Agent Skills
› espressif/esp-claw
› cap_llm_inspect_image
cap_llm_inspect_image
GitHub指导 Agent 使用 inspect_image 工具分析设备本地图片,通过指定路径和提示词识别图像内容、文字或场景,适用于视觉分析需求。
Trigger Scenarios
用户询问图片、照片或截图中的内容
需要对本地文件系统中的图像进行视觉分析
Install
npx skills add espressif/esp-claw --skill cap_llm_inspect_image -g -y
SKILL.md
Frontmatter
{
"name": "cap_llm_inspect_image",
"metadata": {
"cap_groups": [
"cap_llm_inspect"
],
"manage_mode": "readonly"
},
"description": "How to inspect a local image with inspect_image."
}
Image Inspection
Use this skill when the user wants the device to inspect a local image and describe what is visible.
When to use
- The user asks what is in an image, photo, screenshot, or camera frame.
- The image already exists on the device filesystem.
- The task needs visual analysis rather than plain file reading.
Available capability
inspect_image: analyze one local image from an absolute path using a prompt that says what to inspect.
Calling rules
- Call
inspect_imagedirectly. - Always pass an absolute local file path in
path. - Always pass a clear
promptthat tells the model what to look for. - Confirm or discover the image path first if it is not already known.
- Do not pass remote URLs or non-image files.
Path guidance
- Prefer real local paths already stored on the device.
- If the exact path is unknown, inspect storage first with file capabilities such as
list_dir. - Common roots in this demo include
<storage_root>/inbox,<storage_root>, or other application-managed storage paths.
Example
{
"path": "<storage_root>/inbox/photo.jpg",
"prompt": "Describe the main objects in this image and mention any visible text."
}
Notes
- Keep the prompt specific. For example: identify objects, read visible text, describe a scene, or check whether a target item appears.
- If the image is blurry or uncertain, report that uncertainty instead of over-claiming.
Version History
- 06eb576 Current 2026-07-24 19:54


