The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs users to install the infsh CLI by piping a script downloaded from https://cli.inference.sh directly into the shell (sh). This method executes a remote script on the host without intermediate verification.
[EXTERNAL_DOWNLOADS]: Fetches multiple tools and dependencies from external providers:
Downloads the uv Python tool from Astral's official installation endpoints (astral.sh).
Retrieves the fnm Node.js version manager from Vercel's official installation script.
Fetches the nvm manager from its official GitHub repository.
[COMMAND_EXECUTION]: Employs shell commands for app lifecycle management (infsh app init, infsh app test, infsh app deploy). It also provides a PowerShell command with ExecutionPolicy ByPass for dependency installation on Windows, which bypasses local script execution restrictions.
[PROMPT_INJECTION]: The skill architecture is vulnerable to indirect prompt injection.
Ingestion points: Untrusted external data enters the app context via AppInput (Python) and RunInput (Node.js) schemas, particularly through prompt and imageUrl fields in SKILL.md and related logic files.
Boundary markers: The provided templates and implementation guides lack explicit delimiters or instructions to ignore embedded commands in the processed data.
Capability inventory: Apps following these templates possess broad system capabilities, including file system access (open, writeFileSync), network requests (httpx, OpenAI), and project management via the infsh CLI.
Sanitization: No evidence of input validation, escaping, or sanitization logic is provided in the recommended development patterns.

building-inferencesh-apps