功能描述

对指定账号下设备通道的实时抓图地址进行AI智能场景分析。AI scene analysis for device channel snapshot URLs under an Imou account. 支持：人形检测、抽烟检测、玩手机检测、工装检测与离岗检测（需预先配置工装模板）、货架检测、垃圾检测、热力图数据...

使用说明 (SKILL.md)

\r \r

Imou Multimodal Analysis\r

Name: Imou Open Multimodal Analysis
Author: imou-openplatform

\r AI scene analysis for device channel snapshot URLs (or any image URL) under an Imou account: human detection, smoking detection, phone-using detection, workwear and absence detection (with pre-configured workwear repository), shelf detection, trash detection, heatmap statistics, and face analysis. Also manage detect repositories and targets (create/list/delete repository; add/list/delete targets).\r \r

Quick Start\r

\r Install dependency:\r

pip install requests\r
```\r
\r
Set environment variables (required):\r
```bash\r
export IMOU_APP_ID="your_app_id"\r
export IMOU_APP_SECRET="your_app_secret"\r
export IMOU_BASE_URL="your_base_url"\r
```\r
\r
**API Base URL (IMOU_BASE_URL)** (required; no default—must be set explicitly):\r
- **Mainland China**: Register a developer account at [open.imou.com](https://open.imou.com) and use the base URL below. Get `appId` and `appSecret` from [App Information](https://open.imou.com/consoleNew/myApp/appInfo).\r
- **Overseas**: Register a developer account at [open.imoulife.com](https://open.imoulife.com) and use the base URL for your data center (view in [Console - Basic Information - My Information](https://open.imoulife.com/consoleNew/basicInfo/myInfo)). Get `appId` and `appSecret` from [App Information](https://open.imoulife.com/consoleNew/myApp/appInfo). See [Development Specification](https://open.imoulife.com/book/http/develop.html).\r
\r
| Region         | Data Center     | Base URL |\r
|----------------|-----------------|----------|\r
| Mainland China | —               | `https://openapi.lechange.cn` |\r
| Overseas       | East Asia       | `https://openapi-sg.easy4ip.com:443` |\r
| Overseas       | Central Europe  | `https://openapi-fk.easy4ip.com:443` |\r
| Overseas       | Western America | `https://openapi-or.easy4ip.com:443` |\r
\r
**Note**: AI APIs are value-added; apply for access via Imou if needed.\r
\r
Run analysis on an image URL (e.g. device channel snapshot from live `coverUrl` or any accessible URL):\r
```bash\r
# Human detection\r
python3 {baseDir}/scripts/multimodal_analysis.py analyze HUMAN "https://example.com/snapshot.jpg"\r
\r
# Smoking detection\r
python3 {baseDir}/scripts/multimodal_analysis.py analyze SMOKING "https://example.com/snapshot.jpg"\r
\r
# Phone-using detection\r
python3 {baseDir}/scripts/multimodal_analysis.py analyze PHONE "https://example.com/snapshot.jpg"\r
\r
# Workwear detection (optional repositoryId and threshold)\r
python3 {baseDir}/scripts/multimodal_analysis.py analyze WEAR "https://example.com/snapshot.jpg" [--repository-id REPO_ID] [--threshold 0.8]\r
\r
# Absence detection (requires workwear repositoryId)\r
python3 {baseDir}/scripts/multimodal_analysis.py analyze ABSENCE "https://example.com/snapshot.jpg" --repository-id REPO_ID [--threshold 0.8]\r
\r
# Shelf detection\r
python3 {baseDir}/scripts/multimodal_analysis.py analyze SHELF "https://example.com/snapshot.jpg"\r
\r
# Trash detection\r
python3 {baseDir}/scripts/multimodal_analysis.py analyze TRASH "https://example.com/snapshot.jpg"\r
\r
# Heatmap (threshold required; optional exclude repository IDs)\r
python3 {baseDir}/scripts/multimodal_analysis.py analyze HEATMAP "https://example.com/snapshot.jpg" --threshold 0.8 [--exclude-repos ID1,ID2]\r
\r
# Face analysis\r
python3 {baseDir}/scripts/multimodal_analysis.py analyze FACE "https://example.com/snapshot.jpg"\r
```\r
\r
Repository and target management:\r
```bash\r
# Create detect repository (face | human for workwear)\r
python3 {baseDir}/scripts/multimodal_analysis.py repo create "MyWorkwearLib" human\r
\r
# List repositories (paginated)\r
python3 {baseDir}/scripts/multimodal_analysis.py repo list [--page 1] [--page-size 20]\r
\r
# Delete repository\r
python3 {baseDir}/scripts/multimodal_analysis.py repo delete REPOSITORY_ID\r
\r
# Add target to repository (image URL or base64 type)\r
python3 {baseDir}/scripts/multimodal_analysis.py target add REPOSITORY_ID "TargetName" "https://image.url" [--type url]\r
