migrate-new

/migrate:new

You are starting a new Next.js migration. The user has provided a source URL as the first positional argument and optionally a --source-repo <path> flag.

Step 1 — Ask the three wizard questions

All three are skippable with Enter (accept default).

1. Target directory. Check if the current working directory is empty. If non-empty, ask: "Use current dir or ./[slugified-domain]/?" Default: subfolder if CWD is non-empty, current dir if empty.

2. Source code access (skip if --source-repo was already passed). Ask: "Do you have the source code repo? Path (optional):" Default: skip, use the url-only input mode. If provided, use the url-plus-repo input mode.

3. Pages to migrate. Ask: "Pages to migrate — all discovered pages, or comma-separated URLs/paths? [all]" Default: all. If the user gives paths or URLs, preserve them as ${INITIAL_PAGE_SELECTION} exactly as a comma-separated list. The initial crawl/probe pass normalizes paths against ${URL} and filters the crawl before probing.

Step 2 — Ensure the target scaffold exists before migration state

Before invoking the entry script, check ${TARGET_DIR} for package.json and src/app/layout.tsx.

If either is missing and the target directory is empty, scaffold before invoking the entry script, before .migration/ exists. This ordering avoids create-next-app conflicts with .migration/SESSION_LOG.md and other migration state.

If either is missing and the target directory is not empty, stop before creating .migration/ and print:

Target lacks a Next.js App Router scaffold. Create the scaffold first in [TARGET_DIR], then re-run /migrate:new.

Do not create .migration/ before the scaffold exists.

Step 3 — Invoke the entry script

Run the Node entry point with collected answers:

node --experimental-strip-types ${PLUGIN_DIR}/lib/new-migration.ts \
  --url "${URL}" \
  --target "${TARGET_DIR}" \
  --input-mode "${INPUT_MODE}" \
  --initial-page-selection "${INITIAL_PAGE_SELECTION}" \
  ${SOURCE_REPO:+--source-repo "${SOURCE_REPO}"}

If ${PLUGIN_DIR} is not set by the harness, resolve it from the plugin install path.

Step 4 — Run the LLM inventory decision pass

Before reporting the review artifact, load the draft inventory and raw discovery evidence, then invoke the inventory-decider agent. The agent returns InventoryCorrection[] for semantic names, splits/merges, prop-intent notes, and initial migration decisions.

Apply those corrections with regenerateInventoryArtifacts({ targetDir: TARGET_DIR, corrections, userFeedback: "Initial LLM inventory decision pass" }) so the raw decision is recorded in SESSION_LOG.md and .migration/decisions/.

If blocking names remain, run another LLM decision pass or ask the user for the missing semantic name. Do not approve the inventory automatically.

Step 5 — Report the Component Inventory Review

On success, parse the JSON outcome printed by the entry script and report:

Migration initialized at [TARGET_DIR]/.migration/. Open the Component Inventory Review at [reviewHtmlPath]. When you are ready, describe any name or grouping changes in chat. To approve the inventory, say so in chat.

Treat the review as LLM-led and tool-supported: raw discovery tools gather evidence and enforce approval gates, while the LLM owns semantic grouping, naming, prop intent, migration decisions, and refinement choices based on that evidence.

If the entry script fails (e.g., .migration/ already exists), surface the error message verbatim and stop.