v2.1.1 - Token Counting Fix

🐛 Bug Fixes

Token Counting Accuracy

• Fixed token counting to match Claude Code UI's /context command
• Now correctly counts: compact summary content + user message content + assistant usage tokens
• Achieves 96% accuracy compared to UI (6,341 vs 6,600 tokens)

API Improvements

• Added count_session_tokens() function for consistent API interface
• Fixed isCompactSummary field name mismatch (was incorrectly is_compact_summary)
• Improved SQL queries to include cache_read_input_tokens for accurate counts

One-liner Usage

from claude_parser import load_latest_session, count_session_tokens

# Get current session tokens
tokens = count_session_tokens(load_latest_session())

---

🤖 Generated with Claude Code