virtual-card-art-checker

Virtual Card Art Checker

This skill reviews virtual/digital card art submissions against Visa Digital Card Brand Standards (September 2025) and Rain's internal requirements. It produces a structured compliance report, suggests RGB fallback color values, and generates a visual output image for review.

---

⚠️ #1 REJECTION REASON — 56px Visa Brand Mark Margin

The most common reason Visa rejects card art is the Visa Brand Mark (including "VISA" text and the product identifier like "Signature", "Platinum", etc.) being too close to the card edges. The mark must be at least 56px from the nearest edges — even 1px under is a rejection. The Python script now includes a two-pass Visa Brand Mark distance measurement that:

1. Locates the mark using brightness + density filtering (separates text from decoration)
2. Measures exact pixel distance from the mark to the top/bottom and right card edges
3. Reports FAIL if distance < 56px, WARN if 56-58px (borderline)

Borderline placements (56-58px) have been rejected by Visa in practice — treat WARN as a strong signal to adjust. Always visually verify with extreme scrutiny. When in doubt, fail the card — it is better to send back a card for adjustment than to let a borderline case through to Visa and get rejected.

---

Your job

When a user shares a card art image (or multiple images), you will:

1. Locate or download the image file so you can run technical checks on it
2. Run the technical spec checker (Python script — fast, objective)
3. Visually inspect the image using your vision capabilities
4. Review the suggested RGB fallback colors (background, foreground, label)
5. Generate the Card Art Checker Results image — a single-page PNG with the card art, numbered location markers for warnings/failures, overall status, tech spec table, and visual design compliance table — output as a single-page PDF
6. Show the results image to the user
7. Produce a structured compliance report with a clear pass/fail for every check

Do this even if the user says "quick check" or seems to want a brief review — the full checklist is important and shouldn't be skipped. A card that fails compliance can delay an issuer's launch.

---

Step 1: Get the image file

The image can arrive in three ways. Handle each one:

A) Image attached directly in the conversation

Run the following Python to find the uploads folder and any image files in it — the session name changes every session, so never hardcode it:

import glob, os

# Dynamically find the uploads folder (session name varies every run)
sessions = sorted(glob.glob('/sessions/*/mnt/uploads/'))
uploads_dir = sessions[0] if sessions else None

image_extensions = ['*.png', '*.jpg', '*.jpeg', '*.webp',
                    '*.PNG', '*.JPG', '*.JPEG', '*.WEBP']
image_files = []
if uploads_dir:
    for ext in image_extensions:
        image_files.extend(glob.glob(os.path.join(uploads_dir, ext)))

if image_files:
    print(f"Found: {image_files}")
else:
    print(f"No image files found in: {uploads_dir}")

Use the first image found as the source file for Step 2. If multiple images are present, ask the user which one to check.

If no image file is found in uploads (i.e. the image was pasted inline rather than uploaded via the file picker), do NOT fall back to visual-only mode — instead proceed to inline reconstruction (Step 1B below) to create a working file so the spec checker can run.

A2) Inline reconstruction (when no file is found on disk)

Cowork does not write inline-pasted images to disk — they only exist as base64 in the conversation context. To work around this, reconstruct a working PNG file using PIL based on what you can observe visually. This is not pixel-perfect, but it allows the spec checker to run and eliminates ⚠️ Unverified results wherever possible.

Before writing the reconstruction code, visually inspect the image and determine:

1. Background color — the dominant solid fill color (as R, G, B integers)
2. Color mode — is it RGB (solid, no transparency) or RGBA (has transparent areas)?
3. Approximate dimensions — estimate width × height in pixels:
   • Standard virtual card art is 1536 × 969. If the image looks like a standard landscape card and there's no reason to think otherwise, use these dimensions.
   • If the image is clearly a different size or aspect ratio, estimate accordingly.
4. Filename — derive a sensible filename from the card name/context (e.g. visa_platinum_virtual.png)

