Figma Shot Board

figma-shot-board — screenshots → labeled Figma board

Turn N local images into one tidy Figma page: header (title + subtitle) and labeled sections, each a horizontal row of image frames at their natural size.

Inputs to establish first

• Image files, their order, and a section grouping (section label → files).
• Scale divisor: captures taken at deviceScaleFactor: 2 should be placed at natural size ÷ 2 (logical 1x). Default divisor 2 for playwright 2x captures, 1 for plain images.
• Target: an existing Figma file URL (extract fileKey), or a new file (needs plan/team choice — whoami lists plans; ask if multiple).
• Page: new page name (recommended; never disturb existing pages) or an existing page node-id.

Workflow

1. Preflight: call whoami (auth check + plan keys; also the tool to cite when rate-limited). MANDATORY prerequisite skills: load figma-create-new-file before any create_new_file call, and figma-use before any use_figma call.
2. Target file/page: for a new file use create_new_file (lands in drafts — tell the user it can be dragged into a project). For a page inside a design file use use_figma with figma.createPage() (design files only).
3. Get upload URLs: upload_assets(fileKey, count=N) → N single-use submit URLs (expire in 10 min — upload promptly).
4. POST the files with scripts/post_uploads.sh (multipart file field — the filename becomes the Figma layer name, so name files descriptively before uploading). Keep URL↔file order aligned.
5. Fix placement (the big gotchas) via one use_figma call:
   • Uploaded frames land on the first/current page as 400×300 frames with the image as a FILL (cropped!) — NOT at natural size.
   • For each frame: read fills → figma.getImageByHash(fill.imageHash) → await img.getSizeAsync() → frame.resize(w/divisor, h/divisor).
   • Move to the target page with targetPage.appendChild(frame) (works across pages; collect frames from whatever page they landed on).
6. Build the board (respect ~10 ops per use_figma call — split into 2-3 calls): root figma.createAutoLayout('VERTICAL', {itemSpacing: 56, padding…}) with a light fill, positioned away from (0,0); header (title 28 Semi Bold + subtitle 14 gray); per section a VERTICAL auto-layout (label 18 Semi Bold + HORIZONTAL row, counterAxisAlignItems: 'MIN', itemSpacing 24) and row.appendChild(frame) for its images. Font loads first (Inter styles are Semi Bold, not SemiBold).
7. Verify: await root.screenshot() in the last call and eyeball it. Return the page URL: https://www.figma.com/design/<fileKey>/<name>?node-id=<pageId with ':'→'-'>.
8. Relocation ("move this board to file X"): rebuild in the target file (repeat 3-7 — image hashes are per-file, re-upload is required), then mark the old page/file clearly (rename to "MOVED → …"); the MCP cannot delete files — tell the user to delete the old draft.

Gotcha table

References

• references/scenarios.feature — BDD spec. Read only when auditing or amending this skill; not needed for normal execution.
• scripts/post_uploads.sh — multipart POST loop for submit URLs (usage: post_uploads.sh mapping.tsv where each line is submitUrl<TAB>filepath).

Retrospective

After the board is delivered, reflect on the run:

1. Consider: mid-session corrections (wrong file/team, re-uploads, layout redo)?
2. Ask the user (in Japanese): 「今回のFigmaボード作成のフィードバック (1-5の評価、気になった点、または何もなければEnter)」 If the rating is < 5, ALWAYS follow up: 「なぜその評価ですか？ (改善のために具体的に教えてください)」 and record the answer verbatim as Rating reason.
3. If feedback was given OR corrections occurred: create feedback/ next to this SKILL.md if missing (resolve via git rev-parse --show-toplevel), create feedback/log.md with the standard header if missing, and prepend an entry (Skill Version / Task / Outcome / Rating / Rating reason / Corrections / Issues / User Note). Confirm in one short Japanese sentence.
4. If skipped and no corrections occurred, end without recording.