ras-commander: Agent for Claude Code

Conversation Index Scanner

Scan conversation history index quickly and efficiently.

Purpose

Scan ~/.claude/history.jsonl to:

• Count prompts by time period
• Identify active projects
• Extract prompt text for pattern matching
• Build activity timeline

Data Source

File: ~/.claude/history.jsonl

Format: JSONL (one JSON object per line)

{"display": "user prompt text", "timestamp": 1733847562000, "project": "C:\\path\\to\\project", "pastedContents": {}}

Scanning Tasks

1. Basic Statistics

# Count total prompts
prompt_count = sum(1 for line in open(history_file))

# Filter by date
cutoff_ms = (datetime.now() - timedelta(days=7)).timestamp() * 1000
recent = [p for p in prompts if p['timestamp'] >= cutoff_ms]

2. Project Distribution

from collections import Counter
projects = Counter(Path(p['project']).name for p in prompts)

3. Time Distribution

from datetime import datetime
dates = Counter(
    datetime.fromtimestamp(p['timestamp']/1000).strftime('%Y-%m-%d')
    for p in prompts
)

4. Prompt Text Extraction

# Extract all prompt text for pattern analysis
prompt_texts = [p['display'] for p in prompts]

Output Format

{
  "total_prompts": 2240,
  "time_range": {"start": "2025-09-27", "end": "2025-12-13"},
  "projects": [
    {"name": "ras-commander", "count": 450},
    {"name": "hms-commander", "count": 120}
  ],
  "daily_activity": [
    {"date": "2025-12-13", "count": 45},
    {"date": "2025-12-12", "count": 38}
  ],
  "prompts": ["prompt text 1", "prompt text 2", ...]
}

Performance Notes

• File is ~1-2 MB for active users
• Stream-parse the file (do not load entirely into memory)
• Complete index operations in <5 seconds
• Cache results for subsequent analysis

Cross-References

Agents (collaborate with):

• conversation-insights-orchestrator -- Coordinates analysis workflows
• conversation-deep-researcher -- Deep analysis after initial scan