---
name: jor-browser-review
description: Review the finished change in the current coding task with Chrome DevTools MCP, preferably through one independent subagent. Use when the user asks for browser review of current work.
disable-model-invocation: true
metadata:
  based-on: "original by Jorge (browser-review)"
---

In Codex, use `$jor-...` for these commands, or `$jor-stack:jor-...` when the skill picker shows the plugin namespace.

# Browser review

Use the active task's scope, acceptance criteria, changed files, and supplied URL or route. Follow the current project's instructions. If the target or a runnable URL cannot be determined, ask only for the missing detail.
Start, stop, or restart dev servers only when the user explicitly requests it, including in worktrees. If the server is down and startup is not authorized, report the repo's start command and stop the browser check.

For every browser run, read and follow [browser session safety](references/browser-session.md), including when no project verification skill exists. Include these rules in any delegated review brief.

Delegate one bounded, read-only review when the harness supports subagents. Prefer a strong available model at high reasoning when available; otherwise use an available capable model and supported effort. Give the reviewer the needed task context. If subagents are unavailable, perform the review in the current agent and say that it was not independent.

Use **Chrome DevTools MCP** to check the relevant viewport sizes, variants, and interactions; inspect layout, overflow, console errors, and failed requests where applicable. Viewport sizes: read `.agents/skills/verify-<project>/SKILL.md` and follow it. If that file is missing, read `.cursor/skills/verify-<project>/SKILL.md` and follow that instead. Use its `doctor` first, its `breakpoints` (360, 768, 1280, 1440, 1920 px wide) for screenshots, and its feature map for the routes, entry points, and end states of the features the change touches. If neither file exists, tell the user to run `/jor-verify`, and check the same five widths: 360, 768, 1280, 1440, 1920. If Chrome DevTools MCP is unavailable, report the blocker instead of substituting another browser tool. Do not edit code, commit, or start another implementation task.

Include a **Human verification** section in the result, delegated or local, with the exact navigable URL for each reviewed page or variant, plus the viewport and interaction to check. Include it even when there are no findings. Wait for any delegated result. Pass those URLs through to the user with concise, prioritized findings, reproduction steps, expected and actual behavior, and evidence. State what was checked and what could not be checked. If there are no findings, say so without claiming untested cases passed. Leave fixes to an authorized implementation pass.

For a large change, the user can request the optional full `/jor-review` after browser findings and fixes are settled. Do not run it automatically.