The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references the installation of the belt-sh/cli via npx and provides links to installation documentation on GitHub (inference-sh/skills). These are standard procedures for utilizing the inference.sh platform described in the guide.
[COMMAND_EXECUTION]: The allowed-tools configuration permits the use of the belt CLI tool. The skill body contains numerous examples of belt commands used to run inference on LLM, image, and video models. These examples serve educational purposes and align with the skill's stated goal of teaching prompt engineering.
[INDIRECT_PROMPT_INJECTION]: The skill provides templates for processing external data, such as code reviews and article summaries. While this represents a surface for indirect prompt injection (where instructions hidden in the data might influence the model), the skill correctly demonstrates the use of clear delimiters (e.g., Code:, Reviews:) and role-based constraints to mitigate these risks. As a guide, it does not programmatically handle untrusted input.
Ingestion points: User-provided content passed to model prompts via the --input flag in shell examples.
Boundary markers: Examples demonstrate using headers and brackets (e.g., [article text], Reviews:) to separate instructions from data.
Capability inventory: The skill utilizes the belt CLI to send data to remote AI inference APIs.
Sanitization: The guide teaches users how to structure prompts to avoid confusion between instructions and data, though it does not provide automated sanitization scripts.

prompt-engineering