The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs developers to install the CLI and other requirements by piping remote scripts from cli.inference.sh, astral.sh, fnm.vercel.app, and GitHub directly into system shells (sh, bash, iex). This occurs in SKILL.md and references/cli.md. While these are standard installation methods for the referenced technology stacks (Astral/UV, Node.js version managers), executing unverified remote code is a significant security risk.
[COMMAND_EXECUTION]: In references/cli.md, the installation instructions for Windows users include a command that explicitly bypasses the PowerShell execution policy (-ExecutionPolicy ByPass). This configuration allows the execution of scripts that would normally be restricted by the operating system's security settings.
[PROMPT_INJECTION]: The skill documents an architecture for building AI applications that process untrusted data (user prompts), creating an attack surface for indirect prompt injection.
Ingestion points: Data enters the agent context through the AppInput and RunInput schemas defined in references/python-app-logic.md and references/node-app-logic.md.
Boundary markers: The templates do not implement delimiters or explicit instructions to ignore embedded commands in the processed input.
Capability inventory: Deployed apps have access to the file system (node:fs, PIL), network operations (httpx), and sensitive credentials (environment secrets).
Sanitization: The provided logic templates do not include validation or sanitization of external content, leaving security entirely to the developer's implementation.

building-inferencesh-apps