note Name]] - why it matters
- [[Note Name]] - why it matters

## Threads
[What's still being thought about? What might develop?]

## Raw Notes
[Brief context, observations, anything worth remembering that isn't an extraction]

Insight Note

---
title: Insight - [Clear Name]
date: YYYY-MM-DD
tags:
  - insight
  - [domain]
type: insight
---

# [The Insight in One Line]

[The insight fully expressed in 1-3 sentences. Should be immediately clear.]

## Why This Matters
[What does this change? What does it enable?]

## Context
[Brief: what led to this realization]

## Related
- [[Other Notes]]

Decision Note

---
title: Decision - [Clear Name]
date: YYYY-MM-DD
tags:
  - decision
  - [domain]
type: decision
status: active
---

# Decision: [What Was Decided]

## Context
[Why did this decision come up?]

## Choice
[What was decided—be specific]

## Rationale
[Why this over alternatives]

## Alternatives Rejected
- [Option] - why not

## Related
- [[Other Notes]]

Pattern Note

---
title: Pattern - [Clear Name]
date: YYYY-MM-DD
tags:
  - pattern
  - [domain]
type: pattern
---

# Pattern: [Name]

## When to Use
[What situation triggers this pattern?]

## The Pattern
[The approach, clearly described]

## Example
[Concrete instance of this pattern in action]

## Related
- [[Other Notes]]

Model Note

---
title: Model - [Clear Name]
date: YYYY-MM-DD
tags:
  - model
  - [domain]
type: model
---

# Model: [Name]

## The Model
[The framework or mental model, clearly explained]

## When It Applies
[What situations does this help you think about?]

## How to Use It
[Practical application]

## Related
- [[Other Notes]]

Question Note

---
title: Question - [The Question]
date: YYYY-MM-DD
tags:
  - question
  - [domain]
type: question
status: open
---

# [The Question]

## Why This Matters
[Why is this worth returning to?]

## Current Thinking
[What do you currently believe? What's uncertain?]

## Related
- [[Other Notes]]

Quality Checklist

Before finalizing:

Each note passes all three tests (resonance, atomicity, production)
Notes are distilled, not captured (in your own words, self-contained)
Titles are clear and simple (the "easy to name" test)
Future-you could understand this without context
Links connect related ideas
Nothing forced—if it didn't resonate, it's not there

Remember

Some days have nothing. That's normal. Don't force extractions.
Patterns emerge over time. The value isn't in one note—it's in the accumulation.
Distillation > Capture. The note should be the knowledge, not a pointer to it.
Trust resonance. If something made you pause, there's probably something there.