python3 {baseDir}/scripts/multimodal_analysis.py target add REPOSITORY_ID "TargetName" "BASE64_DATA" --type base64\r
\r
# List targets in repository\r
python3 {baseDir}/scripts/multimodal_analysis.py target list REPOSITORY_ID [--page 1] [--page-size 20]\r
\r
# Delete target from repository\r
python3 {baseDir}/scripts/multimodal_analysis.py target delete REPOSITORY_ID TARGET_ID\r
```\r
\r
## Capabilities\r
\r
1. **Human detection**: Detect whether the image contains human figure(s).\r
2. **Smoking detection**: Detect whether someone is smoking in the image.\r
3. **Phone-using detection**: Detect whether someone is using a phone.\r
4. **Workwear detection**: Detect whether personnel are in compliance with workwear (optional workwear repository and threshold).\r
5. **Absence detection**: Detect absence from post (requires pre-configured workwear repository).\r
6. **Shelf detection**: Detect shelf status (e.g. empty/full).\r
7. **Trash detection**: Detect trash overflow.\r
8. **Heatmap**: Get heatmap statistics for regions (threshold required; optional exclude repository IDs to filter by workwear).\r
9. **Face analysis**: Face detection/analysis.\r
10. **Detect repository**: Create (face/human), list by page, delete.\r
11. **Target in repository**: Add (URL or Base64), list by page, delete.\r
\r
## Request Header\r
\r
All requests to Imou Open API must include the header `Client-Type: OpenClaw` for platform identification.\r
\r
## API References\r
\r
| API | Doc |\r
|-----|-----|\r
| AI overview | https://open.imou.com/document/pages/f1b9a3/ |\r
| Dev spec | https://open.imou.com/document/pages/c20750/ |\r
| Get accessToken | https://open.imou.com/document/pages/fef620/ |\r
| humanDetect | https://open.imou.com/document/pages/93rflk/ |\r
| smokingDetect | https://open.imou.com/document/pages/kf70sq/ |\r
| phoneUsingDetect | https://open.imou.com/document/pages/jf78o9/ |\r
| workwearDetect | https://open.imou.com/document/pages/2jisd8/ |\r
| absenceDetect | https://open.imou.com/document/pages/29dicv/ |\r
| shelfStatusDetect | https://open.imou.com/document/pages/2oud87/ |\r
| trashOverflowDetect | https://open.imou.com/document/pages/cdmfd6/ |\r
| heatmapDetect | https://open.imou.com/document/pages/fdjfg9/ |\r
| faceAnalysis | https://open.imou.com/document/pages/28d7ug/ |\r
| createAiDetectRepository | https://open.imou.com/document/pages/34ff11/ |\r
| listAiDetectRepositoryByPage | https://open.imou.com/document/pages/5e8222/ |\r
| deleteAiDetectRepository | https://open.imou.com/document/pages/5esi8a/ |\r
| addAiDetectTarget | https://open.imou.com/document/pages/ikdf78/ |\r
| listAiDetectTarget | https://open.imou.com/document/pages/278dkj/ |\r
| deleteAiDetectTarget | https://open.imou.com/document/pages/odty82/ |\r
\r
See `references/imou-ai-api.md` for request/response formats.\r
\r
## Tips\r
\r
- **Token**: Fetched automatically per run; valid 3 days. Do not cache across runs unless you implement expiry handling.\r
- **Image input**: Use `type` "0" for image URL, "1" for Base64. Snapshot URL can be device channel live cover URL (e.g. from imou-open-device-video skill `liveList` / `bindDeviceLive` streams[].coverUrl) or any accessible image URL.\r
- **Workwear / Absence**: Create a human-type repository first, add workwear target images, then pass `repositoryId` to workwearDetect and absenceDetect.\r
- **Heatmap**: `threshold` in (0,1]. Use `excludeRepositoryIds` to exclude matched workwear persons (e.g. staff) and count only valid customers.\r
- **detectRegion**: Optional; up to 3 regions, each 3–6 points (normalized 0–1). Omit to analyze full image.\r
\r
## Data Outflow\r
\r
| Data | Sent to | Purpose |\r
|------|---------|--------|\r
| appId, appSecret | Imou Open API | Obtain accessToken |\r
| accessToken, image URL or Base64, repositoryId, threshold, etc. | Imou Open API | AI detection and repository/target management |\r
\r
All requests go to the configured `IMOU_BASE_URL`. No other third parties.\r

安全使用建议

