Mobile Engineering Agents

Appearance

Workflow: Optimize Performance ​

Measure → find the bottleneck → fix → verify. Never optimize on intuition.

Objective ​

Resolve a performance problem (slow launch, janky scrolling, memory growth, battery drain) with a measured before/after improvement and no correctness regressions.

Inputs ​

  • Symptom + context (which screen/flow, device, OS, reproduction steps).
  • A target/budget if one exists (e.g. launch < 400ms, 60fps scroll).

Outputs ​

  • Root-cause analysis, the fix, and before/after measurements.

Step-by-Step Process ​

  1. Reproduce + measure baseline (Performance Expert) — pick the right Instrument (Time Profiler, Allocations/Leaks, Animation Hitches, Energy) and capture numbers.
  2. Identify the dominant bottleneck — fix the biggest cost first.
  3. Apply the targeted fix — e.g. move work off main thread, break a retain cycle, bound a cache, downsample images, defer launch work (see skills/performance/).
  4. Re-measure under the same conditions; compare to baseline.
  5. Guard against regression — add a budget/metric (MetricKit, a perf test) where feasible.
  6. Review — ensure no correctness/security trade-off was made silently.

Validation Steps ​

  • Before/after measurements show a real improvement on the target metric.
  • No new correctness failures; tests still pass.
  • The change targets the measured bottleneck, not a guess.

Failure Scenarios ​

  • No measurable improvement → wrong bottleneck; re-profile.
  • Improvement with a regression → revert; find a non-destructive approach.
  • Can't reproduce the slowness → match device/OS/data scale of the report.

AI Agent Instructions ​

  • Always attach a baseline measurement before changing anything.
  • Fix the dominant cost first; avoid micro-optimizations that don't move the metric.
  • Keep UI work on the main thread and heavy work off it; never trade correctness/security for speed without explicit sign-off.
  • Re-measure and report before/after numbers.

Acceptance Criteria ​

  • [ ] Baseline measured with the appropriate Instrument.
  • [ ] Dominant bottleneck identified and addressed.
  • [ ] Before/after numbers show improvement.
  • [ ] No correctness/security regression; tests pass.
  • [ ] Regression guard added where feasible.