Conciseness

The body is lean and assumes Claude's competence — no explanations of what tokens, hooks, or caches are — and every sentence carries instruction (host-specific behavior, flags, anti-fabrication rules). The repeated "savings unknown" statements are deliberate guardrail emphasis across different hosts and flags, not padding, so anchor 5 fits better than anchor 4.

Actionability

As an instruction-only skill it gives fully concrete guidance: exact paths (src/hooks/caveman-stats.js), exact commands (/stats model, /stats session), exact flags (--all, --since 7d, --share), and unambiguous do/don't rules per host. This meets the top anchor's standard of specific, executable direction covering the common cases.

Workflow Clarity

Host branching is clear and each branch's single action is explicit (Claude Code: print hook report verbatim; Gemini: direct to /stats; others: native report or state unavailability), with no destructive or batch operations requiring validation. It is not 5 because there is no fallback instruction for the case where the hook supplies no additionalContext, leaving a minor gap in the primary path.

Progressive Disclosure

The skill is under 50 lines with no bundle files, and the five short paragraphs are well organized by topic (hook mechanics, host fallbacks, report contents and limits, flags, file-size caveat). Per the simple-skill guideline this earns the top anchor; the mentioned paths (src/hooks/*.js, docs/HONEST-NUMBERS.md) are repo files the skill operates on, not nested skill references.

Specificity

"Show recorded output and cache-read token usage and mode attribution" and "locate the host's native usage report" list several concrete, specific actions in third person. It falls short of anchor 5 because body-documented capabilities (--all, --since, --share, statusline mode display) are absent from the description.

Completeness

The "what" is clear (shows recorded output, cache-read tokens, mode attribution; locates native usage reports), but the only "when" guidance is "Trigger: /caveman-stats", a command name with no conditions or natural trigger phrases — a missing "Use when..." clause caps this at 3. Not score 2 because the "what" is specific, not vague.

Trigger Term Quality

Natural terms users would say are present: "token usage", "cache-read", "session", "usage report". A few common variations are missing ("token count", "cost", "spend", "context usage"), matching anchor 4 rather than anchor 5's comprehensive synonym coverage.

Distinctiveness Conflict Risk

"Recorded output and cache-read token usage and mode attribution for the current Claude Code session" carves out a clear niche (Caveman-mode session stats) with a distinct slash-command trigger, so conflict with other skills is minimal. It is more distinct than anchor 4's "minor overlap risk" example.