Set up Studio as a Claude and Codex marketplace for data-liberation

.agents/plugins/marketplace.json

@@ -0,0 +1,20 @@
+{
+  "name": "studio",
+  "interface": {
+    "displayName": "studio"
+  },
+  "plugins": [
+    {
+      "name": "data-liberation",
+      "source": {
+        "source": "local",
+        "path": "./packages/data-liberation-agent"
+      },
+      "policy": {
+        "installation": "AVAILABLE",
+        "authentication": "ON_INSTALL"
+      },
+      "category": "Coding"
+    }
+  ]
+}

.agents/skills/codex.marketplace.json

@@ -0,0 +1 @@
+../plugins/marketplace.json

.agents/skills/marketplace.json

@@ -0,0 +1 @@
+../plugins/marketplace.json

.buildkite/commands/run-lint.sh

@@ -14,3 +14,16 @@ npm run lint
 
 echo '--- :typescript: Typecheck'
 npm run typecheck
+
+# The data-liberation plugin ships its MCP server as a committed esbuild
+# bundle (the plugin installer copies the package from git verbatim — no
+# npm install, no build). Rebuild it and fail if the committed artifact
+# doesn't match src/, so the two can never drift. The build is
+# byte-deterministic for a given lockfile, so a clean tree means fresh.
+echo '--- :package: Verify data-liberation MCP bundle is fresh'
+npm -w data-liberation run build:mcp-bundle
+if ! git diff --exit-code -- packages/data-liberation-agent/dist/mcp-server.bundle.mjs; then
+  echo "^^^ +++"
+  echo "The committed data-liberation MCP bundle is stale. Run 'npm -w data-liberation run build:mcp-bundle' and commit the updated bundle."
+  exit 1
+fi

packages/data-liberation-agent/.claude-plugin/marketplace.json → .claude-plugin/marketplace.json

@@ -1,12 +1,12 @@
 {
-  "name": "data-liberation",
+  "name": "studio",
   "owner": {
     "name": "Automattic"
   },
   "plugins": [
     {
       "name": "data-liberation",
-      "source": "./",
+      "source": "./packages/data-liberation-agent",
       "description": "Extract content from closed web platforms (GoDaddy Websites & Marketing, Hostinger, HubSpot, Shopify, Squarespace, Webflow, Weebly, Wix) into WordPress-compatible WXR files. Inspect, extract, QA, and import to WordPress."
     }
   ]

.gitignore

@@ -114,6 +114,10 @@ cli/vendor/
 dist
 dist-hosted
 dist-local
+# …except the data-liberation plugin's committed MCP server bundle (the plugin
+# installer copies the package without running npm install or a build). The
+# package-level .gitignore keeps everything else in its dist/ ignored.
+!packages/data-liberation-agent/dist
 
 # Playwright traces
 test-results

packages/data-liberation-agent/.claude-plugin/plugin.json

@@ -1,12 +1,12 @@
 {
   "name": "data-liberation",
   "description": "Extract content from closed web platforms (GoDaddy Websites & Marketing, Hostinger, HubSpot, Shopify, Squarespace, Webflow, Weebly, Wix) into WordPress-compatible WXR files. Inspect, extract, QA, and import to WordPress.",
-  "version": "0.2.2",
+  "version": "1.0.0",
   "author": {
     "name": "Automattic"
   },
-  "homepage": "https://github.com/Automattic/data-liberation-agent",
-  "repository": "https://github.com/Automattic/data-liberation-agent",
+  "homepage": "https://github.com/Automattic/studio",
+  "repository": "https://github.com/Automattic/studio",
   "license": "GPL-2.0-or-later",
   "keywords": [
     "content-extraction",
@@ -23,8 +23,5 @@
     "wordpress",
     "wxr"
   ],
-  "mcp": {
-    "command": "npx",
-    "args": ["tsx", "src/mcp-server.ts"]
-  }
+  "mcpServers": "./.mcp.json"
 }

packages/data-liberation-agent/.codex-plugin/plugin.json

@@ -1,11 +1,11 @@
 {
   "name": "data-liberation",
   "description": "Extract content from closed web platforms (GoDaddy Websites & Marketing, Hostinger, HubSpot, Shopify, Squarespace, Webflow, Weebly, Wix) into WordPress-compatible WXR files. Inspect, extract, QA, and import to WordPress.",
-  "version": "0.2.2",
+  "version": "1.0.0",
   "author": {
     "name": "Automattic"
   },
-  "homepage": "https://github.com/Automattic/data-liberation-agent",
+  "homepage": "https://github.com/Automattic/studio",
   "license": "GPL-2.0-or-later",
   "skills": "./skills/",
   "mcpServers": "./.mcp.json",
@@ -20,7 +20,7 @@
       "Read",
       "Write"
     ],
-    "websiteURL": "https://github.com/Automattic/data-liberation-agent",
+    "websiteURL": "https://github.com/Automattic/studio",
     "defaultPrompt": [
       "Inspect a site for liberation",
       "Extract this site into WordPress-compatible WXR",

