The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill's documentation provides explicit instructions for the user to execute commands with root privileges using sudo to install system-level build dependencies (cmake, g++) on the host environment.- [EXTERNAL_DOWNLOADS]: Fetches the uv tool installer from astral.sh and downloads media assets from external providers like pexels.com during the application preparation phase.- [REMOTE_CODE_EXECUTION]: Implements a remote code execution pattern by piping a downloaded shell script directly to sh (curl | sh) for the installation of the uv package manager.- [COMMAND_EXECUTION]: Orchestrates a variety of system-level commands to manage Docker images (pull and run), configure Python virtual environments, and install third-party packages.- [DYNAMIC_EXECUTION]: Generates application logic, configuration files, and model export scripts at runtime based on user requirements, which are then executed using the command tool.- [INDIRECT_PROMPT_INJECTION]: The skill processes external model identifiers and video URLs from untrusted sources, creating an attack surface for indirect prompt injection.
Ingestion points: Ingests external video URLs and Hugging Face model identifiers in SKILL.md and references/model-preparation.md.
Boundary markers: Does not utilize specific delimiters or isolation techniques for processed external metadata or configuration strings.
Capability inventory: Accesses both write and command tools, allowing for file creation and arbitrary shell execution.
Sanitization: Includes high-level instructional rules to avoid raw user input interpolation into shell commands and requires user confirmation for impactful operations.

dlstreamer-coding-agent