
Run Xcode Accessibility Audit on a Remote Mac
A worked recipe for running Apple's XCUITest accessibility audit on a hosted Mac: the correct XCUIAutomation signature, a narrow suppression pattern, an xcodebuild job, and a short price worksheet.
Mac hosting workflows first — then website hosting guides. Search across all guides or filter by workflow.
Build, test and run Mac software. Browse the full Mac guide directory.

A worked recipe for running Apple's XCUITest accessibility audit on a hosted Mac: the correct XCUIAutomation signature, a narrow suppression pattern, an xcodebuild job, and a short price worksheet.

Capture the accessibility hierarchy from a running iOS Simulator with the xctree CLI, then diff two builds in Python to surface label, role and value regressions tied to WCAG 2.2.

Two complementary checks for macOS Tab focus: a short XCUITest that runs the documented accessibility audit, and a manual Tab-walk worksheet that writes a baseline focus-order CSV you can diff on every pull request.

A deadline-focused walkthrough for designers renting a Mac: check Adobe Fonts activation, install client-supplied OTFs correctly, and compare a document font manifest before you open the layout.

One xcodebuild flag turns on AddressSanitizer; a short grep pipeline lifts the violations out of the build noise so the stack trace is the first thing you read.

Push an AE render queue to a rented Mac with aerender, a log-based success check, and a Slack notification. Keep your local workstation free.

Stop runaway agent actions before they cause damage. Map every tool call by risk level, add a human gate for destructive operations, and log every approval decision durably.

A practical polling-agent build for a hosted Mac: launchd scheduling, a Retry-After-aware retry loop, a rate-budget worksheet, and the Figment plan that fits.

Three layers keep an overnight agent inside a monthly budget: the provider's own spend cap, a Python ceiling checked after every call, and a launchd watchdog that unloads the job before daily spend crosses your line.

Pair a launchd KeepAlive SuccessfulExit policy with a wrapper that counts relaunches and resumes from a checkpoint, so a crashing agent cannot burn through API quota.

A signed delivery-receipt schema and a short, reviewable reference script for shipping hosted Mac agent results to any API, with a 12-month cost worksheet against CloudMac and a direct Figment plan recommendation.

A direct Brewfile workflow for hosted Mac automation: dump from a known-good machine, detect drift with brew bundle check and cleanup, and restore the same toolchain on a Figment Mac with one command.

A profile-per-account directory layout, a strict account-identifier validator for the launcher, a Playwright collision check that seeds and reads cookies through the public API, and a short operator acceptance checklist for hosted Mac browser agents.

A practical pattern for isolating pasteboard state between hosted Mac jobs: capture-and-clear wrappers, temp-file handoff for secrets, and a verification step.

A per-directory storage budget, an allowlist that protects project data, and a scheduled launchd check that fires before the disk is actually full on a hosted automation Mac.

A small Python detector watches Chrome .crdownload, Firefox .part, and Safari .download temp files, confirms the final size is stable, and optionally checks a SHA-256 before passing the path downstream.

A direct cutover playbook for relocating a hosted Mac automation agent: drain the old machine, export configuration and state, rotate every secret, re-register the CI runner, and verify online status before releasing traffic.

A small alert helper for hosted Mac automation: one-line integration, a 5-minute dedup lock, a consecutive-failure counter, and a Slack-webhook payload that respects how incoming webhooks actually work.

A concrete two-job pattern for human-reviewed Mac automation: the agent uploads a structured review packet, GitHub Actions holds the ship job in a protected environment, and a named reviewer approves or rejects before any downstream write.

A SQLite idempotency-key ledger and a three-delivery worked example that keeps hosted Mac agent jobs safe when the same task arrives more than once.

A concrete SQLite WAL schema, a resume-aware Python runner, and a kill-and-replay test script that proves your hosted Mac agent never re-processes finished work.

A staged checklist for rotating agent API keys on a hosted Mac: issue a new key, store it in Keychain, smoke test, promote, then revoke — with rollback at each step.

A practical multipart upload worksheet for a hosted Mac: initiate, list, and abort S3 parts; use rclone retries; drive a byte-offset checkpoint script; and schedule nightly jobs with launchd.

