Chat History Summarizer

Overview

This skill extracts user prompts and Claude's actions from Claude Code conversation logs, generating a concise, structured summary document. It's useful for:

• Documenting what was accomplished in a session
• Creating a record for future reference or handoff
• Reviewing methodology and approach taken
• Learning from past interactions

Prerequisites

Install the claude-conversation-extractor tool:

pip install claude-conversation-extractor

Workflow

Step 1: List Available Sessions

claude-extract --list

This shows all available sessions with:

• Session ID
• Project folder
• Modification date
• Message count
• Size
• Preview of first message

Step 2: Search for Specific Session (Optional)

claude-extract --search "keyword"

Search by keyword to find relevant sessions.

Step 3: Export the Session

# Basic export
claude-extract --extract <session_number>

# Detailed export (includes tool outputs)
claude-extract --extract <session_number> --detailed

# Export as HTML
claude-extract --format html --extract <session_number>

Default output location: ~/Desktop/Claude logs/

Step 4: Generate Summary

Read the exported file and create a structured summary with:

1. Session Metadata

   • Session ID
   • Date
   • Total message count

2. User Inputs & Claude Actions (organized by interaction rounds)

   • Each user prompt (quoted)
   • Concise summary of Claude's actions (tools used, files created/modified)

3. Generated Files List (if applicable)

   • File paths and descriptions

4. Tool Usage Statistics (optional)

   • Tools used and frequency

5. Key Outcomes (optional)

   • Main deliverables or conclusions

Output Template

# [Session Title/Topic]

**Session ID:** [id]
**Date:** [date]
**Total Messages:** [count]

---

## User Inputs & Claude Actions Summary

### Round 1: [Brief Title]

**User Input:**
> [Exact user prompt, quoted]

**Claude Actions:**
1. [Action 1 - tool used, brief description]
2. [Action 2 - tool used, brief description]
3. [Files created/modified if any]

---

### Round 2: [Brief Title]
...

---

## Generated Files

| File | Path | Description |
|------|------|-------------|
| [name] | [path] | [brief description] |

---

## Tool Usage Statistics

| Tool | Count | Purpose |
|------|-------|---------|
| [tool] | [n] | [main use] |

---

*Summary extracted from conversation log on [current date]*

Key Principles

1. Quote user inputs exactly - Preserve the user's original wording
2. Be concise on Claude actions - Focus on what was done, not how
3. Group by interaction rounds - Each user input starts a new round
4. Highlight key deliverables - Files created, reports generated, etc.
5. Skip warmup/system messages - Only include substantive exchanges

Extracting from Raw Logs

When reading exported markdown files:

1. Look for ## 👤 User markers to identify user inputs
2. Look for ## 🤖 Claude markers for Claude responses
3. Look for 🔧 Using tool: to identify tool usage
4. Ignore:
   • System reminders (<system-reminder>)
   • Warmup messages
   • Agent internal messages
   • Tool input/output details (unless specifically relevant)

Example Search Patterns

# Find user messages
grep -n "## 👤 User" exported-log.md

# Find tool usage
grep -n "🔧 Using tool:" exported-log.md

# Find Write operations (files created)
grep -n "Using tool: Write" exported-log.md

Tips

• For very long sessions, focus on key milestones rather than every exchange
• Group related exchanges into logical "rounds" even if interrupted
• Note when context was compacted or session was continued
• Include any user choices/preferences that shaped the outcome