The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to run node scripts/design-check.mjs <artifact.html> where the artifact path is supplied by the user via $ARGUMENTS. This pattern creates a command injection surface if the file path is not strictly validated or escaped by the platform, potentially allowing for arbitrary shell command execution using metacharacters.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted HTML and CSS content during design reviews. A malicious artifact could contain hidden instructions or specific text patterns designed to influence the agent's logic when it evaluates the findings reported by the analysis script.
Ingestion points: scripts/design-check.mjs reads the file content from a path provided in the arguments.
Boundary markers: There are no delimiters or instructions to ignore embedded commands in the reporting output.
Capability inventory: The skill has the ability to read local files, execute shell commands, and interact with a browser environment via Playwright.
Sanitization: The script uses regular expressions for static analysis but does not sanitize or filter the content before the agent processes the findings.

review-design