Content
85%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A well-structured performance-audit skill with a clear evidence-driven workflow, executable snippets, and clean reference splitting. Minor trims of concept-restating sentences and one or two concrete examples for underspecified fixes (image downsampling, off-main-actor work) would close the remaining gaps.
Suggestions
Trim sentences that restate known SwiftUI semantics (e.g., 'Identity controls view lifetime and state', 'Observation tracks properties read during view evaluation') — keep only the prescriptive guidance that follows them.
Add a short concrete example for the two remediation items currently given only as direction: image downsampling (e.g., ImageIO thumbnail API) and an off-main-actor decode/parse helper.
Consider moving the eight-item Common Mistakes list or the lazy-container guardrails into a reference file to further slim SKILL.md, since the checklist and smells table already cover the same ground at audit time.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and skill-specific: a smells table, guardrail lists, version-specific flags ('-LogForEachSlowPath YES', '.onGeometryChange (iOS 16+)'), and non-obvious techniques like Self._printChanges() and observation-splitting. Not 5 because a few sentences restate concepts Claude already knows (e.g., 'Identity controls view lifetime and state', 'Observation tracks properties read during view evaluation') and could be trimmed. | 4 / 5 |
Actionability | Executable code snippets are copy-paste ready ('#if DEBUG let _ = Self._printChanges() #endif', the split-reads ProfileView, ForEach with .id(item.stableID)) and tooling is concrete (named Instruments lanes, Cmd+I, -LogForEachSlowPath YES). Not 5 because two remediation items ('Downsample images before rendering', moving work off MainActor) give direction without a concrete example, leaving minor gaps. | 4 / 5 |
Workflow Clarity | A decision tree plus a clearly numbered 5-phase workflow (review, profile, diagnose, remediate, verify) with built-in checkpoints: findings are labeled hypotheses until trace-confirmed, and the Verify step re-runs the same capture against baseline. A review checklist closes the loop. This matches the top anchor including explicit validation and a feedback loop. | 5 / 5 |
Progressive Disclosure | The body is a navigable overview with a TOC, a well-signaled inline pointer to the profiling reference, and a References section listing five real one-level-deep files (verified to exist; reference files link only to external URLs, not nested .md chains). Deep WWDC/Instruments material is appropriately split out while the working checklists stay inline. Matches the top anchor. | 5 / 5 |
Total | 18 / 20 Passed |