Then run this Python, substituting in your observations:

from PIL import Image
import os

# --- Fill these in based on visual inspection ---
bg_color     = (R, G, B)           # e.g. (255, 255, 255) for white
color_mode   = "RGB"               # "RGB" or "RGBA"
width        = 1536                # estimated width in pixels
height       = 969                 # estimated height in pixels
filename     = "card_art_inline.png"
# ------------------------------------------------

out_path = f"/tmp/{filename}"
img = Image.new(color_mode, (width, height), bg_color)
img.save(out_path, "PNG", dpi=(72, 72))
print(f"Saved reconstruction to: {out_path} ({width}x{height}, {color_mode})")

Use out_path as the source file for Step 2. In the final report, mark the Dimensions row as ⚠️ Estimated (not ✅ Pass or ❌ Fail) and add a note: "Inline paste — dimensions estimated visually. Upload via file picker for pixel-accurate verification." All other checks (format, DPI, visual checklist) can be reported normally.

B) URL provided (e.g., a Slack link, Google Drive link, or direct image URL)

Try to download the image using Python:

import requests, os, re
url = "<the URL>"
# Try to infer filename from URL
filename = re.split(r'[/?#]', url)[-1] or "card_art_image"
if '.' not in filename:
    filename += ".jpg"  # fallback extension
out_path = f"/tmp/{filename}"
r = requests.get(url, timeout=15)
r.raise_for_status()
with open(out_path, 'wb') as f:
    f.write(r.content)
print(f"Downloaded to {out_path}")

If the URL is behind authentication (e.g., Slack, Google Drive) and the download fails or returns HTML instead of image bytes, use inline reconstruction (Step A2 above). Note the URL source in the report and mark Dimensions as ⚠️ Estimated.

C) File path provided explicitly

Use the path directly with the spec checker script.

---

Step 2: Run the technical spec checker

Once you have a file path, locate and run the spec checker script. The script is extracted alongside the skill — find it dynamically (session name changes every run):

import glob, subprocess, sys

# Find the spec checker script — never hardcode the session name
scripts = glob.glob('/sessions/*/virtual-card-art-checker/scripts/check_technical_specs.py')
if not scripts:
    scripts = glob.glob('/sessions/*/virtual-card-art-checker-extracted/virtual-card-art-checker/scripts/check_technical_specs.py')

if scripts:
    script_path = scripts[0]
    result = subprocess.run([sys.executable, script_path, "<path-to-image>"],
                            capture_output=True, text=True)
    print(result.stdout)
else:
    print("Script not found — will need to extract from skill zip first")

If the script isn't found, extract the skill zip first:

SKILL_ZIP=$(find /sessions/*/mnt/uploads/ -name "virtual-card-art-checker.skill" 2>/dev/null | head -1)
SESSION_DIR=$(echo "$SKILL_ZIP" | grep -oP '/sessions/[^/]+')
unzip -o "$SKILL_ZIP" -d "${SESSION_DIR}/virtual-card-art-checker/"

This outputs JSON with:

• Pass/fail for: dimensions, file format, DPI (calculated from image resolution)
• Bleed zone analysis: programmatic detection of content in the 56px Visa Brand Mark margin zone — if this check fails, the brand mark is almost certainly extending past the boundary and the card should be flagged as ❌ Fail for the margin check in Step 3
• Suggested RGB fallback colors (background, foreground, label)
• Path to the generated output review image

Save the JSON output — you'll use it in the report.

If Pillow isn't installed, the script installs it automatically.

Note: The script no longer generates a separate review PNG. The results PDF (generated in Step 5) is the sole visual output — it contains the card art with bleed border, PAN overlay, color swatches, and all compliance tables in a single document.

---

Step 3: Visual inspection checklist

With the image in front of you, check each of the following visually. Be thorough — some of these are easy to miss:

✅ Required elements (must be present)

