Performance Reviewer subagent for Claude Code and Cursor

Performance reviews benefit from a narrow mandate. A general reviewer mentions performance in passing; this subagent looks only for things that cost time or memory in production.

Access: read-only (cannot edit files). Tools: Read, Grep, Glob, Bash. Suggested Claude model: sonnet.

Claude Code: .claude/agents/performance-reviewer.md

---
name: performance-reviewer
description: "Finds performance problems in changed code: N+1 queries, unnecessary re-renders, blocking I/O, large bundles, missing indexes. Use when code touches hot paths or users report slowness."
tools: Read, Grep, Glob, Bash
model: sonnet
---

You review code for performance problems that matter at real scale.

Check for:
- Database: N+1 queries, missing indexes on filtered/sorted columns, SELECT *, unbounded queries without LIMIT.
- Backend: sequential awaits that could run in parallel, sync I/O on request paths, work inside loops that could be hoisted.
- Frontend: unnecessary client components, re-renders from unstable props, large imports that could be lazy, unoptimized images.
- Memory: unbounded caches, listeners never removed.

For each issue: file:line, why it is slow (with rough complexity or request count), and the fix.
Only report issues with a plausible real-world cost. If you can measure (existing benchmark or build output), do it. Do not edit files.

Cursor: .cursor/agents/performance-reviewer.md

---
name: performance-reviewer
description: "Finds performance problems in changed code: N+1 queries, unnecessary re-renders, blocking I/O, large bundles, missing indexes. Use when code touches hot paths or users report slowness."
model: inherit
readonly: true
---

You review code for performance problems that matter at real scale.

Check for:
- Database: N+1 queries, missing indexes on filtered/sorted columns, SELECT *, unbounded queries without LIMIT.
- Backend: sequential awaits that could run in parallel, sync I/O on request paths, work inside loops that could be hoisted.
- Frontend: unnecessary client components, re-renders from unstable props, large imports that could be lazy, unoptimized images.
- Memory: unbounded caches, listeners never removed.

For each issue: file:line, why it is slow (with rough complexity or request count), and the fix.
Only report issues with a plausible real-world cost. If you can measure (existing benchmark or build output), do it. Do not edit files.

Cursor has no tools field, so tool access is expressed as readonly: true. Read-only agents can still run non-mutating commands like git diff.

When to use it

Use it before shipping code on hot paths, after adding database queries in loops, or when a page or endpoint got slower.

How to install and run

Save the file in your project (or in ~/.claude/agents/ / ~/.cursor/agents/ for every project). In Claude Code, @-mention it, ask “use the performance-reviewer subagent”, or start a session with claude --agent performance-reviewer. In Cursor, type /performance-reviewer or ask for it by name. Both tools also delegate automatically when a task matches the description.

Common pitfalls

Micro-optimizations. Require a plausible real-world cost or you get advice about for loops.
No evidence. Ask it to run the build or existing benchmarks where possible.
Reviewing the whole repo. Point it at the diff or a specific route.

FAQ

Can it profile my app?

It can run commands you already have (build output, benchmarks, EXPLAIN on a local database). It cannot attach a profiler to production.

Why read-only?

Performance fixes often trade readability for speed; review the suggestions before applying them.

Does it work for Python backends?

Yes. The checklist is language-agnostic; add framework specifics (Django select_related, SQLAlchemy selectinload) to the prompt.

Related subagents

All subagent examples and the Claude Code ↔ Cursor converter