Skip to main content
hc
Submit
skillsSource-backedReview first Safety ✓ Privacy ✓

Claude Code Computer Use GUI QA Capability Pack Skill

Expert Claude Code computer use GUI QA capability pack that applies documented computer-use MCP enablement, per-app session approval, screenshot checkpoints, and macOS permission workflows for native app validation—not an official Anthropic QA product.

by kiannidev·added 2026-06-16·
Claude CodeCodexWindsurfGeminiCursorCLI
HarnessClaude CodeCodexWindsurfGeminiCursorCLI
Level:expertType:capability-packVerified:validated
Review first — review before installing

Open the source and read safety notes before installing.

Safety notes

  • Computer use controls real desktop apps with per-app session approval—not sandboxed Bash.
  • Review sentinel warnings before approving Terminal, Finder, or System Settings access.
  • Press Esc or Ctrl+C to stop computer use and release the machine-wide lock immediately.
  • Only one Claude Code session can hold the computer-use lock at a time.

Privacy notes

  • Screenshots are downscaled before model upload but may still capture sensitive UI content.
  • Your terminal window is excluded from screenshots per official documentation.
  • Redact customer or proprietary UI details before sharing screenshot evidence externally.
  • Other visible apps are hidden while Claude works; still minimize sensitive windows before starting.

Prerequisites

  • Interactive Claude Code session on macOS v2.1.85+ with Pro or Max plan per computer-use docs.
  • computer-use MCP server enabled through /mcp for the project.
  • macOS Accessibility and Screen Recording permissions granted when prompted.
  • Disposable QA target app or simulator without live customer data on screen.

Schema details

Install type
package
Reading time
9 min
Difficulty score
78
Troubleshooting
Yes
Breaking changes
No
Source repository stats
Scope
Source repo
Package metadata
Skill and platform metadata
Skill type
capability-pack
Skill level
expert
Verification
validated
Verified at
2026-06-16
Retrieval sources
https://code.claude.com/docs/en/computer-usehttps://code.claude.com/docs/en/skillshttps://code.claude.com/docs/en/features-overviewhttps://code.claude.com/docs/en/securityhttps://github.com/anthropics/claude-codehttps://developers.google.com/search/docs/fundamentals/creating-helpful-content
Tested platforms
ClaudeClaude CodeCodexCursorGeneric AGENTS
PlatformSupportInstall path
claude-codeNative.claude/skills/<skill-name>/SKILL.md
codexNative.agents/skills/<skill-name>/SKILL.md
windsurfNative.windsurf/skills/<skill-name>/SKILL.md
geminiNative.gemini/skills/<skill-name>/SKILL.md or .agents/skills/<skill-name>/SKILL.md
cursorAdapter.cursor/rules/<skill-name>.mdc
cliManualAGENTS.md or tool-specific context file
Full copyable content
# Trigger
"Apply the Claude Code computer use GUI QA capability pack for this flow."

# Required output
1) Enablement and macOS permission checklist
2) Per-app approval and sentinel review notes
3) Screenshot-indexed test steps and observations
4) Stop/rollback actions taken
5) Privacy-safe QA summary

About this resource

Knowledge Freshness

This capability pack is grounded in Claude Code computer-use, skills, features overview, and security documentation verified on 2026-06-16. Computer use is a research preview on macOS with plan and version requirements that can change; prefer live official docs over cached workflow assumptions.

Retrieval Sources

Source Verification Notes

Verified against official Claude Code computer-use documentation on 2026-06-16:

  • Computer use runs through the built-in computer-use MCP server, disabled until enabled in /mcp, and requires interactive macOS sessions on Pro or Max plans.
  • macOS Accessibility and Screen Recording permissions are required before control begins.
  • Per-app session approval gates which applications Claude may control; sentinel warnings flag Terminal, Finder, and System Settings as high-reach approvals.
  • A machine-wide lock allows only one computer-use session at a time; Esc or Ctrl+C stops control and releases the lock.
  • Screenshots are downscaled automatically; the terminal window is excluded from captures.
  • Official examples cover native build validation, layout bug reproduction, and iOS Simulator tap-through flows.

Scope Note

This is a community reusable QA workflow skill—not an official Anthropic test product. It applies documented computer use enablement, approval, screenshot, and stop steps from code.claude.com. Browser-only automation is covered by Claude in Chrome documentation instead.

Core Workflow

  1. Confirm eligibility: macOS, Claude Code v2.1.85+, Pro/Max plan, interactive session.
  2. Enable computer-use in /mcp if not already enabled for the project.
  3. Grant Accessibility and Screen Recording when macOS prompts; restart if needed.
  4. Define the GUI flow under test with explicit pass/fail criteria and screenshot checkpoints.
  5. Start the task; approve only the target apps when session prompts appear.
  6. Review sentinel warnings before approving high-reach apps such as Terminal.
  7. Capture screenshot-indexed observations at each documented checkpoint.
  8. Stop with Esc or Ctrl+C if UI content looks suspicious or the flow completes.
  9. Produce a privacy-safe QA summary with findings—not a formal CI test harness.