• Visa Brand Mark: clearly visible, legible, not distorted or stretched
• Visa Brand Mark position: in the upper-left or upper-right corner only — no lower-edge placement allowed
• Visa Brand Mark margin (CRITICAL — #1 rejection reason): Every part of the Visa Brand Mark — including the outermost edges of letters like the tips of the "A" in "VISA" and every character in the product identifier ("Signature", "Platinum", "Infinite", "Debit") — must be fully inside the 56px boundary on ALL sides (top, bottom, left, right edges of the card). Zero tolerance: if ANY pixel of the brand mark text touches or crosses the red dashed 56px boundary line shown in the review image, this is ❌ FAIL. Do NOT use approximate language ("looks about right", "appears close enough"). Zoom in on the boundary edges near the brand mark and check each edge explicitly. The script also runs a programmatic bleed zone analysis — if it flags content in the margin zone, treat that as strong evidence of failure. This margin requirement applies ONLY to the Visa Brand Mark, not to other logos or design elements
• Visa Brand Mark size: must match one of the two allowed size options (see below)
• Product identifier present and placed: Every card must display a Visa product identifier — one of Visa Platinum, Visa Signature, Visa Infinite (consumer cards) or Corporate (business/corporate cards). Rain no longer offers the Classic tier, so a product identifier is always required — there is no valid case where it can be absent. The identifier must be placed directly below or immediately adjacent to the Visa Brand Mark, anchored to the same upper corner (upper-left or upper-right). Flag as ❌ Fail if: (a) no identifier is visible on the card, (b) the identifier is in the opposite corner from the Brand Mark, (c) the identifier is separated from the Brand Mark by unrelated graphics or large empty space, or (d) the identifier is anywhere in the lower-left personalization zone. Casing should match the canonical forms; if it deviates (e.g., VISA PLATINUM, signature), include a ⚠️ note in the result.
• Issuer logo: clearly present (may bleed to edge — no margin requirement)

Visa Brand Mark Size Options

Only two size options are permitted. Check which one is used and verify it matches:

Option One — Visa Brand Mark with Debit Identifier (for Debit cards):

Measurement	Value	Description
C	109 px	Height of Visa Brand Mark
D	56 px	Distance from nearest card edge to Visa Brand Mark
E	56 px	Distance from baseline of debit identifier to top of Visa Brand Mark
F	50 px	Minimum height of debit identifier

Option Two — Visa Brand Mark (standalone or with Product Identifier) (for Signature, Platinum, Infinite, etc.):

Measurement	Value	Description
C	142 px	Height of Visa Brand Mark
D	220 px	Distance from top of Visa Brand Mark to baseline of product identifier (when present)
E	56 px	Distance from nearest card edge to Visa Brand Mark

When visually inspecting, estimate whether the Visa Brand Mark height appears to be approximately 109px (Option One) or 142px (Option Two) relative to the card dimensions. The mark should be roughly 11.2% (Option One) or 14.7% (Option Two) of the card height (969px). If it looks significantly smaller or larger than either option, flag it as ❌ Fail.

🚫 Prohibited elements (must NOT be present)

• EMV chip graphic: no chip contact image or chip artwork
• Hologram imagery: no static hologram pictures, foil-style graphics, or dove hologram
• Magnetic stripe graphic: no magnetic stripe imagery
• Cardholder name: no name on the card
• Full PAN / card number: no card number digits
• Expiry date: no expiry date
• Physical card photography: no photograph or highly detailed realistic illustration of a physical card
• Embossed attribute labels: no text labels describing embossed-only features
• Lower-left area clear: the lower-left area of the card is reserved for card personalization (last 4 PAN digits overlay) and must not contain discrete marks or graphics that would compete with or reduce readability of overlaid PAN digits — this includes issuer logos, brand names, icons, text, or opaque design elements. Background patterns and subtle decorative elements ARE allowed — thin lines, curves, gradients, textures, guilloche patterns, and other elements that are clearly part of the card's overall background design do NOT trigger this check, even if they pass through the lower-left zone. The key question is: would the element meaningfully reduce the legibility of white or light-colored PAN digits overlaid on top of it? If the answer is no (e.g., a thin decorative curve on a dark background), this check passes. If the answer is yes (e.g., a logo, icon, or high-contrast graphic), flag as ❌ Fail.
• Design elements clear of product identifier: no artwork, logos, or design elements obscuring or touching the Visa product identifier text (Signature, Platinum, Infinite, etc.)

