Conciseness

The body is lean throughout: a 4-bullet Mindset, a compact tier table, one-line escalator checks, terse 'Why:' clauses on anti-patterns, and three tight worked examples. It explains no concepts Claude already knows and every section carries decision-relevant signal, matching the 'every token earns its place' anchor.

Actionability

Fully executable guidance for an instruction-only skill: a 6-step procedure with exact rules ('Move up exactly one tier when any escalator fires, capped at Opus'), a tie-break list, a copy-paste output template with a filled example, and worked examples covering all three tiers. As an instruction-only skill, the absence of code is not penalized and the guidance is maximally specific.

Workflow Clarity

The procedure is a clearly sequenced loop with explicit checkpoints: step 1 handles missing input ('Ask for it when the argument is empty'), step 6 enforces justification ('Always name the escalator or matrix row that decided it'), and the confidence rule provides an error-recovery feedback loop ('Ask one clarifying question instead of guessing when confidence is low'). The validation cap for destructive/batch operations does not apply to this recommendation skill.

Progressive Disclosure

SKILL.md is a genuine overview: the base-tier matrix, escalator summaries, tie-breaks, output format and calibrating examples are inline, while deeper material sits in four real one-level-deep files (references/decision-matrix.md, escalators.md, examples.md, reference.md), each clearly signaled inline ('Read the full matrix', 'See escalators for edge cases', 'More cases are in worked examples') and catalogued in a descriptive References section.

Specificity

Concrete action ('Recommend the cheapest Claude model tier (Haiku, Sonnet, Opus)') with a stated method ('task-type matrix plus complexity escalators (scope, stakes, ambiguity, stakeholders)') and covered domains ('Covers technical, business, creative work'). It is one primary action with rich supporting specifics rather than the multiple distinct concrete actions of the 5 anchor, but well above the 1-2-action coverage of the 3 anchor.

Completeness

Explicitly answers what ('Recommend the cheapest Claude model tier... using a task-type matrix plus complexity escalators') and when ('Use when asked "which model", "pick model", "model for", or before starting a costly task') with concrete trigger phrases, matching the 5 anchor exactly. The 'when' clause is specific and quoted, not the merely-present 'when' of the 4 anchor.

Trigger Term Quality

Explicit quoted triggers 'which model', 'pick model', 'model for' are phrases users naturally say, plus 'before starting a costly task'. Common synonyms such as 'choose a model', 'model selection', or 'cheap model' are missing, so coverage is good but not comprehensive as the 5 anchor requires.

Distinctiveness Conflict Risk

Clear niche (Claude tier recommendation) with distinct triggers, but 'model for' is broad enough to fire on cross-vendor or general model-choice questions that the skill itself excludes, and it sits near model/API/pricing skills. Minor overlap risk matches the 4 anchor rather than the minimal-conflict 5 anchor.