Conciseness

The body explains nothing Claude already knows and pads nothing; every line is operative, and detail is delegated to references ("references/build.md defines yields and artifacts", the quality floor "neither lives here"). This matches anchor 5's lean, competence-assuming standard; anchor 4's "minor instances of over-explanation" are absent.

Actionability

For an instruction-only skill the guidance is fully concrete: an exact command ("git -C <repo root> check-ignore -q .context/compound-engineering/" with the trailing-slash caveat), exact run paths (".context/compound-engineering/ce-prototype/<date>-<slug>/" and its fallback root), named files to read per phase, and explicit branch rules (seeing vs driving, overlay vs isolated, write-back vs recap) — matching anchor 5 for an instruction-only skill.

Workflow Clarity

Four gated phases run in clear sequence (Scope the question, Build it, Keep the decisions, Apply or continue) with explicit validation steps ("Do not build until authorized", the check-ignore probe before the root is resolved, the confidence gate on capsule writes) and error-recovery loops (unclean undo → "name the files you left modified"; unclear relatedness → recap instead of write-back). The product-tree overlay path carries its own validation, so the destructive-operation cap does not apply.

Progressive Disclosure

All six references cited in the body (scoping, build, preview, annotation-loop, craft-floor, write-back) exist in references/, each with an explicit load condition ("before you ask the user anything", "for a seeing question", "when the user applies"). The 1275-line server script and 837-line overlay assets stay out of SKILL.md, and reference cross-links are peer-level ownership pointers rather than nested content chains — matching anchor 5's well-signaled one-level-deep structure.

Specificity

"Build a throwaway prototype to answer how something should work, feel, or read" names a concrete action and its three test dimensions, going beyond anchor 3's generic one-to-two-action examples, but coverage rests on a single core action rather than the several specific actions of anchor 5.

Completeness

Both parts are explicit: the what ("Build a throwaway prototype to answer how something should work, feel, or read") and a literal "Use when committing the wrong answer would be expensive to unravel and a cheap sketch cannot settle it" clause. It falls short of anchor 5 because the when is a situational cost condition rather than concrete user-facing trigger phrases.

Trigger Term Quality

"prototype" and "cheap sketch" are terms users naturally say, and the use-when clause frames the triggering situation, but common synonyms such as "mockup", "spike", "demo", or "proof of concept" are missing — good coverage with a few natural terms absent, matching anchor 4 rather than anchor 5's comprehensive synonym list.

Distinctiveness Conflict Risk

The negative triggers ("Not a rough visual probe during brainstorming, not for deciding what to build, not polishing a feature that already works, not implementing the real thing") explicitly carve out adjacent skills, giving a clear niche with minimal conflict risk, matching anchor 5.