Tier-2 Review: ios-design-review

What we attempted

We picked up ios-design-review (cluster: general, source: skillsmp.com listing for garrytan-gstack-ios-design-review-skill-md) with a composite score of 4.2 / 5.0 and dimension scores that looked, on paper, like a strong candidate: trigger clarity 4.5, scope precision 4.5, output specificity 4.0, self-containment 4.0, reusability 3.5.

We ran the standard D-048 test harness against it: an install step and a smoke-invocation step, executed inside a clean Docker container. Neither test produced a pass, a partial, or a fail. Both were skipped, and the reason for each skip is the same root cause — so this review is really about one failure mode, not two.

What failed

The harness could not execute either test because the artifact we were given is not runnable. The SKILL.md content available to us is truncated to a single line:

Visual design audit for iOS apps on real hardware. (gstack)

That is a description, not an interface. There is:

• No install command. No package name, no repository URL, no npm/pip/brew/git clone line, nothing the harness could shell out to.
• No invocation syntax. No CLI entrypoint, no function signature, no prompt template, no README pointer. We cannot construct a call.
• No runnable surface at all. The harness had nothing to bind to, so install skipped and smoke-invocation skipped in sequence.

The second skip is partly environmental and partly intrinsic. Even if the interface had been documented, the skill explicitly targets "real hardware" iOS devices. A clean Docker container has no iOS device, no Xcode device bridge, no simulator with the right runtime. So the smoke test would have had to be skipped on capability grounds regardless. The first skip, however, is not environmental — it is a packaging problem, and it is the one that matters.

What we observed

• 0 tests passed, 0 partial, 0 failed. Both skipped. There is no execution signal here, positive or negative.
• Observed dependencies: none. That is not a compliment. It means the harness found nothing to declare as a dependency, because it found nothing at all.
• The composite 4.2 and its dimension scores are therefore theoretical. They were derived from the listing and the one-line description, not from any observed behavior. We are reporting them as context, not as validated results.

To be explicit about the failure mode: this is a truncation / packaging failure, not a logic failure. We cannot say the skill is broken. We can only say we could not reach it. The distinction matters and we are not going to blur it.

Rating caveat

The 4.2 / 5.0 composite should be treated as unverified until a physical re-run resolves the above. A reviewer with an iOS device and the untruncated SKILL.md could plausibly confirm or contradict every dimension score. Until that happens, the number is a prior, not a measurement. We are not marking tests as passed, and we are not marking them as failed — they did not run.

Does the skill still seem valuable in principle?

Yes, provisionally. A visual design audit skill scoped to real-hardware iOS apps addresses a real gap: simulator rendering and screenshot diffing miss touch-target ergonomics, real color reproduction, and device-specific layout quirks that only show up on glass. The dimension profile — high trigger clarity and scope precision, moderate reusability — is consistent with a focused, opinionated tool rather than a general-purpose one, which is the right shape for this job.

The value proposition is intact. What is missing is the packaging: an install line, an invocation contract, and a documented hardware requirement the harness can check against. Fix those three and this becomes testable. Leave them out and it stays a description of a skill rather than a skill.

Verdict: valuable in principle, unverified in practice, blocked on packaging and hardware.

Tests simulated against README claims; pending physical re-run in Docker harness. Ran 2026-10-04.

Overall: broken. 0 tests passed, 0 partial, 0 failed; both tests skipped because the truncated SKILL.md exposes no install command, invocation syntax, or runnable interface, and the skill requires real iOS hardware unavailable in Docker.

Test	Status	Notes
install	skipped	SKILL.md content is truncated to a single-line description ('Visual design audit for iOS apps on real hardware. (gstack)') with no documented install command, package name, or repository URL, so no install step can be executed.
smoke-invocation	skipped	No README, CLI entrypoint, or invocation syntax is present in the provided SKILL.md; the skill also targets 'real hardware' iOS devices, which cannot be exercised in a clean Docker container.