cyberneticlibrary

Diagnose your Claude spending and usage

token-doctorskillsetup L281
techwolf-ai/ai-first-toolkit
What it does

Diagnose and report Claude token spend patterns and hotspots

Best for

Identifying spend concentrations and antipatterns when your Claude bill is unexpectedly high.

Inputs
  • · session transcripts from ~/.claude/projects/*/*.jsonl and ~/Library/Application Support/Claude/.../audit.jsonl
  • · time window (default 90 days)
Outputs
  • · Stage 1: terminal report with doctor-style ASCII diagram
  • · Stage 2: Markdown report with per-project analysis
Requires
  • · Python 3 stdlib
  • · optional Cowork transcript parser
Preconditions

Claude Code or Cowork transcripts exist; Python 3 available

Failure modes
  • · No session data (missing transcript paths)
  • · Cache rebuilds underestimated
  • · Re-read ratio skewed by automated sessions
Trust signals
  • · Uses real message.usage from transcripts, not estimation
  • · Traffic-light vitals (🔴🟡🟢) calibrated against org averages
  • · Per-project emoji assigned deterministically from by_cwd_top