Conversation Logging

Conversation Logging Skill

Automatically log all Claude Code interactions to structured markdown files using global hooks. Enables conversation replay, debugging, and context sharing between sessions.

What Gets Logged

• User prompts: Every prompt submitted to Claude
• Tool executions with context: Tool name plus specific details:
  • Read/Write/Edit: File paths and content previews
  • Bash: Commands and output (first 10 lines)
  • Glob/Grep: Search patterns
  • TodoWrite: Task lists with status indicators
  • Task: Subagent type and prompt preview
  • WebSearch/WebFetch: Queries and URLs
  • AskUserQuestion: Questions asked
• Session metadata: Working directory, timestamps, session IDs
• Claude responses: Excerpts from Claude's actual responses (last 500 chars)

Each Claude instance gets its own log file to prevent conflicts when running multiple sessions simultaneously.

Installation

Step 1: Copy Hook Script

The logging script is provided in this skill's scripts/ directory. Copy it to your hooks directory:

# Create hooks directory if it doesn't exist
mkdir -p ~/.claude/hooks

# Copy the script from this skill
cp /path/to/my-skills/skills/conversation-logging/scripts/log-conversation.sh ~/.claude/hooks/

# Make it executable
chmod +x ~/.claude/hooks/log-conversation.sh

Or create it manually at ~/.claude/hooks/log-conversation.sh:

#!/bin/bash
# Global hook to log Claude Code conversation events
# Handles multiple concurrent Claude instances by using unique session IDs

LOG_DIR="$HOME/.claude/conversation-logs"
TIMESTAMP=$(date '+%Y-%m-%d %H:%M:%S')
DATE=$(date '+%Y-%m-%d')

mkdir -p "$LOG_DIR"

# Read JSON input from stdin
INPUT=$(cat)

# Extract key fields using jq if available
if command -v jq &> /dev/null; then
    EVENT=$(echo "$INPUT" | jq -r '.hook_event_name // "unknown"')
    CWD=$(echo "$INPUT" | jq -r '.cwd // "unknown"')
    TOOL=$(echo "$INPUT" | jq -r '.tool_name // ""')
    PROMPT=$(echo "$INPUT" | jq -r '.prompt // ""')
    SESSION_ID=$(echo "$INPUT" | jq -r '.session_id // ""')
else
    EVENT="unknown"
    CWD=$(pwd)
    SESSION_ID=""
fi

# Create unique log file per session
# Format: YYYY-MM-DD-session-XXXXX.md
if [ -n "$SESSION_ID" ]; then
    # Extract first 8 chars of session ID for readability
    SHORT_SESSION="${SESSION_ID:0:8}"
    LOG_FILE="$LOG_DIR/${DATE}-session-${SHORT_SESSION}.md"
else
    # Fallback: use PID if session_id not available
    LOG_FILE="$LOG_DIR/${DATE}-pid-$$.md"
fi

# Initialize log file with header if it doesn't exist
if [ ! -f "$LOG_FILE" ]; then
    cat > "$LOG_FILE" <<EOF
# Claude Code Conversation Log
**Date:** $DATE
**Session ID:** ${SESSION_ID:-unknown}
**Started:** $TIMESTAMP

---

EOF
fi

# Log based on event type
case "$EVENT" in
    UserPromptSubmit)
        cat >> "$LOG_FILE" <<EOF

## [$TIMESTAMP] User Prompt
**Working Directory:** \`$CWD\`

\`\`\`
$PROMPT
\`\`\`

EOF
        ;;

    PostToolUse)
        if [ -n "$TOOL" ]; then
            echo "### [$TIMESTAMP] Tool: \`$TOOL\`" >> "$LOG_FILE"
            echo "" >> "$LOG_FILE"
        fi
        ;;

    Stop)
        echo "### [$TIMESTAMP] Response Complete" >> "$LOG_FILE"
        echo "" >> "$LOG_FILE"
        ;;
esac

exit 0

Then make it executable:

chmod +x ~/.claude/hooks/log-conversation.sh

Step 2: Configure Global Hooks

Add hooks to ~/.claude/settings.json:

{
  "hooks": {
    "UserPromptSubmit": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "~/.claude/hooks/log-conversation.sh"
          }
        ]
      }
    ],
    "PostToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "~/.claude/hooks/log-conversation.sh"
          }
        ]
      }
    ],
    "Stop": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "~/.claude/hooks/log-conversation.sh"
          }
        ]
      }
    ]
  }
}

If you already have other settings in your settings.json, merge the hooks section carefully.

