Exemplary token efficiency — "Respond terse like smart caveman. All technical substance stay. Only fluff die." — with no concept explanations and every token earning its place. Not 2 because there is no unnecessary explanation to tighten.

Concrete, executable guidance: explicit drop-lists ("articles (a/an/the)", "filler (just/really/basically/actually/simply)"), specific abbreviations, a pattern template ("`[thing] [action] [reason]. [next step].`"), and Not/Yes examples. Absence of code is fine for an instruction skill.

A simple single-purpose skill that is unambiguous and well-organized (Persistence, Rules, Examples, Auto-Clarity Exception); the exception lists trigger conditions and a resume step, and the destructive example includes the verification note "Verify backup exist first".

Under 50 lines with no need for external references, organized into clearly labeled sections. Per the simple-skill guideline this is sufficient for a top score.

Lists multiple concrete mechanisms — "Cuts token usage ~75% by dropping filler, articles, and pleasantries" and "keeping full technical accuracy" — rather than vague language. Not 2 because it is comprehensive and concrete, not partial.

Explicitly answers what ("Ultra-compressed communication mode... Cuts token usage ~75%") and when ("Use when user says... or invokes /caveman"), plus a persistence clause. Not 2 because both what and when are explicit.

Natural user phrases with good variation coverage: "caveman mode", "talk like caveman", "use caveman", "less tokens", and "/caveman". Not 2 because common variations are well covered.

Distinct niche with very specific triggers and an explicit anti-trigger ("A one-off 'be brief' is not a trigger"), making wrong-skill activation unlikely.