マルチモーダル画像分析

# Multimodal Image Analysis

## When to use

Analyze authorized images for text, objects, scenes, and evidence while separating observations from inference.

## Workflow

1. Confirm the user owns the input or has permission to process it. Collect the source material, audience, output format, and acceptance criteria.
2. Separate observed facts and user-provided constraints from inference. Ask for missing information instead of inventing it.
3. Define the requested observations, extraction fields, coordinate system, confidence needs, and unsupported modalities before analysis.
4. Use <skill-tag>skill:e2013556-b0b7-50ca-880d-5cf16661f953</skill-tag> for the connected analysis workflow. Return evidence, confidence, and limitations separately.

## Output contract

Return the input assumptions, the approved plan, the generated or analyzed result, uncertainty and limitations, and a concise verification checklist. Preserve source links and provider-returned identifiers when available. Stop if the required connected capability is unavailable; do not substitute local scripts, package installation, or embedded credentials.

## Safety and review

Do not upload private or sensitive media without explicit authorization. Do not claim certainty beyond the evidence. Require review before externally visible, paid, destructive, or account-changing actions.

## Source and license

This Aident-native re-authoring is based on [the pinned upstream source](https://github.com/countbot-ai/countbot/tree/2a9def3602c17a0daa1f5e0450112d25d342a69c/workspace/skills/image-analysis), discovered through [ModelScope](https://modelscope.cn/skills/@countbot-ai/image-analysis), and used under MIT. It preserves the upstream workflow goal while removing local executable, credential, and package assumptions.