The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's core functionality involves analyzing external URLs and local codebases, which are untrusted data sources. Maliciously crafted CSS comments or web content could contain instructions aimed at subverting the agent's audit logic.
Ingestion points: Processes local styling files (CSS, Tailwind) and external competitor websites via browser tools.
Boundary markers: The skill does not define delimiters or specific instructions to ignore embedded commands in the data it audits.
Capability inventory: Reads local files, accesses the network via browser MCP, and writes generated reports and HTML previews.
Sanitization: There is no evidence of sanitization for the styling patterns or web content extracted during the audit process.
[DATA_EXFILTRATION]: The skill reads local codebase patterns and performs network operations to research competitor sites. This combination of read access to local source code and external network access represents a data exposure risk if the agent is directed to malicious or unintended domains.
[COMMAND_EXECUTION]: The skill instructs the agent to generate and write multiple files to the filesystem, including a self-contained interactive HTML preview page and JSON token sets based on the analysis.
[NO_CODE]: The skill consists entirely of markdown instructions and does not include any accompanying executable scripts or binary files.

design-system