Agents on a hosted Mac collect credentials in their log stream faster than any cleanup script can chase them. Define safe JSON fields, redact at write time, and run a corpus of known-bad lines in CI before the agent ships.
Figma publishing, AI editing, uploading HTML, moving between platforms.
Seven tools can turn a Figma design into a live website, and they differ most in how much you rebuild along the way. This guide orders them on that one job, then says which tool fits a CMS, code export or staying inside Figma instead.
A new way to update websites starts with the AI you already use. You describe the change. Your AI writes it. One missing step turns that answer into a website.
A report-sharing workflow with a public-data check, a reader test, and a prompt that keeps the report portable.
Claude can produce a website two different ways — a single-page artifact you share as a claude.ai link, or a full set of project files that Claude Code writes for you. Neither one is, by itself, a live site on your own domain. This guide covers what each route actually produces, how an artifact is shared, and the separate deployment step every route needs, sourced from Anthropic's own documentation.
Figma Make and Lovable both turn a prompt into a working prototype or web app with real code, but they start from different places: Make lives inside Figma and can attach an existing design directly, while Lovable is a standalone builder that imports a Figma file through a separate integration. This guide compares the starting point, what each builds, publishing and hosting, code access, and pricing, sourced from each vendor's own pages.
Replit and Bolt both let an AI agent build and deploy an application from a prompt, but Replit wraps that in a full cloud development environment with four distinct deployment types, while Bolt is a narrower builder that publishes straight to its own hosting. This guide compares what each builds, deployment and hosting, GitHub and database, and pricing, sourced from each vendor's own pages.
Base44 and Bolt both turn a prompt into a working application with a database behind it, but they diverge on where a project can start from and how the backend is built. Bolt documents importing an existing GitHub repository as a starting point; Base44's GitHub integration connects an app you already built in Base44, not an external codebase. This guide compares what each builds, backend and database, hosting, code export, and pricing, sourced from each vendor's own pages.
v0 prices Free at $0, Plus at $30/user/month (shown with a $90 reference price), and Business at $100/user/month, plus a custom Enterprise tier and a legacy Premium plan being phased out. This guide covers what each plan unlocks, how v0's prepaid credit system works, and its per-model token rates, sourced from v0's own pricing and docs pages.
Figma has two AI routes of its own - the Figma MCP server and Figma Make - and several outside vendors, including Anima, Locofy, Builder.io and v0, each convert a design to code their own way. This guide covers what each one actually produces and what still needs a developer, sourced from each product's own docs.
Figma-to-HTML plugins split into two structurally different kinds: ones that export code you host yourself, and ones that publish a hosted page directly. This guide covers what each kind actually produces, using two real plugins' own pages as examples, and what to check before you rely on either.
Figma and Framer aren't really fighting for the same job: Figma is a general design and prototyping tool, and Framer is a site builder that happens to include its own design canvas. This guide covers what each does, how a Figma design gets into Framer, and what each costs, from each vendor's own pages.
Figma Sites has no standalone price tag: it's gated to a Full seat on a paid Figma plan, and its custom-domain and bandwidth terms are still governed by a beta policy tied to a 2025 date. This guide covers exactly what Figma's pricing page and help center say.
Figma Make doesn't have its own price tag - it comes bundled with a Full seat on a paid Figma plan, and its real cost shows up as AI credits per month. This guide covers what each plan and seat includes, checked against Figma's own pricing page and help center.
Figma Sites and Framer both let you design on a canvas and publish the result as a real website, with a CMS and custom code layered on top. This is a side-by-side of what each one's own documentation says about the editor, CMS, code, custom domains, and pricing.
A custom 404 page has one real job by Google's own standard: tell the visitor clearly that the page is gone, and help them get somewhere useful. Here's what Google's crawling documentation says about that, what a soft 404 actually is, and the exact steps Squarespace, Wix, and Webflow each document for building one.
An Open Graph image is the picture that shows up when your page's link is shared on social platforms, set with the og:image meta tag from the open ogp.me protocol. This guide covers the tags, the sizes each platform actually documents, and how to check the preview before you share.
Google doesn't state a meta description character limit anywhere in its own documentation. It does explain, in its own words, when it uses your meta description versus page content, and why the displayed snippet gets cut off anyway.
Google's Search Central documentation does not state a character limit for title tags. It does explain, in its own words, where a title link comes from, when Google rewrites it, and why long titles get cut off on the results page.
10Web's AI website builder turns a prompt into a live WordPress site, hosted on 10Web's own managed stack, editable by chat, visual point-and-click, or direct code. This guide covers what it actually generates, its editing modes, and its current plans, sourced only from 10web.io.
A website redesign rarely costs search traffic because the new design is different — it costs traffic because a URL, a page title, or a piece of ranking content quietly disappeared during the move. This guide sets the order that protects against that: inventory what you have, decide what happens to every URL, fix content before layout, test before launch, then confirm the move in Search Console.
OpenAI documents two different ways ChatGPT can produce a website: ChatGPT Sites, a public-beta feature that creates, publishes, and hosts a site with its own URL, and writing/code blocks, which generate portable HTML you deploy yourself. This guide covers what each one does, who can use it, and its documented limits, sourced from OpenAI's own help center.
Cursor and Lovable solve different problems that both involve AI writing code: Cursor is an editor for a codebase you already have or are building by hand, while Lovable generates a full application and hosts it for you from a chat prompt. This guide compares who each is for, where the code lives, hosting, pricing, and what you can take with you.
Bolt and v0 both generate a working, database-backed application from a prompt with real, exportable code, but they diverge on how deeply Git and deployment governance sit inside the product. Once a v0 project is connected to GitHub, every publish goes through a working branch, a pull request, and CI; without GitHub, v0 deploys straight to Vercel. Bolt edits more directly and owns its own hosting and database infrastructure. This guide compares what each builds, Git and deployment mechanics, hosting, databases, and pricing, sourced from each vendor's own documentation.
Lovable prices Free at $0, Pro at $25/month, and Business at $50/month on its default monthly view, plus a custom-priced Enterprise tier. This guide covers what each plan unlocks, how Lovable's single shared credit balance works, and what the annual billing toggle changes, sourced from Lovable's own pricing page.