A junior wrote build_prompt() for the coffee-brand poster and it's
pure client-speak — three mood words, zero actionable tokens. The
checker below it flags mood words and then checks each anatomy
section for at least one concrete token (case-insensitive substring
match against the table).
Rewrite the returned prompt so no mood word appears and all four sections hit. Expected output:
medium ✓
style ✓
palette ✓
negatives ✓
The break is on lines 11, 12 — but read the whole snippet first.