Measure React Renders

Render Performance Comparison

Step 1: Create the baseline worktree

git fetch origin main
git worktree add /tmp/sculptor_baseline origin/main
cd /tmp/sculptor_baseline/sculptor/frontend && npm install --silent

Step 2: Write the scenario file

If the user provides one, use it. Otherwise write a Python file exporting:

• DESCRIPTION: str — what is being measured
• TARGET_COMPONENTS: list[str] — component names to highlight
• setup(page, base_url, workspace_id, task_id) — navigate + wait
• action(page) — perform the action to measure

Component names must be unminified source names (e.g. SplittableSectionComponent for the memo-wrapped SplittableSection export, AlphaChatInterface if not wrapped).

See scenarios/panel_resize.py for an example.

Step 3: Run the comparison

uv run --project sculptor python \
  .claude/skills/measure-react-renders/scripts/perf_compare.py \
  --baseline-dir /tmp/sculptor_baseline \
  --current-dir "$(pwd)" \
  --scenario path/to/scenario.py

Use --skip-build if frontends are already built.

The script builds both frontends with --minify false (preserves component names), starts two isolated backends, injects __REACT_DEVTOOLS_GLOBAL_HOOK__ via Playwright's addInitScript before React loads, runs the scenario, and prints a comparison table.

Step 4: Cleanup

git worktree remove /tmp/sculptor_baseline

Scenario example

import time

DESCRIPTION = "Section resize render cascade"
TARGET_COMPONENTS = ["AlphaChatInterface", "SectionGrid", "SplittableSectionComponent"]

def setup(page, base_url, workspace_id, task_id):
    page.goto(f"{base_url}/#/ws/{workspace_id}/agent/{task_id}")
    page.wait_for_load_state("networkidle")
    time.sleep(5)

def action(page):
    handles = page.locator('[role="separator"]').all()
    handle = handles[-1]
    handle.focus()
    time.sleep(0.3)
    for _ in range(5):
        page.keyboard.press("ArrowLeft")
        time.sleep(0.15)
    for _ in range(5):
        page.keyboard.press("ArrowRight")
        time.sleep(0.15)