The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions to download and execute installation scripts directly from the web using shell pipes. This is used to install the belt CLI from https://cli.inference.sh, the uv package manager from https://astral.sh, the fnm Node.js manager from https://fnm.vercel.app, and the nvm manager from https://raw.githubusercontent.com.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to provide templates for AI applications that ingest and process external data (e.g., AppInput.prompt in inference.py and RunInput.prompt in inference.js).
Ingestion points: External data enters through the run method in SKILL.md, references/python-app-logic.md, and references/node-app-logic.md.
Boundary markers: The provided templates do not include explicit boundary markers or instructions for the agent to ignore embedded commands within the input data.
Capability inventory: The templates include capabilities for file system writing (/tmp/output.png), network requests via httpx, and model execution.
Sanitization: The instructional templates do not demonstrate sanitization or validation of the input content before processing.
[COMMAND_EXECUTION]: The development workflow relies on the execution of the belt CLI tool for scaffolding projects (belt app init), testing (belt app test), and deploying to the cloud (belt app deploy).
[DYNAMIC_EXECUTION]: The Python debugging and optimization guidelines in references/python-patterns.md recommend programmatically modifying the Python module search path using sys.path.append to resolve local dependencies at runtime.

building-inferencesh-apps