Capability Scope

  • computer-use MCP enablement checklist.
  • macOS permission and plan eligibility verification.
  • Per-app approval and sentinel review notes.
  • Screenshot-indexed GUI validation steps.
  • Lock, stop, and rollback guidance.
  • Privacy-safe QA reporting.

Compatibility

Native

  • Claude Code / Claude: use as an Agent Skill when running documented computer use GUI validation on macOS interactive sessions.

Manual Adaptation

  • Codex, Cursor, Generic AGENTS: apply the checklist when reviewing computer use QA plans against public Claude Code documentation.

Required Inputs

  • Target application, simulator, or native build to validate.
  • Expected user flow steps and acceptance criteria.
  • QA environment free of unrelated sensitive windows.
  • Owner for final merge or release sign-off.

Production Rules

  • Never run computer use on production admin desktops with live customer data.
  • Approve only apps required for the scoped flow; deny broad Terminal access unless needed.
  • Stop immediately if on-screen content suggests prompt injection or unexpected dialogs.
  • Do not share raw screenshots externally without redaction.
  • Verify another session is not holding the computer-use lock before starting.
  • Treat findings as draft until a human validates screenshots and repro steps.

Review Matrix

Flow type Approval focus Screenshot checkpoint
Native build launch Target app only Post-launch main window
Layout bug repro Target app + resize Clipped state capture
Simulator onboarding Simulator app Each onboarding screen
IDE/build tools Sentinel on Terminal Build success/failure UI
Unexpected system dialog Stop and escalate Dialog before any click

Output Contract

  1. Enablement and permission checklist results.
  2. Apps approved and sentinel warnings reviewed.
  3. Screenshot-indexed step observations.
  4. Stop/rollback actions taken.
  5. Privacy-safe QA summary for human sign-off.

Troubleshooting

Issue: computer-use missing from /mcp Fix: Confirm macOS, v2.1.85+, Pro/Max plan, claude.ai auth, interactive session.

Issue: Permission prompt repeats Fix: Quit Claude Code fully after granting Screen Recording; retry Try again.

Issue: Another session holds the lock Fix: Finish or exit the other session; lock releases when the process ends.

Issue: Text too small after downscaling Fix: Increase in-app font or control size per docs—not display resolution alone.

Duplicate Check

Checked content/skills/, content/guides/, content/agents/, and open PRs. computer-use-from-the-claude-code-cli-for-gui-qa is a step-by-step guide. computer-use-gui-test-agent is a reusable agent prompt. No skills entry provides this computer use GUI QA capability pack with review matrix and output contract grounded in official computer-use documentation.

Editorial Disclosure

Submitted as an independent source-backed HeyClaude content entry by kiannidev. It is based on public Claude Code computer-use documentation and the public Anthropic claude-code repository. No paid placement, referral link, affiliate link, or vendor sponsorship is used.

#claude-code#computer-use#gui-qa#desktop-testing#capability-pack

Source citations

Add this badge to your README

Show that Claude Code Computer Use GUI QA Capability Pack Skill is listed on HeyClaude. Paste this Markdown into your README — it renders the badge and links back to this page.

Listed on HeyClaude
[![Listed on HeyClaude](https://heyclau.de/badge/skills/claude-code-computer-use-gui-qa-capability-pack.svg)](https://heyclau.de/entry/skills/claude-code-computer-use-gui-qa-capability-pack)

How it compares

Claude Code Computer Use GUI QA Capability Pack Skill side by side with 3 alternatives on trust, install, platform support, and disclosed safety notes — all from reviewed registry metadata.

FieldClaude Code Computer Use GUI QA Capability Pack Skill

Expert Claude Code computer use GUI QA capability pack that applies documented computer-use MCP enablement, per-app session approval, screenshot checkpoints, and macOS permission workflows for native app validation—not an official Anthropic QA product.

Open dossier 		Claude Agent SDK Session Storage Capability Pack Skill

Expert Claude Agent SDK session storage capability pack for designing, reviewing, and rolling out Agent SDK session storage with source-backed checklists, production rules, and privacy-safe output contracts.

Open dossier 		Claude Code Chrome Browser QA Capability Pack Skill

Expert Claude Code Chrome browser QA capability pack applying documented Claude in Chrome enablement, browser automation boundaries, and web QA checkpoint workflows.