This skill appears to do what it says: it calls Imou Open API endpoints to analyze images and manage detection repositories. Before installing: 1) Understand that IMOU_APP_ID/IMOU_APP_SECRET are full API credentials — a compromised key lets the skill list/create/delete repositories and upload/delete targets (including face/workwear data). 2) Consider using a dedicated Imou developer account with limited data and permissions for this integration. 3) Be mindful of privacy and legal rules for biometric/face data in your jurisdiction; do not upload sensitive images without consent. 4) Note the small doc inconsistency about IMOU_BASE_URL; the code falls back to a default base URL. 5) If you want extra caution, review the included Python files (imou_client.py, multimodal_analysis.py) and run them in an isolated environment (container or limited VM) before granting production credentials.

功能分析

Type: OpenClaw Skill Name: imou-multimodal-analysis Version: 1.0.0 The skill is a standard API wrapper for Imou's AI multimodal analysis services, including human detection, face analysis, and repository management. The code in `imou_client.py` and `multimodal_analysis.py` correctly implements the Imou Open API specification, using environment variables for credentials and communicating exclusively with official Imou endpoints (e.g., `openapi.lechange.cn`). No evidence of malicious intent, data exfiltration, or prompt injection was found.

能力评估

✓ Purpose & Capability

Name/description match what the code implements. The scripts call Imou Open API endpoints for human/smoking/phone/workwear/absence/shelf/trash/heatmap/face analysis and repository/target management; the two required env vars (IMOU_APP_ID, IMOU_APP_SECRET) are appropriate for this purpose.

ℹ Instruction Scope

Runtime instructions only reference the declared env vars and run the included Python CLI. They do not attempt to read unrelated system files or secrets. Minor documentation inconsistency: SKILL.md in one place says IMOU_BASE_URL is required with no default, but the client code actually uses a reasonable default (https://openapi.lechange.cn) if IMOU_BASE_URL is not set.

✓ Install Mechanism

No installer that downloads arbitrary code; dependency is requests (pip), which is reasonable and declared. The skill is instruction + included Python scripts; there are no external URLs or extract/install steps that would write unknown code at runtime.

ℹ Credentials

Only IMOU_APP_ID and IMOU_APP_SECRET are required (IMOU_BASE_URL optional). These are proportional for calling Imou APIs. Important: those credentials provide administrative API access (token issuance and repository/target management) — installing the skill gives it ability to create/list/delete repositories and add/delete face/workwear targets in the Imou account, so treat them like sensitive credentials.

✓ Persistence & Privilege

always is false and the skill does not request persistent platform privileges. It does not modify other skills or system-wide configs. It runs as a user-invoked CLI and can be invoked autonomously per platform default, which is expected for skills.

版本历史

v1.0.0

imou-multimodal-analysis 1.0.0 - Initial release. - Provides AI analysis on device snapshot/image URLs: human, smoking, phone-using, workwear, absence, shelf, trash, heatmap, and face detection. - Supports manage detect repositories: create, list (paginated), delete. - Supports manage targets in repository: add (image URL or base64), list (paginated), delete. - CLI and environment variable setup instructions included. - Documentation links and API usage examples provided.

元数据

Slug imou-multimodal-analysis

版本 1.0.0

许可证 MIT-0

累计安装 0

当前安装数 0

历史版本数 1

常见问题

Imou Open Multimodal Analysis 是什么？

对指定账号下设备通道的实时抓图地址进行AI智能场景分析。AI scene analysis for device channel snapshot URLs under an Imou account. 支持：人形检测、抽烟检测、玩手机检测、工装检测与离岗检测（需预先配置工装模板）、货架检测、垃圾检测、热力图数据... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 161 次。

如何安装 Imou Open Multimodal Analysis？

在 OpenClaw 或 Claude Code 对话框中运行命令「/install imou-multimodal-analysis」即可一键安装，无需额外配置。

Imou Open Multimodal Analysis 是免费的吗？

是的，Imou Open Multimodal Analysis 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。

Imou Open Multimodal Analysis 支持哪些平台？

Imou Open Multimodal Analysis 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。

谁开发了 Imou Open Multimodal Analysis？

由 Imou-OpenPlatform（@imou-openplatform）开发并维护，当前版本 v1.0.0。

Imou Open Multimodal Analysis