Xiaobenyang Image Toolkit
Paste the following prompt into your AI chat to install this skill:
Please install @org-rn88lg3j/pic using the guide at https://skillhub.cn/install/skillhub.md.
About this skill
What problem it solves
In engineering workflows, image information often appears as scattered OCR text, document fields, bounding boxes, and classification labels. If a multimodal model simply receives an image, the output may be inconsistent, hard to validate, or missing structured fields for sensitive documents. Xiaobenyang Image Toolkit wraps these capabilities into Python tool functions that an agent can call, while the LLM handles intent understanding and routing.
How it works
The core capabilities fall into four groups:
- General OCR: scripts.tools.ocr and scripts.tools.ocr_pro accept image URLs or Base64 payloads for documents, screenshots, and signage.
- Document and business OCR: ID cards, passports, HK/MO/TW travel permits, bank cards, business licenses, driver licenses, and vehicle licenses return structured fields, with validation such as ID or bank-card checksums.
- Detection and recognition: pedestrians, vehicles, fire, phones, e-bikes, helmets, reflective vests, gestures, birds, insects, plants, dishes, and pet emotions, returning boxes, confidence scores, or labels.
- Vision-model tasks: ImageNet classification, COCO object detection, oriented object detection, human pose estimation, and instance segmentation for tasks requiring boxes, masks, or keypoints.
The workflow is: check XBY_APIKEY, select the matching scripts.tools.* function, extract required parameters such as dataUrl or dataBase64, call it, and normalize the raw response before presenting it. If parameters are missing, ask the user instead of guessing.
Boundaries
The toolkit depends on an external API rather than local offline inference. Before use, ensure the key is available and that sensitive images are handled according to permissions and compliance requirements. It covers recognition, detection, and classification tasks well, but is not a substitute for complex cross-image reasoning or private-domain model training. Results can still vary with image quality, occlusion, or resolution, so production use should include human review.
Use Cases
- Reviewers extract fields from ID, passport, and bank card photos and validate number checksums.
- QA engineers inspect production screenshots for fire, helmets, and reflective vests with boxes and risk labels.
- Data engineers structure text, license plates, and object labels from screenshots for database import.
- Algorithm engineers compare COCO detection, pose estimation, and segmentation outputs across models.
Best For
- Backend engineers in document review who need stable JSON extraction from photos.
- Algorithm engineers analyzing surveillance frames who need fire, pedestrian, and vehicle boxes.
- Application engineers building multimodal agents who need intent-based vision API routing.
- Data engineers processing receipts and form images who need OCR, plates, and labels for import.
Related Skills
Organizes files by extension into subfolders like Documents, Code, and Archives, then outputs a report.
Extract tables, formulas, charts, and layout from invoices, reports, papers, and multi-column documents.
Generates a multi-sheet Excel report containing only structured data tables from byteplan-analysis results.
Automatically sort directory files into type-based folders, with dry-run preview, reports, and JSON custom rules.