.lovable/plan.md

Goal

For every image included in the downloaded ZIP, also add:

• <same-basename>.txt — UTF-8 tags file (space-separated, one line, matches the image filename exactly except for extension).
• <post-id>.md — human-readable markdown containing all extracted post metadata (all tags, source link — blank if missing, file size, sample link, original file link, rating, score, dimensions, uploader, created date, md5, etc.).

Approach

The server currently only receives {id, fileUrl, ext} per post. To write rich metadata into the ZIP without a second network round-trip per post, the frontend will forward the full normalized post (plus a raw metadata bag) to /api/download-zip, and the server will emit the extra files alongside each image.

To capture "any data related to that post", searchPosts will additionally return the raw JSON entry for each post (as raw: unknown). The Markdown generator will pretty-print known fields first, then append the full raw JSON as a fenced code block so nothing is lost.

New optional fields (sampleUrl, fileSize, width, height, md5, uploader, createdAt) get added to SiteConfig.fields as optional dotted paths, defaulted for the built-in e621 site. Existing custom sites keep working because every new field is optional.

Files to change

src/lib/grabber/sites.ts

• Extend FieldMap with optional: sampleUrl?, fileSize?, width?, height?, md5?, uploader?, createdAt?.
• Extend NormalizedPost with the same optional fields plus raw: unknown.
• Update normalizePost() to resolve each new path when present and attach raw (the original entry).
• Update BUILTIN_SITES e621 entry fields with: sampleUrl: "sample.url", fileSize: "file.size", width: "file.width", height: "file.height", md5: "file.md5", uploader: "uploader_id", createdAt: "created_at".

src/lib/grabber/search.functions.ts

• Extend the SiteSchema.fields zod object with the new optional string paths.
• No handler logic changes needed — normalization already carries raw and new fields through.

src/components/grabber/AddSiteDialog.tsx

• Add the new optional field keys to the (["id",...] as const) list on line 73 so users can map them for custom sites (sampleUrl, fileSize, width, height, md5, uploader, createdAt).
• Extend DEFAULT (lines 22–31) with sensible Danbooru mappings (sampleUrl: "large_file_url", fileSize: "file_size", width: "image_width", height: "image_height", md5: "md5", uploader: "uploader_name", createdAt: "created_at").

src/routes/api/download-zip.ts (major rewrite)

• Expand BodySchema.posts item to accept the full normalized post shape (all new optional fields + tags: string[] + raw: unknown).
• For each post:
  1. Fetch the image bytes as today. Build image filename base = "<siteName>_<id>" and <base>.<ext>.
  2. Write <base>.txt — strToU8(tags.join(" ")) (UTF-8). Uses the exact same basename as the image so tag files pair 1:1.
  3. Write <id>.md — pretty markdown with:
     • # Post <id> heading
     • Table/list of: Source (blank line if missing), Rating, Score, Dimensions (width x height), File size (human-readable KB/MB), MD5, Uploader, Created at, Original file URL, Sample URL, Preview URL, Ext.
     • ## Tags section with the tags space-separated.
     • ## Raw metadata fenced ```json block containing JSON.stringify(raw, null, 2).
• Keep the existing _errors.txt behavior and 200-post cap.
• Sanitizer stays the same; md/txt names use sanitize(id) so filenames match the image exactly.

src/routes/index.tsx

• In downloadZip() (the fetch("/api/download-zip", …) call), send the full post objects instead of a stripped {id,fileUrl,ext} triple. Change:

  posts: chosen.map((p) => ({ id: p.id, fileUrl: p.fileUrl, ext: p.ext })),
  

  to:

  posts: chosen, // full NormalizedPost incl. tags + raw + optional metadata
  

• No other UI changes.

ZIP layout example

grabber-2026-07-17T...zip
├── e621_1234567.jpg
├── e621_1234567.txt          ← "tag1 tag2 tag3 ..."
├── 1234567.md                ← human-readable metadata + raw JSON
├── e621_1234568.png
├── e621_1234568.txt
├── 1234568.md
└── _errors.txt               (only if any fetch failed)

Notes / non-goals

• No new dependencies (still fflate + zod).
• No DB, no auth, no schema changes — this is presentation + payload shape only.
• Custom sites without the new mappings still work; missing fields render as blank lines in the .md and the raw JSON section still contains everything from the API.