AgentSkillsCN

visual-diff

利用 Gemini 的视觉理解能力，对比两张图像，识别其中的差异。当您需要将渲染结果与参考图像进行比对、开展视觉回归测试，或分析同一图像的两个不同版本之间的差异时，可使用此技能。调用方式：/visual-diff <参考图像> <实际图像>。

SKILL.md
--- frontmatter
name: visual-diff
description: Compare two images using Gemini's visual understanding to identify differences. Use when comparing rendered output against reference images, visual regression testing, or analyzing what's different between two versions of an image. Invoke with /visual-diff <reference-image> <actual-image>

Visual Diff with Gemini

Compare two images using Gemini 3 Pro's visual understanding.

Invocation

Parse arguments: /visual-diff <reference-path> <actual-path>

Steps

  1. •Verify both paths exist (use ls or Read tool)
  2. •Run gemini comparison:
bash
gemini --yolo --model gemini-3-pro-preview --output-format text \
  "Compare these two images. First is REFERENCE (expected), second is ACTUAL (rendered).

Report:
1. Overall Match: identical/minor/moderate/major difference?
2. Missing Elements: in reference but not actual?
3. Extra Elements: in actual but not reference?
4. Position Differences: same element, different location?
5. Size/Scale Differences: wrong size?
6. Text Differences: missing, wrong, or different text?
7. Specific Coordinates: approximate pixel locations of issues.

Be specific. Focus on what needs fixing.

Reference: <reference-path>
Actual: <actual-path>"
  1. •Summarize findings to user