Kami Package Detection

A free skill by Kami SmartHome. Get notified the moment a package arrives at your door. Detects packages, parcels, and bags from RTSP camera streams using AI...

Security Scan

The skill's files, dependencies, and runtime instructions are consistent with a local ONNX-based RTSP package-detection tool; nothing in the package attempts unexpected network exfiltration or privilege escalation.

Purpose & Capability

Name/description (package detection from RTSP) aligns with required binaries (python3), Python dependencies (onnxruntime, opencv-python-headless, numpy), and included code (yolo_world_onnx.py). The skill legitimately needs access to an RTSP stream and a local ONNX model file for its stated purpose.

Instruction Scope

SKILL.md instructs only local setup (bash setup.sh), placing the ONNX model, and running the script against an RTSP URL. The instructions do not ask the agent to read unrelated system files or environment variables. Minor inconsistencies in documentation: SKILL.md claims the ONNX model is included in the package, while README states the model is user-provided/not included; the registry version in metadata (1.0.5) differs from the SKILL.md frontmatter (1.0.4). These are documentation mismatches, not evidence of malicious behavior.

Install Mechanism

No remote installer is invoked by the skill bundle. setup.sh creates a local .venv and installs packages from requirements.txt via pip; required packages are standard (onnxruntime, opencv-python-headless, numpy). There are no downloads from arbitrary URLs or embedded installers that extract remote archives.

Credentials

The skill requests no environment variables or external credentials. It needs network access only to whatever RTSP URL the user supplies — this is expected for a camera-based detection tool. No unrelated secrets or config paths are requested.

Persistence & Privilege

The skill does not request always:true and does not modify other skills or system-wide configurations. It logs to a local file in the skill directory (yolo_world_onnx_log.log) which is reasonable for debugging; there is no evidence of persistent backdoors or privilege escalation.

Assessment

This package appears to do exactly what it claims: run a YOLO ONNX model on an RTSP camera and print detections to stdout. Before installing, consider: 1) Supply a trusted RTSP URL — the script will connect to that camera and could process private video. 2) The model file may not actually be included despite one doc line saying it is; ensure you have a valid yolov8s-worldv2.onnx in the skill directory before running. 3) The script writes a local log file (yolo_world_onnx_log.log) in the skill directory. 4) The code uses only standard Python packages installed into a local .venv (setup.sh). If you want extra assurance, review the full yolo_world_onnx.py file (no network exfiltration or unexpected remote endpoints were found) and run the skill in a network-isolated environment first.

Get notified the moment a package arrives at your door.

Monitor your camera feed for packages, parcels, backpacks, handbags, and suitcases. When detected, returns the object class and bounding box as JSON — ready for automation.

Features

• 📦 Package & parcel detection
• 🧳 Suitcase / backpack / handbag recognition
• 🏠 Doorstep & reception monitoring
• ⏱ Configurable detection duration
• 🔔 JSON output for easy integration

Scenarios

• Doorstep delivery waiting
• Office reception package management
• Warehouse cargo monitoring
• Temporary item watch

Installation

bash setup.sh

Creates .venv/ and installs onnxruntime, opencv-python-headless, numpy. Idempotent.

Prerequisites

• python3 and python3-venv installed
• yolov8s-worldv2.onnx model file in the skill directory
• RTSP camera online and reachable

Model

The yolov8s-worldv2.onnx model file is included in the skill package. If missing, re-download this skill from ClawHub:

clawhub install kami-package-detection

Alternatively, download the YOLOv8s-World v2 .pt model from Ultralytics YOLO-World and export it to ONNX yourself. Make sure the class names used during export match the --class_names parameter:

pip install ultralytics
yolo export model=yolov8s-worldv2.pt format=onnx imgsz=320
cp yolov8s-worldv2.onnx /path/to/kami-package-detection/

Parameter Confirmation

Before running, confirm these parameters with the user:

Parameter	Default	Description
--rtsp_url	rtsp://127.0.0.1/live/TNPUSAQ-757597-DRFMY	RTSP camera URL
--conf_threshold	0.25	Confidence threshold (0.0–1.0)
--class_names	parcel package "delivery box" person "Cardboard box" "Packaging Box" backpack handbag suitcase	Classes to detect
--run_time	60	Max seconds; 0 = unlimited

Ask the user: do any parameters need to be changed?

Usage

.venv/bin/python yolo_world_onnx.py \
  --rtsp_url rtsp://your-camera-address \
  --run_time 60

Output (stdout JSON)

{
  "detections": [
    {
      "class_name": "parcel",
      "bbox": {"x1": 100, "y1": 200, "x2": 300, "y2": 400}
    }
  ]
}

Field	Type	Description
class_name	string	Detected object class
bbox.x1, y1, x2, y2	int	Bounding box coordinates

Exit Codes

Code	Meaning
0	Target detected, JSON output written
1	Error (model missing, RTSP failure, runtime exception)
2	Timeout, no target detected within --run_time

Troubleshooting

• bash: .venv/bin/python: No such file or directory → Run bash setup.sh
• Model file not found → Place yolov8s-worldv2.onnx in the skill directory
• Cannot open video → Check camera is online and --rtsp_url is correct