dart-perf
GitHub用于复现性能场景、定位瓶颈并优化 Dart 代码。遵循标准方法论，通过 A/B 门控和壁钟时间测量验证性能主张，生成包含基线、候选版本及兼容性分析的详细报告。
Trigger Scenarios
Install
npx skills add dartsim/dart --skill dart-perf -g -y
SKILL.md
Frontmatter
{
"name": "dart-perf",
"description": "DART Perf: reproduce workloads, attribute bottlenecks, optimize with behavior guards, and report performance claims"
}
DART Performance
Required Reading
@docs/ai/principles.md @docs/onboarding/profiling.md
Follow docs/onboarding/profiling.md § "Performance Methodology", the shared
standard for agents and contributors. Reproduce the user's scenario and state
deviations before optimizing. Choose exact-count A/B gates for regressions and
controlled wall-time repeats for speed claims; pair every claim with behavior
guards. Attribute engine and integration costs by phase before choosing a
change, then verify the full workload and downstream compatibility.
For exact-count commands and policy, use the owner's "Revision Comparisons"
section. For visible simulation changes, load dart-verify-sim and follow the
linked verification route. For publication, follow the linked evidence flow;
uploading requires explicit approval.
Output
Leave reproducible evidence: named DART baseline and candidate revisions, scenario/deviations, setup and timed windows, count-gate results, wall-time median/min-max when measured, behavior results, attribution, and compatibility limitations. Report the user effect first with plots using the linked "PR Descriptions" guidance. Public comparisons use DART revisions only.
Version History
- f630c91 Current 2026-10-18 20:26
