FAQ

Are the token counts exact?

No. Every count is characters divided by four. They rank sinks against each other; they do not match a bill.

What does "carried" mean in score?

A token that enters the context early is sent again on every later turn, as a cache read. score multiplies each sink by the turns left after it. Cache reads are billed at about a tenth of the input price, so a large carried number is still worth cutting.

Why does one read appear twice in score?

It no longer does. In 0.1.0 and 0.2.0 a second whole read of a large file was listed both as whole-read and as repeat read, and both went into the total. 0.2.1 counts it once, as a repeat read. Update the plugin, or read older totals as an upper bound.

Does score read my transcripts anywhere else?

No. It reads the JSONL file on disk and prints to the terminal. Nothing is sent anywhere.

Do the hooks slow anything down?

With TOKENWRIGHT_HOOKS unset, each hook starts Python and exits at once. Not measured.

Does it work with other agents?

The skills are plain folders that Copilot, Codex, Cursor and Gemini CLI can read. tw.py score and repeats read Claude Code transcripts only. Other agents are not tested.