The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill recommends installing the CLI using curl -fsSL https://cli.inference.sh | sh. This executes a shell script hosted on the vendor's domain.
[COMMAND_EXECUTION]: The instructions frequently execute the infsh command-line tool to interact with AI models. This tool is explicitly listed in the skill's allowed-tools configuration.
[EXTERNAL_DOWNLOADS]: The skill fetches assets and installers from vendor-controlled subdomains, including cli.inference.sh, dist.inference.sh, and cloud.inference.sh. It also references external skill installation via npx from the vendor's repository.
[PROMPT_INJECTION]: The skill demonstrates techniques that create an indirect prompt injection surface.
Ingestion points: User-provided content (e.g., [code], [article text]) is interpolated into templates for tasks like code review or summarization.
Boundary markers: The templates use text labels such as Code:\n[code] to separate instructions from data, which provides limited protection against adversarial input.
Capability inventory: The skill uses the infsh tool to send these constructed prompts to external LLMs.
Sanitization: There are no explicit sanitization or escaping steps provided for the data being inserted into the prompts.

prompt-engineering