📐 Layout & orientation

• Orientation: horizontal (landscape) — required for Visa review submissions
• Aspect ratio: looks proportional to a standard credit card (roughly 1.586:1 width:height)

🎨 Design quality

• Color: appears to be full-color (not grayscale or monochrome)
• Legibility: Visa Brand Mark and any text elements are legible at card size
• Visa Brand Mark contrast: the Visa Brand Mark (including the product identifier text — Signature, Platinum, Infinite, Debit, etc.) must have strong color contrast against the card's background. Both the "VISA" wordmark AND the product identifier must be clearly readable. Check that neither blends into or is hard to distinguish from the background color. Common issues: gray/silver text on light backgrounds, dark text on dark backgrounds, or a product identifier in a different color than the Visa wordmark that lacks sufficient contrast. If the Visa Brand Mark or product identifier is hard to read against the card background, flag as ❌ Fail and recommend using white or a higher-contrast color.

ℹ️ Important: Bleed rules

• The 56px margin requirement applies ONLY to the Visa Brand Mark
• Issuer logos, design elements, artwork, and other visuals may extend to the card edge (full bleed is allowed)
• Do NOT flag non-Visa elements for being too close to the edge

---

Step 4: Suggested RGB fallback colors

The spec checker script extracts and suggests fallback colors automatically. These are YOUR suggestions based on analyzing the card design — the issuer should confirm they match their brand intent.

Color	What it's used for	How to identify it
Background color	Shown when the card image can't render (low-bandwidth fallback)	The dominant background/fill color of the card design
Foreground color	Variable values like last 4 PAN digits	A color with strong contrast against the background
Label color	Static labels on the card (e.g., "Debit", "Credit", account type)	A color readable against the background

Review the script's suggestions against what you see in the design. If the card has a complex gradient or pattern, note the most representative solid color. Override the script's suggestion in the report if your visual judgment produces a better match.

---

Step 5: Generate the Card Art Checker Results image

After completing the visual inspection (Step 3) and reviewing the colors (Step 4), generate the full results image. This combines everything — the card art with location markers, overall status, tech spec table, and visual design compliance table — into a single-page PDF.

5a) Construct the visual results JSON

Build a JSON object from your visual inspection findings. The structure is:

{
  "overall_status": "APPROVED | REQUIRES CHANGES | APPROVED WITH NOTES",
  "overall_description": "1-2 sentence summary of findings and what needs fixing",
  "visual_checks": [
    {
      "name": "Check name (from Step 3 checklist)",
      "result": "pass | fail | warning",
      "notes": "Brief explanation (optional for passes)"
    }
  ]
}

Location-based markers: For any failure or warning where a specific location on the card caused the issue, add marker_x and marker_y fields (floats from 0.0 to 1.0) to place a numbered marker on the card art. These fields are only for location-specific issues — not every failure gets a marker.

• marker_x: 0.0 = left edge, 1.0 = right edge of card
• marker_y: 0.0 = top edge, 1.0 = bottom edge of card
• Place the marker at the approximate center of where the issue occurs

Examples of checks that should get markers:

• Visa Brand Mark contrast → marker at the Visa logo position
• Visa Brand Mark margin → marker at the edge where margin is insufficient
• Lower-left area not clear → marker at the offending element in the lower-left
• Design elements near product identifier → marker where elements encroach
• Product identifier present and placed → marker at the identifier's actual position when misplaced; no marker when entirely absent (absence check)

Examples of checks that should NOT get markers:

• Landscape orientation (global, not location-specific)
• Full color / grayscale (global)
• No cardholder name (absence check — nowhere specific to point)
• File format / dimensions / DPI (technical, not visual location)

{
  "name": "Visa Brand Mark contrast",
  "result": "fail",
  "notes": "Silver text on pink background — low contrast",
  "marker_x": 0.75,
  "marker_y": 0.12
}

Each marker gets an auto-assigned number (1, 2, 3, ...) that appears both on the card and in the Ref column of the Visual Design Compliance table, creating a visual legend.

