The Agent Skills Directory

[METADATA_POISONING]: The skill's metadata identifies the author as "Anthony Fu" and references his GitHub repositories, whereas the actual skill author is identified as "onmax". This discrepancy is misleading regarding the source and trustworthiness of the instructions.
[INDIRECT_PROMPT_INJECTION]: The skill provides patterns for the agent to interact with and process local project files, environment variables, and build configurations, creating an attack surface where untrusted project data could influence agent behavior.
Ingestion points: Project configuration files (vite.config.ts), environment files (.env), and application source code accessed via import.meta.glob hooks.
Boundary markers: Absent; the instructions do not implement or recommend delimiters to separate instructions from the data being processed.
Capability inventory: The skill facilitates the use of Vite CLI tools for file system writes (vite build), network proxying (server.proxy), and the execution of dynamic JavaScript through the evaluation of configuration files.
Sanitization: Absent; the skill relies on standard tool behavior and provides no additional validation or escaping for ingested data.
[DATA_EXFILTRATION]: The skill provides guidance on using loadEnv with an empty prefix ('') in references/core-config.md and references/build-and-ssr.md. This pattern instructs the agent to load all environment variables from the system without filtering, which can lead to the accidental exposure of sensitive secrets (such as database credentials or private keys) if they are subsequently processed, logged, or bundled into application outputs.

vite