Conciseness

The body is dense and operational — "Count non-test callers exactly with `rg`", the anti-cosmetic gate, and the worktree snippet all assume Claude's competence with no filler explaining concepts Claude already knows. Not 5 because the six-link Related-skills block and the Pass-parameters section carry some tokens that could be tightened further; comfortably above 3 since there is no padded or over-explained section.

Actionability

The git worktree block is copy-paste ready ("git fetch origin pull/<number>/head:pr-<number>-collapse"), the finding format is an exact template, and commands are named (`bun test`, `bun run typecheck`). Not 5 because key mechanics are invoked by name rather than shown — "Count non-test callers exactly with `rg`" gives no actual rg command, and the report shape is deferred to a reference.

Workflow Clarity

The per-iteration ritual is a clear 7-step sequence with explicit validation after each edit ("`bun test` on impacted packages", "`bun run typecheck` on impacted packages"), a revert feedback loop ("If none is true, the change was cosmetic. Revert and find a deeper smell"), a re-grep straggler sweep, and explicit stop conditions. This matches anchor 5's validation-and-recovery pattern.

Progressive Disclosure

The SKILL.md is a clear overview with a "Load on demand" References section where each of the six links (never-touch.md, smell-catalog.md, library-refusal.md, report-format.md, goal-template.md, periodic-monorepo-pass.md) names what it's for, and every referenced file actually exists in ./references/. All references are one level deep and well-signaled, matching anchor 5.

Specificity

"Remove indirection that does not earn its boundary" names the domain and one concrete action, but the action is stated at a single abstract level with no enumeration of what the pass actually does (delete dead exports, inline single-caller helpers, fold files). Anchor 3 ("names domain and 1-2 concrete actions, but not comprehensive") fits; not 4 because it does not list several specific actions, and not 2 because the action is concrete for its domain rather than generic.

Completeness

It explicitly answers both: what — "Remove indirection that does not earn its boundary across a diff or package" — and when — "Use when the user asks to run a collapse pass, simplify this broadly...", with concrete trigger phrases plus an explicit negative boundary ("not when one function's branching is the target"). This matches the anchor-5 example structure.

Trigger Term Quality

"Use when the user asks to run a collapse pass, simplify this broadly, audit dead abstractions, or shrink a surface" gives four natural trigger phrases. Not 5 because common synonyms users might say (e.g. "dead code", "dedupe", "flatten wrappers") are absent; clearly above 3 since multiple natural phrasings are covered.

Distinctiveness Conflict Risk

"collapse pass"/"audit dead abstractions" is a clear niche, and the negative clause "not when one function's branching is the target" explicitly disambiguates from generic simplify/refactor skills, minimizing conflict risk. Matches anchor 5 (clear niche with distinct triggers).