browse

An agent skill by garrytan, from garrytan/gstack. Tags: debugging, screenshots, ui-ux, web.

What it does

Drive a real browser through Aside: open a page, read it, click through a flow, take screenshots, check console errors. (gstack)

Install

With the skills CLI, which installs into Claude Code, Codex, Cursor and other agents:

npx skills add garrytan/gstack --skill browse

Or copy the skill folder into Claude Code's skills directory by hand (~/.claude/skills for every project, or .claude/skills inside one):

git clone --depth 1 https://github.com/garrytan/gstack
cp -r gstack/browse ~/.claude/skills/browse

Safety box score

Grade F, high risk: 74 of 100 risk points, rated 2026-10-05. The grade rates what the skill and its scripts can reach on the machine of whoever installs it, across eight categories from shell execution to secrets access. How the score works.

This skill drives the user's real browser to open pages, click through flows, fill forms, and take screenshots. The main blast-radius risks are that it shells out to local helper scripts in the user's home directory and operates on live logged-in browser sessions with real cookies.

Category levels, from 0 (none) to 3 (high)
Category	Level	Why
Code & shell execution	3 (high)	It shells out to local helpers like `~/.claude/skills/gstack/bin/gstack-skill-start`, `...gstack-skill-end`, and `...gstack-learnings-log`, so installing it grants real command execution outside the repo.
Network access	2 (moderate)	It is a live browser skill that opens named URLs, follows same-origin links, and reads page content, so remote sites receive real browser traffic from the user's session.
Filesystem reach	2 (moderate)	It writes reports and telemetry under home-dot directories such as `~/.gstack/` and `~/.claude/skills/gstack/`, so it touches files outside the project tree.
Secrets & credentials	3 (high)	It explicitly uses `the user's real browser — Aside, with their real cookies and their real logged-in sessions`, so authenticated accounts and session data are in scope.
Privilege & persistence	1 (low)	I see no sudo, service registration, or shell-startup mutation; the heaviest setup path is a user-level `./setup` or browser build.
Instruction hijack surface	2 (moderate)	It tells the agent to honor `GSTACK_INSTRUCTION_BEGIN` blocks from tool output and to treat the skill file as executable instructions, which is a prompt-injection surface.
Irreversible actions	3 (high)	It can click, fill, submit, and on non-local targets mutate a real account after a confirmation gate, so the blast radius includes posting, purchases, and deletes.
Transparency	2 (moderate)	Some core behavior is delegated to unread helpers outside the repo, including the `~/.claude/skills/gstack/bin/...` scripts and `./setup`, so a reviewer cannot inspect the full runtime path here.

Before installing

Inspect `~/.claude/skills/gstack/bin/gstack-skill-start`, `gstack-skill-end`, and `gstack-learnings-log` if they exist on your machine.
Confirm you are okay with the skill using your logged-in browser session and cookies on real sites.
Try it first on a local test page or throwaway account before pointing it at a real service.

Rated 2026-10-05 by an automated review of 87 files from the skill's repository.

Source

Repository: garrytan/gstack (all skills from this repository)
Path: browse/SKILL.md
Branch: main
Updated: 2026-10-05

hallmark — Anti-AI-slop design skill for greenfield pages, audits, redesigns, and design extraction from URLs or screenshots.
chrome-mcp-troubleshooting — Diagnose and fix Claude in Chrome MCP extension connectivity issues.
Make Bot UI — Use when building a custom UI (page, dashboard, buttons) that should wake a Grok Bot over a webhook, when the user must provide a webhook sender key.
review-walkthrough — Generates an interactive HTML walkthrough for reviewing code changes. Use only when explicitly called.
run-smoke-tests — Run Playwright smoke tests, debug failures, and verify fixes
stitch::generate-design — Generate new screens from text prompts or images, edit existing screens with prompts and design system tokens, and generate design variants using Stitch MCP.