The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs users to install the infsh CLI using curl -fsSL https://cli.inference.sh | sh. This is a dangerous pattern that executes a remote script directly in the shell without providing a mechanism for the user to verify the script's contents or integrity before execution.\n- [EXTERNAL_DOWNLOADS]: The skill depends on several external AI models and search services hosted by providers such as fal.ai, Tavily, Exa, and ByteDance. These are accessed via the infsh command-line tool.\n- [COMMAND_EXECUTION]: The skill utilizes npx to install additional skills from inference-sh/skills, which involves downloading and executing Node.js packages from a remote registry.\n- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it accepts untrusted user input for search queries and image generation prompts without sufficient sanitization or boundary markers.\n
Ingestion points: User-provided text in infsh app run commands within SKILL.md.\n
Boundary markers: Absent. The skill does not use delimiters to isolate user-provided data.\n
Capability inventory: Access to the Bash tool to run the infsh CLI, enabling network requests and external service interaction.\n
Sanitization: Absent. No input filtering or escaping is implemented before sending data to remote AI models.

landing-page-design