Content
85%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A well-structured, highly actionable skill body with a clear sequenced workflow, explicit CI validation gates, and clean one-level-deep references. Its only real weakness is conciseness: time-sensitive version pins live in the main flow instead of a version-notes or deprecated section.
Suggestions
Move time-sensitive version pins (e.g., '3.0.3' and the @3.0 / @1.2 URL segments) into a dedicated 'Version notes' or pinned-versions section, or strip them to version-agnostic docs links, to avoid the version-number conciseness penalty.
Consolidate the small spine/reference repetition — 'spikes hide in averages and cause cert failures' appears in both the Overview and the Anti-patterns table, and 'disable VSync' appears in both Step 3 and Step 7 — into a single location.
Consider folding the per-parameter explanations of WarmupCount/MeasurementCount/IterationsPerMeasurement into the unity-profiler.md reference, keeping only the call-site code in the spine.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is largely lean and assumes Claude's competence, but it embeds time-sensitive version numbers in the main flow — 'com.unity.test-framework.performance': '3.0.3' and the @3.0 / @1.2 URL segments — which the rubric penalizes unless they sit in a deprecated/version-notes section; this is 'mostly efficient but could be tightened', not the verbose score-1 case. | 2 / 3 |
Actionability | Fully executable, copy-paste-ready guidance: complete C# Measure.Method / Measure.Frames / GC.Alloc blocks, the 'unity -batchmode -runTests ...' CI command, stat-command and ProfileGPU tables, and a concrete budget-report template. | 3 / 3 |
Workflow Clarity | A clearly sequenced 7-step process (Unity capture → analyze → automated tests; Unreal stat triage → Insights; GPU budgets; CI gate) with explicit validation checkpoints — 'fail the build if any PerformanceMeasurement sample set has a median or p99 exceeding the declared threshold' and 'A FAIL on p99 triggers a regression investigation'. | 3 / 3 |
Progressive Disclosure | A concise spine with an Overview, seven steps, template, anti-patterns, and limitations, deferring module detail to two clearly signaled, one-level-deep references (references/unity-profiler.md, references/unreal-insights.md) that contain real content rather than further redirects and explicitly note 'Cited once here rather than repeated in the spine'. | 3 / 3 |
Total | 11 / 12 Passed |