The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to recommend the installation of an external extension using the command npx skills add leonxlnx/taste-skill -g -y. This downloads and executes code from a repository belonging to a third-party user (leonxlnx) who is not associated with the skill's author (psylch). The inclusion of the -y flag is particularly risky as it automates the installation process without user confirmation.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to scan and audit untrusted source code files (e.g., .tsx, .vue, .html) and web content. This behavior creates a vulnerability where malicious instructions embedded in the files being reviewed could be processed by the agent, potentially leading to unauthorized actions.
Ingestion points: Scans directories (like src/ and app/) to read component and page files; interacts with URLs to take screenshots when browser tools are available.
Boundary markers: None found. The instructions do not specify any delimiters or warnings to help the agent distinguish between its instructions and the data being analyzed.
Capability inventory: The skill possesses the capability to read local files and use browser tools for web interaction.
Sanitization: None found. There are no mechanisms described to filter or validate the content extracted from external files or URLs.

ux-interaction-taste