Step 3: Verify Installation

Run any Claude Code command and check the logs:

# List all conversation logs
ls -lh ~/.claude/conversation-logs/

# View the most recent log
cat $(ls -t ~/.claude/conversation-logs/*.md | head -1)

Note: Each Claude session creates a unique log file with format YYYY-MM-DD-session-XXXXX.md to prevent conflicts between concurrent instances.

Usage

Viewing Conversation Logs

Most recent session:

cat $(ls -t ~/.claude/conversation-logs/*.md | head -1)

All sessions from today:

ls -1 ~/.claude/conversation-logs/$(date +%Y-%m-%d)-*.md

View specific session:

# List today's sessions to find the one you want
ls ~/.claude/conversation-logs/$(date +%Y-%m-%d)-*.md

# Then view it
cat ~/.claude/conversation-logs/2026-01-05-session-a1b2c3d4.md

Recent logs (last 5 sessions):

ls -lt ~/.claude/conversation-logs/ | head -6

Search across all logs:

grep -r "search term" ~/.claude/conversation-logs/

Providing Context to Claude

When you want Claude to understand a previous conversation:

Option 1: Reference most recent session

Read my last conversation log:
@$(ls -t ~/.claude/conversation-logs/*.md | head -1)

What were we working on?

Option 2: Reference specific session

Read my conversation from this morning:
@~/.claude/conversation-logs/2026-01-04-session-a1b2c3d4.md

Continue where we left off.

Option 3: Copy relevant sections

Here's what happened in my last session:

[paste relevant log sections]

Continue from where we left off.

Option 4: Search and reference

# Find the conversation about a topic
grep -l "GitHub Issues" ~/.claude/conversation-logs/*.md

Then reference that file with @ in Claude Code.

Debugging Failed Sessions

When Claude encounters errors or you need to replay a session:

# Find failed tool executions in most recent session
grep -A 5 "Tool: Bash" $(ls -t ~/.claude/conversation-logs/*.md | head -1)

# See what prompts led to errors across all today's sessions
grep -B 10 "error" ~/.claude/conversation-logs/$(date +%Y-%m-%d)-*.md

# Find specific session with errors
grep -l "error" ~/.claude/conversation-logs/*.md

Resuming Work Across Sessions

Use conversation logs to pick up where you left off:

# View most recent session
cat $(ls -t ~/.claude/conversation-logs/*.md | head -1)

# View yesterday's sessions
ls ~/.claude/conversation-logs/$(date -d yesterday +%Y-%m-%d)-*.md

# Share with Claude
claude -p "Read @$(ls -t ~/.claude/conversation-logs/*.md | head -1) and summarize what we were working on"

Pruning Old Logs

Keep your logs directory clean by removing old logs:

Interactive prune (with confirmation):

# Delete logs older than 14 days (default)
~/.claude/hooks/prune-logs.sh

# Delete logs older than 30 days
~/.claude/hooks/prune-logs.sh 30

# Delete logs older than 7 days
~/.claude/hooks/prune-logs.sh 7

Dry run (preview without deleting):

~/.claude/hooks/prune-logs.sh 14 --dry-run

What it does:

• Shows list of logs to be deleted with dates
• Displays total size to be freed
• Asks for confirmation before deleting
• Supports dry-run mode to preview

Setup the prune script:

# Copy from skill
cp /path/to/my-skills/skills/conversation-logging/scripts/prune-logs.sh ~/.claude/hooks/

# Make executable
chmod +x ~/.claude/hooks/prune-logs.sh

# Create an alias (optional)
echo "alias prune-claude-logs='~/.claude/hooks/prune-logs.sh'" >> ~/.bashrc
source ~/.bashrc

# Now you can just run:
prune-claude-logs

Automated cleanup with cron:

# Add to crontab to run monthly
crontab -e

# Add this line to delete logs older than 30 days on the 1st of each month
0 0 1 * * $HOME/.claude/hooks/prune-logs.sh 30 <<< "y"