5b) Save the JSON and call the script

Save the visual results JSON to a temp file, then call the script with --visual-results-file:

import json, glob, subprocess, sys

visual_results = {
    "overall_status": "...",       # from your assessment
    "overall_description": "...",  # from your assessment
    "visual_checks": [ ... ]       # from your Step 3 inspection
}

# Save to temp file
results_json_path = "/tmp/visual_results.json"
with open(results_json_path, "w") as f:
    json.dump(visual_results, f)

# Find the script
scripts = glob.glob('/sessions/*/virtual-card-art-checker/scripts/check_technical_specs.py')
if not scripts:
    scripts = glob.glob('/sessions/*/virtual-card-art-checker-extracted/virtual-card-art-checker/scripts/check_technical_specs.py')

if scripts:
    script_path = scripts[0]
    result = subprocess.run(
        [sys.executable, script_path, "<path-to-image>",
         "--visual-results-file", results_json_path],
        capture_output=True, text=True
    )
    print(result.stdout)
    if result.stderr:
        print(result.stderr)

This generates <filename>_card_art_checker_results.pdf in the same directory as the input image.

---

Step 6: Show the results image

The script generates a Card Art Checker Results image (<filename>_card_art_checker_results.pdf). Always show this image to the user using the Read tool. The single-page PNG contains:

1. Card Art Review Pane (top) — the card art with:

   • The 56px bleed border (red dashed line)
   • Sample "•••• 6789" PAN in the suggested foreground color
   • Numbered markers (colored circles) at exact locations where warnings/failures occur
   • RGB fallback color values displayed to the right with swatches

2. Overall Status — colored badge (green/red/orange) with summary description

3. Technical Specifications table — dimensions, format, DPI with pass/fail status

4. Visual Design Compliance table — all visual checks with:

   • Ref column: numbered markers matching those on the card art above
   • Check name, result (colored PASS/FAIL/WARN), and notes
   • Legend note linking the Ref numbers to the card markers

This gives the issuer a complete, self-contained compliance document.

After presenting the results, always offer to open the PDF for the user:

"Would you like me to open the results PDF?"

If the user accepts, open it with: open -a Preview "<path_to_results.pdf>"

---

Step 7: Produce the compliance report

Output a report in this exact format. Be direct — issuers need to know exactly what to fix.

---

Virtual Card Art Compliance Report

File: [filename] Reviewed against: Visa Digital Card Brand Standards (Sept 2025) + Rain requirements

---

Technical Specs

Check	Result	Detail
Dimensions (1536×969px)	✅ Pass / ❌ Fail / ⚠️ Estimated	Actual: [W×H] or "Could not verify — file not on disk"
File format (PNG)	✅ Pass / ❌ Fail / ⚠️ Unverified	Actual: [format] or inferred from filename/URL
DPI (≥72 for digital display)	✅ Pass / ❌ Fail / ⚠️ Unverified	Calculated: [value] DPI (from pixel width ÷ 3.375″)
56px Margin Zone (Visa Brand Mark)	✅ Pass / ⚠️ Warn / ❌ Fail	Distance measurement from Visa Brand Mark to card edges. FAIL < 56px, WARN 56-58px (borderline)

---

Visual Design Compliance

Check	Result	Notes
Visa Brand Mark present	✅ / ❌
Visa Brand Mark position (upper-left or upper-right only)	✅ / ❌	No lower-edge placement allowed
Visa Brand Mark size (Option One: 109px / Option Two: 142px height)	✅ / ❌ / ⚠️	Must match one of the two allowed size options
Visa Brand Mark margin (56px from edges — #1 rejection reason)	✅ / ❌	Zero tolerance — any brand mark content past the 56px boundary is a hard fail. Cross-reference with bleed zone analysis.
Visa Brand Mark contrast against background	✅ / ❌	Both "VISA" and product identifier must be clearly readable
Product identifier present and placed	✅ / ❌ / ⚠️	Required: one of Visa Platinum / Signature / Infinite (consumer) or Corporate (business), placed directly below/adjacent to the Brand Mark in the same upper corner
Issuer logo present	✅ / ❌	May bleed to edge — no margin requirement
No EMV chip graphic	✅ / ❌
No hologram imagery	✅ / ❌
No magnetic stripe graphic	✅ / ❌
No cardholder name	✅ / ❌
No PAN / card number	✅ / ❌
No expiry date	✅ / ❌
No physical card photography	✅ / ❌
Lower-left area clear (no marks/graphics)	✅ / ❌	Reserved for PAN personalization — no logos, icons, or text; background patterns/decorative lines OK
Design elements clear of product identifier	✅ / ❌	No artwork touching Signature/Platinum/Infinite text
Landscape orientation	✅ / ❌
Full color (not grayscale)	✅ / ❌

---

Suggested RGB Fallback Colors

These values are suggested by the checker based on analyzing the card design. They serve as fallbacks when the card image cannot render due to low bandwidth or connectivity issues.

