Every message I've typed into Claude Code, pulled from
~/.claude/projects/, stripped of pasted logs and
harness-injected text, and counted.
Size and colour both encode frequency. Hover any word for its exact counts.
| # | Word | Uses | Messages |
|---|
Ranked by uses, capped at 3 per message.
Two-word sequences, filler words excluded.
Rate per 10,000 words typed that month, so a busy month doesn't just look like more of everything. Each panel is on its own scale.
Messages typed per repo.
Messages by hour of day, local time.
Words used far more inside one repo than across everything else — the vocabulary that makes a project distinctive rather than just busy.