The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches AI models and sample video content from trusted official repositories belonging to Intel and the Open Edge Platform on GitHub (raw.githubusercontent.com).
[COMMAND_EXECUTION]: Utilizes shell scripts (install.sh, validate_env.sh, sample_start.sh) to manage the containerized stack, configure hardware access for GPU/NPU, and orchestrate the analytics pipeline.
[INDIRECT_PROMPT_INJECTION]: The skill architecture defines an attack surface for indirect prompt injection by interpolating user-provided inputs into configuration files and processing live data streams.
Ingestion points: User parameters provided in the prompt (e.g., object labels, RTSP URLs, and alerting rules) and real-time MQTT detection metadata ingested by Node-RED.
Boundary markers: The validate_env.sh script serves as a validation boundary for environment variables before the stack is deployed.
Capability inventory: Capabilities include Docker container management, shell command execution for pipeline control, and JavaScript-based logic execution within Node-RED.
Sanitization: Input sanitization is performed via regex validation for pipeline names, MQTT topics, and IP addresses, alongside range checks for numeric parameters like source counts and group IDs.
[CREDENTIALS_SAFE]: Implements secure secret management by generating a unique random 16-character hex password for the Coturn/MediaMTX TURN server during the installation phase and explicitly guiding the user to maintain sensitive configurations within a gitignored .env file.

metro-ai-app-recipe