Log Format

Each log file has a unique name per session: YYYY-MM-DD-session-XXXXXXXX.md

Log contents are organized chronologically with rich context:

# Claude Code Conversation Log
**Date:** 2026-01-05
**Session ID:** a1b2c3d4-5678-90ab-cdef-1234567890ab
**Started:** 2026-01-05 14:23:10

---

## [2026-01-05 14:23:15] User Prompt
**Working Directory:** `/home/user/project`

Implement the login feature

### [2026-01-05 14:23:16] Tool: `Read` - `/home/user/project/src/auth.js`

### [2026-01-05 14:23:17] Tool: `Bash`
```bash
git status

On branch main
Changes not staged for commit:
  modified:   src/auth.js

[2026-01-05 14:23:18] Tool: Write - /home/user/project/src/login.js

import { authenticate } from './auth.js';

export async function login(username, password) {
  ret...

[2026-01-05 14:23:20] Tool: TodoWrite - 3 tasks

• Create login function
• [>] Add error handling
• Write tests

[2026-01-05 14:23:25] Claude Response

I've implemented the login feature with proper authentication. The new login.js file handles user authentication and includes error handling...

[2026-01-05 14:25:30] User Prompt

...

**File naming:**
- Session-based: `2026-01-05-session-a1b2c3d4.md` (first 8 chars of session ID)
- Fallback (if no session ID): `2026-01-05-pid-12345.md` (process ID)

## Customization

### Change Log Location

Edit the script's `LOG_DIR` variable:

```bash
LOG_DIR="$HOME/my-custom-logs"

Add More Details

Extend the script to log additional fields from the hook input:

# Log model info
MODEL=$(echo "$INPUT" | jq -r '.model // "unknown"')

# Log session ID
SESSION=$(echo "$INPUT" | jq -r '.session_id // "unknown"')

Filter Specific Tools

Only log certain tools:

PostToolUse)
    # Only log Read, Write, Edit tools
    if [[ "$TOOL" =~ ^(Read|Write|Edit)$ ]]; then
        echo "### [$TIMESTAMP] Tool: $TOOL" >> "$DATE_FILE"
    fi
    ;;

JSON Logs Instead of Markdown

Replace the logging logic to output JSON:

echo "$INPUT" | jq '.' >> "$LOG_DIR/$(date +%Y-%m-%d).jsonl"

Troubleshooting

Logs not being created?

• Check hook script is executable: ls -l ~/.claude/hooks/log-conversation.sh
• Verify settings.json syntax: cat ~/.claude/settings.json | jq .
• Check for hook errors: Hook failures are silent by default

Logs are empty or incomplete?

• Install jq for better parsing: sudo apt install jq (Linux) or brew install jq (macOS)
• Check script permissions on log directory: ls -ld ~/.claude/conversation-logs

Want to disable logging temporarily?

• Comment out hooks in ~/.claude/settings.json
• Or rename the hook script: mv ~/.claude/hooks/log-conversation.sh{,.disabled}

Privacy & Security

Important considerations:

• Logs contain all prompts you send to Claude, including potentially sensitive info
• Logs are stored unencrypted in your home directory
• Logs persist indefinitely unless manually cleaned up

Recommendations:

1. Regular cleanup: Use the prune script to delete old logs

   # Interactive deletion of logs older than 14 days
   ~/.claude/hooks/prune-logs.sh
   
   # Or use find directly
   find ~/.claude/conversation-logs -name "*.md" -mtime +30 -delete
   

2. Exclude from backups: Add to .gitignore or backup exclusions

   echo "conversation-logs/" >> ~/.gitignore_global
   

3. Encrypt sensitive logs:

   gpg -c ~/.claude/conversation-logs/2026-01-05.md
   rm ~/.claude/conversation-logs/2026-01-05.md
   

Advanced: Per-Project Logs

To log per-project instead of globally, use project-specific hooks in .claude/settings.json within each project:

{
  "hooks": {
    "UserPromptSubmit": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "bash -c 'cat >> .claude/conversation.md'"
          }
        ]
      }
    ]
  }
}

Add .claude/conversation.md to your .gitignore.

Quick Reference

See Also

• Claude Code Hooks Documentation - Full hooks reference
• MCP Logging - Alternative logging via MCP servers