The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's documentation directs users to install necessary platform tools and prerequisites using piped shell commands from external sources.
The platform CLI is installed using curl -fsSL https://cli.inference.sh | sh as noted in the main SKILL.md and references/cli.md.
Prerequisites for Python and Node.js development, including uv, fnm, and nvm, are installed via similar piped execution methods from their respective official domains (astral.sh, vercel.app, and raw.githubusercontent.com).
[INDIRECT_PROMPT_INJECTION]: The application boilerplates provided in the skill ingest untrusted user input, establishing an attack surface for indirect prompt injection.
Ingestion points: The inference.py and inference.js templates define a prompt field in the input schemas (AppInput and RunInput) to receive user text.
Boundary markers: The provided examples do not use delimiters or boundary markers to isolate the user-provided prompt from the application instructions.
Capability inventory: The templates include capabilities for network communication (using httpx or OpenAI clients) and file system operations (using the File class), which could be leveraged if an injection is successful.
Sanitization: No sanitization or validation logic for the input prompt is included in the reference implementations.

building-inferencesh-apps