The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides numerous examples for using the belt CLI to interact with AI models. The environment is secured by a restriction in the metadata that limits the Bash tool to only commands prefixed with belt.
[EXTERNAL_DOWNLOADS]: The skill references an installation script and additional skills hosted on the vendor's official GitHub repository (inference-sh/skills). These are legitimate resources for the skill's intended purpose.
[PROMPT_INJECTION]: The provided templates for tasks like code review or content writing use placeholders (e.g., [code], [article text]) to incorporate user data. This introduces a surface for indirect prompt injection where untrusted input could attempt to influence the agent's behavior.
Ingestion points: User-provided content interpolated into placeholders within the Code Review, Content Writing, and Structured Output templates.
Boundary markers: None present in the provided templates to delimit untrusted input from instructions.
Capability inventory: The skill allows execution of belt CLI commands through the Bash tool.
Sanitization: The instructions do not define sanitization or validation logic for the input processed by the templates.

prompt-engineering