AI商品图一致性审计

Use this skill when a seller has AI-generated or AI-edited product images and wants a repeatable preflight for product fidelity and listing-image risks.

Input and output

• - Input: redacted image inspection facts with asset reference, product reference, dimensions, main-subject ratio, watermarks, promotional text, color/shape drift, rights status, and AI disclosure status.
• Output: ai_product_image_audit.json with per-asset checks, score, blockers, and manual review status.

Procedure

1. 1. Inspect the image with an approved local tool or human reviewer and write only redacted facts to JSON.
2. Run python scripts/audit_product_image_fidelity.py --input image_manifest.json --output ai_product_image_audit.json.
3. Pass the audit and platform requirement profile to the publish-gate skill.

Boundaries and recovery

This skill never uploads an image, edits pixels, removes a watermark, changes a product, or claims legal clearance. Unknown observations become review items. Malformed JSON exits non-zero.

Acceptance

The sample must pass a faithful secondary lifestyle asset, while holding an asset with text, watermark, color drift, or unverified disclosure. Output must show external_publish_executed=false.