The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs users and agents to install the global vp CLI using curl -fsSL https://vite.plus | bash and irm https://vite.plus/ps1 | iex. This pattern executes remote scripts directly in the shell without verification or integrity checks.
[REMOTE_CODE_EXECUTION]: The toolchain includes the vp dlx and vpx commands, which download and execute packages or tools from remote registries at runtime.
[COMMAND_EXECUTION]: The vp run (or vpr) command executes arbitrary shell commands defined within the run.tasks block of vite.config.ts or package.json scripts.
[PERSISTENCE]: The vp config command installs Git hooks into the .vite-hooks/ directory. These hooks (such as pre-commit) are configured to automatically run vp staged (which executes shell commands), providing a mechanism for persistent code execution across standard developer workflows.
[PRIVILEGE_ESCALATION]: Setup instructions in setup.md mention the use of apk add libstdc++, which requires root or administrative privileges to execute on supported Linux platforms.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process configuration files that serve as an attack surface for malicious instructions when handled by an agent.
Ingestion points: Project configuration files including vite.config.ts, package.json, and .node-version (as described in monorepo-and-migration.md and setup.md).
Boundary markers: The skill does not specify the use of delimiters or 'ignore' instructions to prevent the agent from following commands embedded in these files.
Capability inventory: The skill provides multiple tools that involve subprocess execution, including vp run, vp build, vp check, vp test, and vp pack.
Sanitization: There is no mention of sanitizing or validating the shell command strings read from configuration files before execution.

vite-plus-best-practices