The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches deployment assets, models, and Docker images from trusted sources. It uses sparse-checkouts to retrieve code and configuration from the official open-edge-platform/scenescape GitHub repository. Additionally, it pulls official Docker images for SceneScape services and downloads pre-trained models using a pinned-version downloader image from the trusted Intel organization.
[COMMAND_EXECUTION]: The skill orchestrates deployment using system commands for infrastructure management. It executes docker compose for container orchestration and uses subprocess calls in Python scripts to manage the deployment lifecycle, including warmup, health checks, and verification. These operations are essential to the skill's primary purpose and use list-based arguments to prevent injection.
[CREDENTIALS_SAFE]: The skill implements secure runtime credential management. It generates random, high-entropy passwords and TLS certificates during the bootstrap phase. Sensitive data, such as the generated .env file and private keys, are stored with restrictive file permissions (0600), while public certificates are correctly set to 0644 to ensure service accessibility.
[DYNAMIC_EXECUTION]: The skill generates environment-specific configurations dynamically. It populates Docker Compose templates and GStreamer pipeline configurations with user-provided parameters. Dedicated validation logic ensures that camera identifiers and stream URLs conform to expected formats before they are written to disk.
[INDIRECT_PROMPT_INJECTION]: The skill manages the attack surface for untrusted data processed at runtime.
Ingestion points: User-provided camera identifiers, RTSP stream URLs, and scene names entered via prompts.
Boundary markers: The skill uses dedicated validation scripts (deploy_inputs.py) to delimit and sanitize inputs before they are interpolated into deployment configuration files.
Capability inventory: Restricts capabilities to local file writes within the deployment directory and specific network operations for health checks and service orchestration.
Sanitization: Implements regex-based validation for identifiers and robust URL parsing for streams to prevent command or path injection.

scenescape-setup