What problem does it solve? Text-level washing of translated manuals misses sensitive data and errors embedded in screenshots: leftover Korean text, real employee emails, customer names, public IPs, UI copy mistakes, and low-resolution or misplaced images. This Skill automates detection of those issues across hundreds of images in a docx manual. ## Core Features & Use Cases - Automated extraction and mapping: Pulls every image from a docx, maps each to its section and surrounding paragraph context, and computes pixel size, effective DPI, duplicates, and reuse statistics. - OCR-based triage: Runs macOS Vision OCR (Korean/Japanese/English) over all images, then flags only suspicious ones (Korean residue, public IPs, emails, keywords, jpg format, low resolution) so reviewers inspect roughly 10-20% of images instead of all of them. - Capture type classification: Classifies each image into nine types (FULL, CONTENT, MODAL, PANEL, INLINE, DIAGRAM, COVER, CLI, TOAST) to support consistent recapture workflows. - Use Case: Before submitting an English product manual, run the pipeline on the docx, review the flagged images in the triage queue, and produce a severity-classified findings report covering washing omissions, Korean residue, UI copy errors, placement issues, and quality defects. ## Quick Start Run an image audit on the attached manual.docx and give me the triage queue of suspicious screenshots with a findings report.