Color Field	RGB Value	Hex	Notes
Background color	rgb([R], [G], [B])	#RRGGBB	[confirm or note if adjusted]
Foreground color	rgb([R], [G], [B])	#RRGGBB	For last 4 PAN digits + variable values
Label color	rgb([R], [G], [B])	#RRGGBB	For static labels on the card

⚠️ These colors are suggested based on analyzing the card design. Please confirm with your designer that they match the intended brand colors before submitting.

Results image: [path to _card_art_checker_results.pdf] — full results with markers, tables, and status.

---

Overall Status

[✅ APPROVED / ❌ REQUIRES CHANGES / ⚠️ APPROVED WITH NOTES]

[1-2 sentence summary. If there are failures, list what needs to be fixed. Be specific.]

---

Notes on edge cases

• RGBA images: Flag as a warning rather than an outright failure — the issuer should confirm transparency is intentional. Recommend converting to RGB with a solid background.
• DPI calculation: DPI is calculated as pixel_width ÷ 3.375 (ISO ID-1 card width in inches). For digital display, the Visa minimum is 72 DPI. At the standard 1536px width, this calculates to ~455 DPI which comfortably exceeds the minimum.
• Missing DPI metadata: Not relevant — DPI is always calculated from pixel dimensions, not read from file metadata.
• Contactless Indicator (⟳ / ))): Allowed per Visa standards, even if the physical card is not contactless enabled. Do NOT flag as a failure.
• Portrait orientation: Only flag if this is being submitted for Visa review. Portrait display in-app is permitted; portrait submission is not.
• Partial card images: Acceptable as long as the user confirms the full image is shown elsewhere in the flow.
• Gradients: Fine. Extract the most representative solid color for the fallback values.
• Bleed to edge: Only the Visa Brand Mark has a 56px margin requirement. All other logos, artwork, and design elements may extend to the card edge without restriction.
• Visa Brand Mark margin (most critical check): This is the #1 rejection reason from Visa. Even partial encroachment — a single letter tip crossing the 56px line — is a hard fail. The script measures exact pixel distance from the Visa Brand Mark to card edges — FAIL if < 56px, WARN if 56-58px (borderline). If the script flags a FAIL or WARN, visually confirm and fail the card. Do NOT pass a card where the brand mark is "close to" or "approximately at" the 56px boundary — if it's borderline, fail it. Visa will reject it.
• Lower-left reserved zone: The lower-left area of the card is reserved for card personalization (last 4 PAN digits). It must be free of discrete, opaque marks or graphics that would reduce PAN readability — no issuer logos, brand names, icons, or text. Background patterns and subtle decorative elements are permitted — thin lines, curves, gradients, textures, and other elements that are clearly part of the card's overall background design may pass through this zone without issue. The test is whether overlaid PAN digits would still be clearly legible. Common failure: issuer logo in the bottom-left. Common false positive to avoid: decorative background curves or lines passing through the zone on a dark background — these do not affect PAN readability and should pass.
• Design elements near product identifier: Artwork, logos, or large design elements must not obscure or touch the Visa product identifier text (Signature, Platinum, Infinite). Flag if elements encroach on the identifier text.
• Product identifier present and placed: Every Rain-issued card must include one of four Visa product identifiers — Visa Platinum, Visa Signature, or Visa Infinite for consumer cards; Corporate for business/corporate cards. Rain no longer offers the Classic tier, so absence of an identifier is an automatic ❌ Fail. The identifier is part of the Visa product lockup and must be placed directly below or immediately adjacent to the Visa Brand Mark, in the same upper corner. Common rejections: identifier missing entirely, identifier drifted into the personalization zone, identifier in the opposite corner from the Brand Mark, or identifier separated from the Brand Mark by unrelated artwork. Verifying the correct identifier for the provisioned product tier requires the program record and is not validated here — but the presence requirement (at least one of the four) is enforced.

---

Reference

Full requirements are in references/visa-requirements.md. Read it if you're unsure about a specific check.