Open dossier 		MCP Client Config Audit Capability Pack Skill

Expert MCP client config audit capability pack for reviewing Claude Code MCP server entries, scope placement, tool approval settings, env var secrets, and startup context load before enabling servers in production repositories.

Open dossier
Trust
Install riskReview firstReview firstReview firstReview first
Notes Safety ✓ Privacy ✓ Safety ✓ Privacy ✓ Safety ✓ Privacy ✓ Safety ✓ Privacy ✓
Categoryskillsskillsskillsskills
Sourcesource-backedsource-backedsource-backedsource-backed
Authorkiannidevkiannidevkiannidevkiannidev
Added2026-06-162026-06-142026-06-162026-06-14
Platforms
Claude CodeCodexWindsurfGeminiCursorCLI
Claude CodeCodexWindsurfGeminiCursorCLI
Claude CodeCodexWindsurfGeminiCursorCLI
Claude CodeCodexWindsurfGeminiCursorCLI
Source repo————
Safety notes✓Computer use controls real desktop apps with per-app session approval—not sandboxed Bash. Review sentinel warnings before approving Terminal, Finder, or System Settings access. Press Esc or Ctrl+C to stop computer use and release the machine-wide lock immediately. Only one Claude Code session can hold the computer-use lock at a time.✓This skill plans Agent SDK session storage; it must not execute destructive changes without explicit approval. Browser, computer-use, and remote surfaces can access sensitive UI state; scope tests carefully. MCP and SDK integrations may exfiltrate data if tool scopes are too broad. The public `anthropics/claude-code` repository ships documentation links to code.claude.com for settings, security, and integration surfaces. Scheduled or autonomous workflows compound risk; cap blast radius in staging first.✓Browser automation can access logged-in sessions—use staging accounts only. Claude chooses Chrome before computer use for web tasks when configured. Review permission prompts before approving navigation on sensitive pages.✓MCP config files can contain API keys, OAuth client secrets, and bearer tokens that must never be committed to git or pasted into public issues. Enabling write or execute MCP tools in shared repositories expands the blast radius for every contributor using Claude Code in that repo. Project-scoped servers affect all collaborators; user-scoped servers follow individual accounts across repositories unless restricted. Removing an MCP server from config does not automatically revoke OAuth tokens issued to remote vendors. This skill recommends config changes; it must not edit MCP settings or rotate secrets without explicit user approval.
Privacy notes✓Screenshots are downscaled before model upload but may still capture sensitive UI content. Your terminal window is excluded from screenshots per official documentation. Redact customer or proprietary UI details before sharing screenshot evidence externally. Other visible apps are hidden while Claude works; still minimize sensitive windows before starting.✓Reviews may expose integration tokens, customer metadata, and internal URLs related to Agent SDK session storage. Telemetry and analytics configs can include account emails; redact before sharing externally. Keep troubleshooting logs in internal channels unless explicitly sanitized. Third-party vendors remain outside Anthropic retention policies; document separately.✓Browser QA sends page content to the model—avoid customer PII in test flows. Screenshots and DOM snapshots may include session tokens in dev tools—close extras.✓MCP config audits often expose internal service URLs, database hostnames, staging environment names, and account identifiers. Tool schema listings loaded at startup can reveal internal API surface area if copied into public support threads. Shared `.mcp.json` files checked into repositories may leak secrets through git history even after redaction in the current commit. Public audit summaries should describe scope and risk categories, not live credentials or full server manifests.
Prerequisites
  • Interactive Claude Code session on macOS v2.1.85+ with Pro or Max plan per computer-use docs.
  • computer-use MCP server enabled through /mcp for the project.
  • macOS Accessibility and Screen Recording permissions granted when prompted.
  • Disposable QA target app or simulator without live customer data on screen.
  • Access to Claude Code or Agent SDK environment where Agent SDK session storage will run.
  • Ability to read project, user, and managed settings relevant to the workflow.
  • Staging repository or sandbox account for safe validation.
  • Platform or security stakeholder available for policy-bound rollouts.
  • Claude in Chrome extension installed per official Chrome documentation.
  • Staging web app URL and test credentials scoped to non-production.
  • Team policy for browser automation on authenticated sessions.
  • Redacted access to project `.mcp.json`, user MCP settings, or enterprise MCP policy for the workspace under review.
  • A list of MCP servers currently configured and the repositories or teams that depend on them.
  • Permission to inspect Claude Code startup behavior with `/context` or equivalent MCP tool listings.
  • Platform or security stakeholder available to approve config changes affecting production repositories.
Install————
Config————
Citations
ClaimUnclaimedUnclaimedUnclaimedUnclaimed

Signals

Loading live community signals…

More like this, weekly

A short, calm digest of reviewed Claude resources. Unsubscribe any time.