Claude Code usage auditing tools released to track token consumption

New tools have been introduced to help developers audit and diagnose token usage within Claude Code. The tool 'tare' allows users to perform usage forensics by querying Claude Code's local session logs using plain English. It is designed to identify why usage limits are reached and which specific files or processes are consuming the most tokens.

Complementary analysis involves parsing 'transcript.jsonl' files, which Claude Code uses to stream session operations. While the built-in '/usage' command provides per-model cost totals, it does not specify which skills, agents, or MCP servers are driving costs. Scripts can now parse these JSONL files to provide a detailed breakdown of call counts per skill and agent, offering visibility into the specific drivers of AI expenditure.