Perf Profiling

Performance Profiling

Use this skill to run the repo’s performance checks (or propose a minimal plan if they don’t exist), compare results to budgets/baselines, and produce an actionable regression summary.

This skill is intentionally strict about “evidence”:

• If you did not measure it, label it as a risk hypothesis.
• If you did measure it, include the exact command and the key output.

Workflow

1. Discover perf tooling, baselines, and budgets.

• Inspect package.json scripts for: perf, benchmark, profile, lighthouse, bundle, analyze, size, stats.
• Look for tooling/config files:
  • next.config.* (bundle/analyze hooks, headers, compression toggles)
  • lighthouserc.*, lighthouse.*
  • size-limit.*, .size-limit.*, bundlesize.*
  • webpack.config.* / analyzer config (if present)
• Check docs under docs/ (or repo root docs) for perf budgets and release gates.

2. Identify the target surface (pick only what’s relevant).

• Next.js:
  • Client JS size per route, shared chunks, and third-party bloat.
  • Route rendering mode (SSR/SSG/ISR) and caching behavior.
  • Server execution hotspots (API routes/server actions) and cold starts (if serverless).
  • Lighthouse and Core Web Vitals (if tracked).
• React Native:
  • App start time and bundle load.
  • JS thread responsiveness (dropped frames) and expensive renders.
  • Memory pressure and large images/lists.
  • Bundle size and source map size (release artifacts).

3. Collect metrics (prefer repo scripts, otherwise propose minimal commands).

• Use the repo's scripts; avoid inventing flags or new dependencies.
• If no scripts exist, propose a minimal plan with clear command candidates and expected outputs, but do not run it unless asked.

4. Compare to budgets/baselines.

• If metrics or budgets are defined, compare current results to those numbers.
• If no budgets exist, call out the missing baseline explicitly.

5. Triage regressions (turn numbers into fixes).

• Separate "measurement noise" from real regressions (e.g., CI variance vs consistent deltas).
• Prefer actionable root causes:
  • New dependency / large asset / code path change
  • Unintentional client-side import of server-only code
  • Increased re-rendering due to unstable props/state
  • Missing memoization in list-heavy views (RN)
  • Next.js route accidentally switched from SSG/ISR to SSR
  • Expensive server action / API route missing caching or doing N+1 work

What “Good” Looks Like (Heuristics)

• Next.js:
  • Avoid pulling large libs into the client for one small feature.
  • Keep per-route client JS growth explainable (new feature weight).
  • Avoid “everything becomes client” migrations (overuse of use client).
• React Native:
  • Keep FlatList item renders cheap, stable, and memoized when needed.
  • Avoid decoding huge images on the JS thread.
  • Avoid repeatedly creating functions/objects in render loops if it triggers re-renders.

Output Format

Provide:

• Metrics Collected (with tool/source)
• Budget Comparison (pass/fail or missing)
• Regressions/Risks (ordered by impact)
• Suspected Causes (if supported by evidence)
• Suggested Follow-ups
• “Next Action” (the single most valuable fix to try first)

Notes

• Keep results scoped to the repo's existing tooling and documented expectations.
• Do not introduce new dependencies unless asked.

Example Commands (only when asked to run)

# Prefer the repo's scripts; these are examples only.

# Next.js (common patterns)
pnpm run build
pnpm run lint

# If the repo already has these scripts/configs:
pnpm run analyze
pnpm run lighthouse
pnpm run size