Process Inbox Images

Every listing we create starts here. We transform raw photos in inbox/ into organized listing folders in listings/, grouping images of the same item together based on visual analysis.

Requirements

Before proceeding, you MUST:

1. •Have images in the inbox/ folder at the project root
2. •Commit to analyzing ALL images before grouping - never assume from filenames
3. •Create descriptive folder names that will work as listing titles

Workflow

digraph process_inbox {
    rankdir=TB;
    node [shape=box];

    start [label="Start" shape=ellipse];
    scan [label="1. IMMEDIATELY scan\ninbox for all images"];
    empty [label="Report: No images found" shape=ellipse];
    analyze [label="2. Analyze EVERY image\nIdentify product and features"];
    group [label="3. Group images by item\nSame product = same group"];
    name [label="4. Generate folder name\nBrand + Product + Feature"];
    create [label="5. Create folder in listings/"];
    move [label="6. Move images to folder"];
    more [label="More groups?" shape=diamond];
    done [label="Report results" shape=ellipse];

    start -> scan;
    scan -> empty [label="no images"];
    scan -> analyze [label="images found"];
    analyze -> group;
    group -> name;
    name -> create;
    create -> move;
    move -> more;
    more -> name [label="yes"];
    more -> done [label="no"];
}

Steps

1. Scan Inbox

IMMEDIATELY scan inbox/ for all image files. Supported formats - check for all of these:

• •.jpeg, .jpg, .png, .heic, .webp

If inbox is empty, report "No images found in inbox" and stop.

2. Analyze Each Image

Use the Read tool on EVERY image - no exceptions. For each image, you MUST identify:

• •Product type: What is this? (tool, furniture, electronics, clothing)
• •Brand/model: Visible brand names, logos, model numbers
• •Distinguishing features: Color, size, condition, accessories
• •Scene context: Background, surface, other items visible

Never stop at the first image. Critical grouping information appears across multiple photos.

Grouping signals to look for:

• •Same product from different angles
• •Same background/surface
• •Sequential filenames (IMG_1234, IMG_1235)
• •Matching brand/color

3. Group Images by Item

Create groups where each group represents ONE listing - no exceptions:

• •Images showing the same product from different angles = ONE group
• •Multiple of the same item type = ONE group (e.g., "3 pairs of boots")
• •Unrelated items = SEPARATE groups

Strong grouping signals (trust these):

• •Same exact product visible
• •Same brand and model
• •Same distinctive color/pattern

Weak grouping signals (always verify with image content):

• •Sequential filenames
• •Similar backgrounds

When unsure, prefer smaller groups. We can always merge listings later - splitting them is harder.

4. Generate Folder Names

Create descriptive folder names for each group. Our naming format - follow exactly:

Format: [Brand] [Product Type] [Distinguishing Feature]

Rules - no exceptions:

• •Title case
• •No special characters except hyphens and spaces
• •Max 50 characters
• •MUST be unique in listings/

Examples:

• •DeWalt Cordless Drill 20V
• •Black Metal Outdoor Bench
• •Vintage Radio Shack Electronics Kits
• •Nike Running Shoes Size 10

5. Create Folders and Move Images

For each group:

1. •IMMEDIATELY create the folder in listings/
2. •Move all grouped images from inbox/ to the new folder

Always use mv to move files - never cp. Images MUST be removed from inbox after processing.

6. Report Results

After processing, ALWAYS report what we did:

Processed X images into Y listing folders:

1. [Folder Name] (N images)
   - IMG_xxxx.jpeg
   - IMG_xxxx.jpeg

2. [Folder Name] (N images)
   - IMG_xxxx.jpeg

Never skip the final report - the user needs to know what happened.

Common Mistakes

Edge Cases

Existing folder with same name: Append a number (e.g., DeWalt Drill 2) or add a distinguishing feature. Never overwrite.

Unidentifiable items: Use descriptive names based on what's visible: Blue Ceramic Vase, Vintage Metal Tool. Never use generic names like "Item" or "Unknown".