Chat History Extractor

Extract user inputs from Claude Code session history (~/.claude/projects/) and organize them into a project's .chats directory with daily markdown files.

When to Use This Skill

Use this skill when the user wants to:

• Extract session history from Claude Code sessions
• Document conversations for future reference
• Create instruction logs organized by date
• Export user inputs from .jsonl session files
• Maintain a changelog of Claude interactions

How It Works

Claude Code stores session data in ~/.claude/projects/{project-path-encoded}/ as .jsonl files. Each file contains messages with:

• type: "user" - User messages
• userType: "external" - External user (not system/agent)
• message.content - The actual message content

This skill extracts meaningful user instructions (filtering out system commands, tool results, and session continuations) and organizes them by date into .chats/{YYYYMMDD}.md files.

Instructions

Step 1: Identify the Project Session Directory

The session directory is derived from the current working directory path:

• Replace / with - in the path
• Prefix with -
• Location: ~/.claude/projects/{encoded-path}/

# Example: /Users/tchen/projects/tubi/titc
# Becomes: -Users-tchen-projects-tubi-titc
# Full path: ~/.claude/projects/-Users-tchen-projects-tubi-titc/

Step 2: Find Session Files by Date

List session files and filter by modification date:

# List all main session files (excluding agent-* files) for a specific date
ls -la ~/.claude/projects/{project-dir}/*.jsonl | grep "Dec 25" | grep -v agent-

Step 3: Extract User Inputs

Extract user messages from jsonl files using jq:

cat {session-file}.jsonl | jq -r '
  select(.type == "user" and .userType == "external" and (.isMeta | not)) |
  .message.content |
  if type == "string" then . else empty end
' | grep -v "^Caveat:" \
  | grep -v "^<command" \
  | grep -v "^<local-command" \
  | grep -v "^This session is being continued" \
  | grep -v "^<user-prompt-submit-hook>" \
  | grep -v "^Analysis:" \
  | grep -v "^$"

Step 4: Create/Update .chats Files

Create markdown files in .chats/ directory with format:

# Instructions

## {task title}

{user instruction}

## {another task title}

{another user instruction}

Step 5: Commit Changes

After creating/updating chat files, commit with:

git add .chats/*.md
git commit -m "docs(chats): add session history for {date range}"

File Format

Each .chats/{YYYYMMDD}.md file should:

• Start with # Instructions header
• Use ## for each major task/instruction
• Include the actual user input text
• Group related instructions under the same heading
• Preserve code blocks and formatting

Example Output

.chats/20251225.md:

# Instructions

## implement feature X

based on @specs/feature-x.md implement all phases entirely

commit the code and test

## fix bug Y

investigate why component Z is not working

use sub agents to analyze the issue in parallel

Filtering Rules

Include:

• Direct user instructions and requests
• Questions about the codebase
• Task specifications and requirements

Exclude:

• System commands (<command-name>, <local-command-stdout>)
• Session continuation messages
• Tool results and agent responses
• Hook notifications (<user-prompt-submit-hook>)
• Empty lines and caveat messages

Workflow Summary

1. Get current working directory
2. Compute encoded project path
3. Find session directory: ~/.claude/projects/{encoded-path}/
4. List sessions grouped by date
5. For each date with sessions:

- Extract user inputs from all sessions - Create .chats/{YYYYMMDD}.md - Organize inputs with descriptive headers

1. Create .chats/ directory if it doesn't exist
2. Optionally commit the changes

Helper Script

You can use this bash snippet to quickly find the project session directory:

# Get encoded project path
PROJECT_PATH=$(pwd | sed 's|/|-|g' | sed 's|^|/|' | sed 's|^/|-|')
SESSION_DIR="$HOME/.claude/projects/$PROJECT_PATH"

# Check if exists
if [ -d "$SESSION_DIR" ]; then
    echo "Session directory: $SESSION_DIR"
    echo "Sessions by date:"
    ls -la "$SESSION_DIR"/*.jsonl 2>/dev/null | grep -v agent- | awk '{print $6, $7}' | sort -u
else
    echo "No session directory found for this project"
fi

Notes

• Session files named agent-*.jsonl are sub-agent sessions and typically don't contain direct user input
• Main session files have UUID-style names (e.g., 01e78099-de0e-4424-845c-518638c8241e.jsonl)
• The .message.content field can be either a string (user text) or an array (tool results)
• Always verify the extracted content makes sense before committing