The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The setup_local_ai.py script fetches installers and metadata from official vendor sources on GitHub.
Fetches the lemonade.msi installer from the lemonade-sdk/lemonade repository.
Retrieves release metadata via the GitHub API for the same repository to identify versioned assets.
Downloads macOS package installers directly from the project's official release page.
[REMOTE_CODE_EXECUTION]: The skill automates the installation of the Lemonade Server by executing the downloaded binaries.
Windows: Runs msiexec to perform a silent installation of the MSI package.
macOS: Executes the installer utility to process downloaded package files.
[COMMAND_EXECUTION]: The setup script utilizes standard system tools to manage the installation environment across different platforms.
Invokes winget on Windows and brew on macOS for package management.
Executes apt-get via a shell pipeline on Linux to add stable repositories and install server components.
[PRIVILEGE_ESCALATION]: Administrative access is requested during the installation process on Linux and macOS to manage system-level services.
Uses sudo for repository configuration and package installation on Linux distros.
Uses sudo to run the system installer on macOS.
[PROMPT_INJECTION]: The skill instructions contain directives intended to persistently steer the agent's behavior regarding multimodal routing.
Directs the agent to prioritize local inference endpoints over cloud providers, even if the user does not explicitly request local processing in a specific turn.
[INDIRECT_PROMPT_INJECTION]: The skill establishes a persistent instruction set within the workspace to govern future agent interactions.
Ingestion points: The script scripts/setup_local_ai.py writes a routing rule block to the workspace AGENTS.md file.
Boundary markers: The injected instructions are enclosed in stable comment markers (amd-skills:local-ai-use) to allow for idempotent updates.
Capability inventory: The skill has capabilities for shell command execution, network communication, and local file modification.
Sanitization: Instruction content is derived from a local template (templates/local-ai-rule.md) and populated with environment-specific parameters like local host and port.

local-ai-use