Transcript Condenser

Transcript Condenser Skill

Condenses verbose Claude Code session transcripts from .claude/logs/ into human-readable summaries.

When to Use This Skill

Claude should invoke this skill when the user:

• Wants to review what happened in a Claude Code session
• Asks about tool usage patterns or statistics
• Mentions analyzing transcripts or logs
• Wants to understand workflow from a past session
• Requests session summaries or reports
• Needs to export session data for analysis

Key Capabilities

Output Formats

• Markdown (default): Human-readable timeline with session metadata and summary statistics
• JSON: Machine-parseable format for automated analysis or integration

Verbosity Levels

• Minimal: Condensed view showing only key events and outcomes
• Standard (default): Balanced detail with truncated long messages
• Detailed: Full content including all tool parameters and complete messages

Filtering Options

• Tool-only view: Show only assistant messages that used tools
• Subagent-only view: Show only subagent invocations and interactions
• System message stripping: Remove system notifications (enabled by default)

Batch Processing

• Process entire directories of transcripts
• Automatically generate output files with consistent naming
• Progress reporting for multiple files

Usage Examples

Basic Usage

# Condense a single transcript to markdown (stdout)
node .claude/skills/transcript-condenser/condense-transcript.js <transcript-file>

# Save condensed version to file
node .claude/skills/transcript-condenser/condense-transcript.js <transcript-file> --output=summary.md

Format Options

# Output as JSON for automated analysis
node .claude/skills/transcript-condenser/condense-transcript.js <transcript-file> --format=json --output=session.json

Verbosity Control

# Minimal output (quick overview)
node .claude/skills/transcript-condenser/condense-transcript.js <transcript-file> --verbosity=minimal

# Detailed output (full content)
node .claude/skills/transcript-condenser/condense-transcript.js <transcript-file> --verbosity=detailed

Filtered Views

# Show only tool usage
node .claude/skills/transcript-condenser/condense-transcript.js <transcript-file> --only-tools

# Show only subagent interactions
node .claude/skills/transcript-condenser/condense-transcript.js <transcript-file> --only-subagents

Batch Processing

# Process all transcripts in a directory
node .claude/skills/transcript-condenser/condense-transcript.js logs/20251017/ --output-dir=condensed/

# Batch process with specific format
node .claude/skills/transcript-condenser/condense-transcript.js logs/20251017/ --output-dir=condensed/ --format=json

Output Structure

Markdown Format

The markdown output includes:

1. Session Metadata: ID, date, branch, duration, working directory, version
2. Timeline: Chronological event list with timestamps (relative to session start)
3. Summary Statistics: Tool usage counts, subagent calls, files modified, errors/warnings

JSON Format

The JSON output provides structured data:

{
  "session": {
    "id": "abc123",
    "date": "2025-10-17",
    "branch": "main",
    "duration": "5m 23s",
    ...
  },
  "timeline": [
    {
      "time": "00:00:15",
      "type": "user",
      "content": "...",
      ...
    }
  ],
  "summary": {
    "toolsUsed": { "Read": 3, "Write": 2 },
    "filesModified": 3,
    ...
  }
}

Automatic Filtering

The script automatically removes noise:

• Meta messages (isMeta: true)
• Hook notifications
• Command clear messages
• Empty stdout
• System reminders (can be disabled with --no-system)

Common Use Cases

1. Quick Session Review: "What did I do in my last session?"

   • Use minimal verbosity for fast overview

2. Tool Usage Analysis: "Which tools did I use most?"

   • Use --only-tools flag

3. Subagent Tracking: "How many times did I call the backend-architect?"

   • Use --only-subagents flag

4. Batch Reporting: "Generate reports for all sessions this week"

   • Use batch processing with output directory

5. Data Export: "Export session data for analysis"

   • Use JSON format for programmatic access

Tips for Claude

• Locate transcript files in .claude/logs/YYYYMMDD/ directories
• Transcript filenames follow pattern: transcript_[subagent_]<sessionId>_<date>_<time>.json
• Suggest appropriate verbosity level based on user's request:
  • "Quick look" → minimal
  • "Detailed analysis" → detailed
  • General review → standard (default)
• For batch processing, create output directory if it doesn't exist
• Always show the output file path when saving to a file

Command Reference

node .claude/skills/transcript-condenser/condense-transcript.js <input> [options]

Arguments:
  input                 Path to transcript file or directory

Options:
  --format=<type>       Output format: "markdown" (default) or "json"
  --output=<file>       Output file (default: stdout)
  --output-dir=<dir>    Output directory for batch processing
  --verbosity=<level>   Level: "minimal", "standard" (default), "detailed"
  --only-tools          Show only tool usage
  --only-subagents      Show only subagent interactions
  --no-system           Strip all system messages
  --include-usage       Include token usage statistics
  --help, -h            Show help message

Requirements

• Node.js 18.0.0 or higher
• No external dependencies (uses built-in Node.js modules only)