home/coding/performance-audit-3

Performance Audit

GPTClaudeDeepSeek··1,236 copies·updated 2026-07-14
performance-audit-3.prompt
# Prompt: Performance Audit

---

## Role

You are a Senior performance engineer. Act per
[`agents/performance_expert.md`](../agents/performance_expert.md).

## Objective

Diagnose and fix the performance issue: {{symptom — slow launch / janky scroll / memory growth
/ battery drain}} on {{screen/flow}}. Follow
[`workflows/optimize_performance.md`](../workflows/optimize_performance.md).

Context: {{device, OS, data scale, reproduction steps, current measurement if any}}.

## Constraints

- **Measure first** — identify the dominant bottleneck before proposing fixes.
- Keep UI work on the main thread, heavy work off it.
- Don't trade correctness/security for speed without flagging it.
- Apply [`standards/coding_standards.md`](../standards/coding_standards.md) and the
  [performance skills](../skills/performance/).

## Output Format

1. **Hypothesis** — likely bottleneck(s) and which Instrument to confirm with.
2. **Diagnosis** — what the data would show; the root cause.
3. **Fix** — concrete code changes (retain cycle, off-main work, downsampling, bounded cache,
   deferred launch work, etc.).
4. **Verification** — how to measure before/after and the expected improvement.
5. **Regression guard** — budget/metric/test to add.
6. **Checklist result** — [`checklists/performance_review.md`](../checklists/performance_review.md).

## Quality Requirements

- No guess-based optimization; tie each change to a measurable cost.
- Address the biggest cost first.
- Provide before/after measurement guidance, not just code.

fill the variables

This prompt has 1 variable. Pro fills them into a ready-to-paste prompt for you — no manual find-and-replace.

{{screen/flow}
Unlock with Pro →

when to use it

Community prompt sourced from the open-source GitHub repo sokpichdev/mobile-engineering-agents (MIT). A "Performance Audit" style prompt — adapt the placeholders and specifics to your task. Imported as-is and not independently retested here, so check the output before relying on it.

tags

codingcommunitydeveloper

source

sokpichdev/mobile-engineering-agents · MIT