Conciseness

The body practices what it preaches: "Respond terse like smart caveman. All technical substance stay. Only fluff die." No concept Claude already knows is explained, every rule is stated once, and examples are minimal Bad/Good pairs. Every token earns its place.

Actionability

For an instruction-only skill, guidance is fully executable: concrete Bad/Good pairs ("Bad: 'The component re-renders because an inline object prop creates a new reference each render.' Good: 'Inline object prop, new ref, re-render. `useMemo`."), an explicit protected-content list ("Never cut: code, commands, paths, API names, error strings (verbatim)..."), and a precise `/caveman status` fallback ("relay the hook's `Caveman mode: <mode>` value; with no hook, the mode you were in (or `off`) plus `(not tracked by this host)`"). The example pairs cover the common compression cases.

Workflow Clarity

A single-action skill (respond in compressed fragments) whose action is unambiguous, satisfying the simple-skill exception. The "Pre-send check" section provides an explicit validation checkpoint before every response ("Two readings? Full sentence. Negations present? Payload verbatim? Flavor word? Cut."), and no destructive or batch operations exist that would cap the score.

Progressive Disclosure

Self-contained single-purpose skill with well-organized sections (Persistence, Floor, Rules, When to break the rules, Pre-send check) and no bundle files or external references needed — nothing here would benefit from being split out. Structure is clear and easy to navigate.

Specificity

Lists several concrete behaviors — "fragments, one word when one word is enough, each fact once" plus persistence behavior ("Stays on until 'stop caveman' or 'normal mode'") — but omits other capabilities from the body (protected content, break-the-rules cases), so coverage has minor gaps. Not 3 because it goes beyond 1-2 actions; not 5 because the capability list is not comprehensive.

Completeness

Explicitly answers both: what ("Caveman at maximum compression: fragments, one word when one word is enough, each fact once") and when ("Invoke only with /ultracave or /caveman ultra"). The 'when' uses concrete trigger phrases and even states the mode's lifetime, matching the anchor for a fully explicit what+when pair.

Trigger Term Quality

Gives the exact natural invocation phrases users would type for this command-only skill — "/ultracave", "/caveman ultra", "stop caveman", "normal mode" — covering both entry and exit. Not 5 because natural-language synonyms a user might say ("terse", "brief", "compress responses") are absent, which would matter if it were model-invoked.

Distinctiveness Conflict Risk

"Caveman at maximum compression" carves a clear niche distinct from a base caveman skill, and invocation is restricted to exact slash commands ("Invoke only with /ultracave or /caveman ultra"), so the risk of triggering the wrong skill is minimal. Third-person voice is used throughout, per the guideline.