The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of shell utilities such as grep, wc, sort, and uniq to perform deterministic audits on HTML and CSS files. It interpolates file paths (e.g., "$FILE") into these commands, which is a powerful capability intended for local analysis.
[REMOTE_CODE_EXECUTION]: The skill executes a local TypeScript file (scripts/layout-check/layout-check.ts) using the bun runtime and arbitrary JavaScript snippets using node -e. As the source for the layout-check script is not provided in the skill bundle, its integrity and safety cannot be independently verified.
[DATA_EXFILTRATION]: The skill initiates a local web server using python3 -m http.server on ports 8899-8905 to serve designs for visual review. This action exposes the contents of the served directory to the local network.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted design artifacts and processes them through shell commands. While variable quoting is applied, malicious filenames or content could potentially be crafted to influence shell execution behavior.
Ingestion points: HTML/CSS design artifacts and file paths processed by the auditing scripts.
Boundary markers: None identified for shell command interpolation.
Capability inventory: Shell command execution (grep), dynamic code execution (node, bun), and local network serving (python3).
Sanitization: Partial mitigation is provided through the use of variable quoting in shell snippets.

design-review