The Agent Skills Directory

[METADATA_POISONING]: The skill metadata identifies the author as 'Anthony Fu' and points to his GitHub repository, which contradicts the actual author context of 'liangmiqwq'. This misattribution impersonates a well-known open-source maintainer.
[INDIRECT_PROMPT_INJECTION]: The skill documents the use of environment variable loading features that increase the surface for sensitive data exposure.
Ingestion points: The loadEnv utility processes project .env files.
Boundary markers: The instructions do not specify any delimiters or safety warnings for the agent when handling the resulting environment data.
Capability inventory: The skill enables full build and server operations through Vite's JavaScript API, which runs with local environment permissions.
Sanitization: The provided examples explicitly demonstrate how to bypass the default VITE_ prefix security filter by using an empty string as a prefix, which results in the exposure of all environment variables including those containing sensitive credentials.
[COMMAND_EXECUTION]: The documentation includes instructions for running CLI commands and invoking programmatic build and server processes that execute code in the local shell.
[EXTERNAL_DOWNLOADS]: The skill references several official resources, including the Vite project, Nitro, and various meta-frameworks. These references target well-known and established technology services.

vite