packages/data-liberation-agent/.gitignore

@@ -4,7 +4,11 @@ skills/replicate-workspace/
 node_modules/
 .DS_Store
 *.log
-dist/
+# dist/ is build output, EXCEPT the committed plugin bundle: the Claude/Codex
+# plugin installer copies this package verbatim (no npm install, no build), so
+# the self-contained MCP server bundle must live in git.
+dist/*
+!dist/mcp-server.bundle.mjs
 docs/plans/
 docs/superpowers/
 qa-screenshots/
@@ -15,6 +19,8 @@ qa-screenshots/
 .worktrees/
 .tmp-test/
 TODOS.md
+# Local discovery log — see CONTRIBUTING.md; intentionally not committed
+DISCOVERIES.md
 
 # Streaming compose-then-install scratch dir
 .tmp-compose/

packages/data-liberation-agent/.mcp.json

@@ -1,9 +1,9 @@
 {
   "mcpServers": {
     "data-liberation": {
-      "command": "npx",
-      "args": ["tsx", "${CLAUDE_PLUGIN_ROOT:-.}/src/mcp-server.ts"],
-      "cwd": "${CLAUDE_PLUGIN_ROOT:-.}"
+      "command": "node",
+      "args": ["${CLAUDE_PLUGIN_ROOT}/scripts/mcp-launcher.mjs"],
+      "cwd": "${CLAUDE_PLUGIN_ROOT}"
     }
   }
 }

packages/data-liberation-agent/AGENTS.md

@@ -50,6 +50,7 @@ When `adminToken` is present, `shopifyAdapter.extract` fetches products via the
 ## Non-obvious Details
 
 - **MCP server runs `tsx src/mcp-server.ts` as ONE long-lived process — editing `src/` does NOT hot-reload it.** ESM modules are cached per process, so a re-called `liberate_*` tool keeps running the code that was loaded at server start. After editing handler/lib source, either restart the MCP server (so all callers pick it up) or, for a quick one-off re-run, drive the handler from a fresh `tsx` process (import the handler, pass a minimal `HandlerContext` with `textResult`/`errorResult`). Unit tests (`vitest`) always use the on-disk source, so they reflect edits immediately — the staleness is only the running MCP server.
+- **`.mcp.json` starts the MCP server through `scripts/mcp-launcher.mjs`: dev checkouts run `src/` via tsx; plugin installs run the COMMITTED bundle `dist/mcp-server.bundle.mjs`.** The launcher only uses the bundle when the package's dependencies don't resolve, so a dev environment never runs a stale dist. The bundle must be committed because plugin installs are plain git copies: Claude's installer restores npm deps only when a per-package `package-lock.json` exists (this workspace package has none), and Codex's installer NEVER restores deps — so `claude/codex plugin marketplace add Automattic/studio` only works with a prebuilt self-contained artifact in the tree. After changing `src/` or dependencies, regenerate with `npm run build:mcp-bundle` and commit the result — CI (Buildkite Lint) rebuilds it and fails on drift. `playwright` and `single-file-cli` stay external to the bundle (guarded dynamic imports that degrade with install guidance). See `scripts/build-mcp-bundle.mjs` for the import.meta.url asset-resolution seam.
 - DLA consumes `@automattic/blocks-engine` through the local pre-publish override `"file:../blocks-engine/packages/blocks-engine"`; standalone `npm install` therefore needs the sibling `blocks-engine` checkout until the package is published. On publish, flip the dependency to the authorized semver range (for example `"^0.x.y"`). Publishing itself remains gated on explicit user authorization.
 - **`tsx scripts/*.ts` that call `page.evaluate` must create the page via `newShimmedPage` from `scripts/_pw.ts`** — tsx's esbuild keepNames rewrites named functions to `__name(...)`, which is undefined in the browser context Playwright serializes into (`ReferenceError: __name is not defined`). The shim installs a `__name` identity via `addInitScript`. vitest-run code is unaffected (its transform doesn't keepNames).
 - WXR builder targets WXR 1.2 spec compliance

packages/data-liberation-agent/CONTRIBUTING.md

@@ -54,7 +54,7 @@ If you can't test against a real site, note that in the PR and explain what you
 
 ### 3. Add to DISCOVERIES.md
 
-This is required. Add a new entry at the top of [DISCOVERIES.md](./DISCOVERIES.md) in this format:
+This is required. DISCOVERIES.md is a local, untracked log (gitignored — it is not committed to the repo); create it at the package root if it doesn't exist yet. Add a new entry at the top in this format:
 
 ```markdown
 ## YYYY-MM-DD — Short title