Build your own reading archive (a personal Readwise)

Build the starter for a local reading archive web app. Requirements:
- Next.js 15 with TypeScript and the App Router, plain CSS modules, no UI kit.
- SQLite stored at data/archive.db using Drizzle ORM, migrations generated with drizzle-kit.
- Schema in src/db/schema.ts defining a highlights table with id, text, author, title, source_file, location, url, tags_json, captured_at, times_seen, next_review_at.
- Home route at / showing row counts read from the highlights table.
- DATABASE_PATH read from .env, ship .env.example, never commit credentials.
- One command, npm run setup, that creates the database, applies migrations, and prints next steps.
- Out of scope: login screens, dark mode, component libraries, Docker.
- Expect drizzle-kit config path trouble, verify the migrations folder exists before moving on.

Add a highlight importer that reads user-exported files into the highlights table in src/db/schema.ts. Requirements:
- POST /api/import accepts multipart uploads of .json, .csv, and .txt files up to 10 MB.
- Parse three formats: JSON arrays of highlight objects, CSVs with text and author columns, and Kindle My Clippings.txt blocks separated by ========== lines.
- Skip duplicates by inserting only rows whose sha256 hash of text plus author is new.
- Page at /import shows an upload form and reports rows imported and duplicates skipped per run.
- Create fixtures/sample-clippings.txt with 5 clippings and fixtures/sample-export.json with 10 highlights, and a test asserting the combined run inserts 15 rows.
- Strip stray page headers and fix encoding artifacts defensively, clippings files are always messy.
- Out of scope: calling Kindle or Readwise APIs directly, OCR from screenshots.

Add tagging, full-text search, and the review queue on top of the importer from step 2. Requirements:
- Full-text search over text and author using SQLite FTS5, exposed at /search?q= with highlighted matches.
- Inline tag editing per highlight, persisted to the tags_json column in src/db/schema.ts.
- Page at /review serving up to 10 due highlights per day, ordered by next_review_at ascending.
- Four grading buttons, Again, Hard, Good, Easy, updating next_review_at with simplified SM-2 intervals of 1, 3, 7, and 14 days.
- Header counter on /review showing how many highlights are due today and how many were reviewed this week.
- Keep search server-side with FTS5 match queries, do not load the whole table into the client.
- Out of scope: Anki sync, configurable algorithm settings screens, streak gamification.

Add capture paths that insert new rows into the highlights table built in step 1. Requirements:
- POST /api/capture accepts JSON with url, title, selection, and tags, validated with zod, inserting with source_file set to bookmarklet.
- Page at /capture renders the bookmarklet snippet as a draggable link that posts the current tab's url, title, and selected text.
- A manual form on /capture for pasting a quote, author, and URL when the bookmarklet misbehaves.
- Rate limit /api/capture to 30 requests per minute per IP to stop runaway loops.
- Store the URL string only, do not fetch page content, paywalls and bot defenses make scraping unreliable and off limits.
- Return clear JSON errors with status codes so the bookmarklet can surface failures.
- Out of scope: browser extensions, mobile share sheets, screenshot capture.

Add exports that cover every column of the highlights table populated by steps 2 through 4. Requirements:
- Write Markdown files to exports/markdown/, grouped into one file per title, filenames slugged from the title with author and date in frontmatter.
- Write exports/backup.json containing every row and column of the highlights table.
- GET /api/export/csv streams CSV with exactly the columns the step 2 importer parses, so imports and exports round trip.
- Page at /export offers three buttons and reports the file count written after each run.
- Slug titles safely, reject path traversal characters, and overwrite cleanly on repeat exports.
- Verify the JSON round trip: restoring exports/backup.json into a fresh database yields identical row counts.
- Out of scope: Notion or Obsidian API sync, scheduled cloud backups.