Token usage
What building this project with Claude Code (Opus 5.5, Pro plan) cost in tokens, where the tokens went, and how one long session compares with starting fresh.
Context re-read on every call
Each model call re-sends the whole conversation so far. Without /clear or /compact, this grows with every turn. Vertical lines mark your prompts.
Table
By project step
Step is estimated from the tools each call used (rules in pipeline/tokens.py). Token counts are measured.
Table
Cost of one call: fresh vs late in a long session
Tokens re-read by a single call.
Fixed overhead every call pays
By prompt
Total tokens processed while answering each of your messages (input, cache write, cache read and output).
Table
Plan usage and method
The written analysis is in TOKEN_USAGE.md.