The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's documentation and UI templates (found in SKILL.md, references/components.md, and assets/templates/landing.html) contain examples of a curl | sh command targeting https://setup.steel.dev. While piped remote execution is a high-risk pattern, in this context it is used exclusively as a static, illustrative example for designing a 'CLI install field' UI component and is not intended for execution by the agent or the included scripts.
[COMMAND_EXECUTION]: The skill provides a Python linter script (scripts/check_brand.py) intended to be run by the agent to verify brand compliance in a project. The script is implemented using only the Python standard library and performs read-only analysis of local files to detect font, color, and layout violations. It does not perform network operations or persistent changes.
[INDIRECT_PROMPT_INJECTION]: The brand linter script processes content from files within the project directory it is directed to scan. Because the script's output (warnings and errors) can include strings extracted from these files, such as CSS class names or hex codes, it creates an attack surface for indirect prompt injection. However, the script does not execute the contents of the files it reads, and its capabilities are limited to providing textual feedback to the agent.

steel-design