# Liberating sites with Studio Code (Bossenger) > [!question] Lingering Questions > - Is the repeated database corruption caused by media write-volume, or by Studio's Playground/SQLite "sandbox" site mode specifically — Bossenger raises both hypotheses and resolves neither on-air. > - Does the visual-regression "verify" step compare against the live source site, or only against screenshots captured earlier in the same run? If the latter, a genuinely broken element could pass verification simply by also being absent from the screenshot. > - Will Automattic ever state a public policy on vendor-neutral plugin recommendations (WooCommerce auto-included, Events Calendar auto-suggested) inside an AI migration tool it owns commercially? > [!tip] Article Ideas > - "What does it take to trust an unattended AI agent with your client's site?" — using the /liberate stream as a live case study in progress-reporting, checkpoints, and failure transparency. > - A short teaching piece on the discovery → checkpoint → extract → import → verify pipeline as a reusable pattern for any checkpointed AI agent skill, independent of WordPress. > [!link] Cross Reference > - [[3. Resources/MOCs/AI Agent Safety MOC]] — silent, unattended, destructive-adjacent failure (database corruption) without per-step user checkpoints > - A separate community livestream on Studio Code by a streamer referred to as "Welcher" covers related ground — no URL captured yet; link once located > [!quote] Vocabulary > - **Data Liberation** — the umbrella WordPress.org project (Matt Mullenweg-announced) aimed at making platform-to-WordPress migration frictionless; distinct from `/liberate`, its Studio Code command surface > - **theme replication vs. blocks and products** — the two `/liberate` reconstruction paths: a high-fidelity non-editable HTML clone vs. native blocks + real WooCommerce product data ("best launchpad for redesign") ## Metadata - Source: [[3. Resources/Readwise/Articles/Liberating Sites With Studio Code|Liberating sites with Studio Code]] - Author: Jonathan Bossenger (WordPress developer, Automattic Studio team) - Type: YouTube livestream (video) - URL: https://www.youtube.com/watch?v=Wz9k1qtExZg ## Author's thesis Studio Code's `/liberate` slash command — an AI agent built on the WordPress Data Liberation project — is a real, working first draft of unattended site migration from Wix/Squarespace/Shopify-class platforms into WordPress: it correctly detects source platforms, offers a meaningful reconstruction-path choice, and pulls structured content through per-platform MCP adapters into a standard WordPress import. On a simple, low-media site it delivers a genuinely useful ~30-minute migration at 50–75% fidelity. On a media-heavy site — arguably the more realistic case for anyone leaving a proprietary builder with years of accumulated photos and video — it repeatedly corrupts its own local database and cannot yet be trusted unattended. ## Key arguments (≤3) 1. **The mechanism works as designed**: discovery → mandatory reconstruction-path checkpoint → per-platform MCP-adapter extraction → WXR import → screenshot-based verification is a coherent, publicly-documented (GitHub `automattic/studio`) pipeline that produces useful byproducts (static HTML mirrors, before/after screenshots) independent of whether the WordPress import itself succeeds. 2. **Reliability tracks media volume, not platform type**: the simple 2-page Wix site completed in ~30 minutes at 50–75% fidelity; the 220+-media-file site corrupted its local SQLite database repeatedly and was aborted live — the single most consequential finding for anyone assessing production-readiness. 3. **An unresolved commercial-incentive question sits inside the tool itself**: theme-replication runs still auto-include WooCommerce (Automattic-owned) while recommending a single competing plugin (The Events Calendar) by name rather than linking a category page — a self-acknowledged ("a marketing decision, not a me decision") but unaddressed conflict of interest. ## Claim atoms ### `/liberate` migrates any site into WordPress via MCP adapters > [!quote] ID:rw_h1 > liberate essentially uses studio code to migrate and rebuild a site from whatever platform into WordPress Establishes the base mechanism — agent + per-platform MCP adapters + reconstruction-path checkpoint — that every other atom depends on. [06:15] ### Media-heavy sites trigger database corruption during migration > [!quote] ID:rw_h2 > the database encountered corruption errors. Database disk images malformed likely triggered by First occurrence on the low-media site during rewrite-rule flush [34:07]; recurs more severely on the 220-media-file site [86:47], ending the live test entirely. ### Small-site migration takes about thirty minutes > [!quote] ID:rw_h3 > it took 18,076 seconds. So 18,76 seconds in minutes. So that's approximately half an hour that it took Concrete timing benchmark for the simplest test case (2-page Wix site, low media). [55:34] ## Permanent candidates - **AI migration agents should report step-level progress, not silent waiting** — the same silence-during-long-operations complaint recurs across both test runs, independent of the corruption bug itself. Strong, portable candidate — promoted below. - The sandbox-vs-native-PHP hypothesis for the corruption cause is a real open question but unconfirmed by Bossenger himself; hold as a fleeting note pending the team's own diagnosis. ## Processing checklist - [x] Verify no existing Literature Note covers this doc (checked, none found) - [x] Extract claim atoms and permanent candidate from the existing deep-read - [ ] Confirm with the `automattic/studio` GitHub repo whether the database-corruption bug has since been filed or fixed - [x] Connect into the relevant MOC — [[3. Resources/MOCs/AI Agent Safety MOC]] ## Links - [[3. Resources/MOCs/AI Agent Safety MOC]] — an unattended agent that corrupts its own working state mid-run, with no per-step visibility, is exactly the trust-calibration failure this MOC tracks - [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]] — applies-in: agentic migration as a category of AI doing structural, judgment-bearing work rather than assistive completion