Azure AI Vision Image Analysis
Guides Python coding agents in implementing Azure AI Vision image understanding.
The skill recommends DefaultAzureCredential, constrains production credential selection, and clearly labels API-key authentication as legacy, supporting least-privilege credential handling. However, image URLs are sent to Azure, while sensitive OCR and people-detection data flows, retention, user confirmation, and permission scope are not explained, so points are deducted.
The material covers sync/async clients, deterministic cleanup, and HttpResponseError handling. Dependency versions, API compatibility, malformed-input behavior, null-field handling, and diagnosable failure feedback remain thin; several examples assume result fields exist. Static review also caps this dimension at 10, so points are deducted.
Triggers, Python SDK scope, image-analysis scenarios, and major visual features are explicit. Non-fit boundaries, regional/service constraints, Chinese-language practices, and conditions for Azure reachability from mainland-China networks are not established, so points are deducted.
The skill has readable sections for installation, authentication, features, errors, best practices, MIT licensing, version metadata, and linked references. It lacks pinned dependency versions, a changelog, explicit maintenance ownership, and a concrete update path, so it does not receive full marks.
The examples directly support URL/file analysis, OCR, detection, captions, and smart cropping and provide a usable starting point. Outputs are mostly print-oriented, with limited production wrapping, cost/privacy tradeoffs, and no statically verifiable representative results; the static cap is 7, so points are deducted.
Concrete code, feature tables, references, and repository-level CI/test evidence provide some auditability. The supplied material does not show committed tests covering this skill's key paths or third-party execution evidence, so verification remains limited.
- analyze_from_url sends images to Azure; confirm compliance, region, and retention requirements before processing personal, identity-document, or sensitive-text images.
- Dependency versions are not pinned, and the material does not prove that every method and result field works with the current SDK version.
- Quotas, billing, regional availability, Chinese OCR/caption quality, and mainland-China network connectivity are not clearly documented.
What does this skill do, and when should you use it?
This skill guides development with the Azure AI Vision 4.0 Image Analysis Python SDK. It covers captions, dense captions, tags, object detection, OCR, people detection, and smart cropping. It includes synchronous and asynchronous client examples and recommends DefaultAzureCredential for authentication. It is suited to developers integrating Azure image analysis into Python applications or services.
It shows agents how to install azure-ai-vision-imageanalysis, configure VISION_ENDPOINT and authentication variables, create synchronous or asynchronous ImageAnalysisClient instances, and analyze images from URLs or local files. The examples retrieve captions, tags, object and people bounding boxes, OCR lines and words, confidence values, and smart-crop regions. The skill also demonstrates HttpResponseError handling, language selection, gender-neutral captions, context-manager usage, and asynchronous credential cleanup.
- A Python developer needs a single-sentence or multi-region description of an image and selects CAPTION or DENSE_CAPTIONS.
- A team building image search, content labeling, or scene recognition uses TAGS to obtain image content tags.
- An application needs to locate objects or people and use their bounding boxes with OBJECTS or PEOPLE.
- A developer extracts text from screenshots, scans, or photographs with the READ OCR feature.
- A team generates thumbnail compositions for several aspect ratios with SMART_CROPS.
What are this skill's strengths and limitations?
- Covers the primary capabilities of Azure AI Vision 4.0 Image Analysis.
- Includes URL and local-file inputs plus synchronous and asynchronous client examples.
- Clearly recommends DefaultAzureCredential and deterministic client and credential lifecycle management.
- Includes guidance on error handling, language, gender-neutral captions, caching, and selecting only required features.
- It targets the Python SDK and is not directly a guide for other language SDKs.
- It requires an Azure endpoint and credentials or an existing keyed deployment; no offline mode is documented.
- The source does not provide independent test coverage or pricing information for this individual skill.
- Images are subject to limits including 20 MB maximum size and dimensions from 50×50 to 16000×16000 pixels.
How do you install this skill?
Run npx skills add microsoft/skills and select the required skill in the wizard. The README says skills are installed in the selected agent directory; the exact destination is determined by the agent configuration. Install the Python SDK with pip install azure-ai-vision-imageanalysis.
How do you use this skill?
Give the agent a task containing a trigger such as “image analysis,” “computer vision,” “OCR,” “object detection,” “ImageAnalysisClient,” or “image caption.” Configure VISION_ENDPOINT, prefer DefaultAzureCredential in production, and call analyze_from_url or analyze through ImageAnalysisClient with the required VisualFeatures. Input images must meet the formats, size, and dimension limits stated in the source.