Content
37%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
This skill provides some useful Kotlin code snippets for common Compose performance fixes (stable keys, remember, lambda-based offset), but falls short on its core promise of teaching how to profile and improve performance. The workflow is generic and lacks a concrete measure-fix-validate loop, and there's no guidance on actually using profiling tools (Layout Inspector, composition counts, macrobenchmarks). Much of the content covers general Compose best practices rather than performance-specific actionable steps.
Suggestions
Add a concrete profiling workflow with explicit steps: enable composition tracing, identify recomposition counts via Layout Inspector, measure with macrobenchmark, fix, re-measure to validate improvement.
Include actual profiling commands and tool usage (e.g., how to enable recomposition counts in Layout Inspector, how to set up a macrobenchmark test, how to read a systrace for jank).
Remove or significantly trim generic guidance (Guardrails, Anti-Patterns) that Claude already knows about Compose development, and replace with performance-specific diagnostic patterns.
Add a feedback loop to the workflow: measure baseline → identify hotspot → apply fix → re-measure → confirm improvement, with specific metrics to check (frame times, recomposition counts, startup time).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The skill is reasonably concise but includes some sections that feel like filler rather than adding unique value. The Workflow and Guardrails sections are generic enough that Claude would already know most of this (e.g., 'respect accessibility semantics,' 'prefer measured performance work over premature micro-optimizations'). The Anti-Patterns section similarly states things Claude already understands. | 2 / 3 |
Actionability | The remediation examples provide concrete, executable Kotlin snippets which is good, but the workflow steps are vague and abstract ('Identify whether the target surface is Compose, View system, or a mixed interoperability screen'). There's no concrete profiling guidance—no actual commands for using Layout Inspector, composition tracing, or macrobenchmark setup despite the skill being about profiling and improving performance. | 2 / 3 |
Workflow Clarity | The workflow is a generic 5-step process that reads more like a UI development checklist than a performance profiling workflow. There are no validation checkpoints, no feedback loops for performance measurement (measure → identify hotspot → fix → re-measure), and the steps don't clearly sequence a performance investigation. For a skill about profiling and improving performance, the absence of a measure-fix-validate loop is a significant gap. | 1 / 3 |
Progressive Disclosure | The skill mentions handoff skills and includes official references, which is good. However, the content is somewhat monolithic—the remediation examples, anti-patterns, and workflow are all inline when some could be split out. The references are well-signaled but the overall structure could better separate quick-start profiling from deeper optimization patterns. | 2 / 3 |
Total | 7 / 12 Passed |