The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The libraries review command scans a user's project files, such as package.json and React components (.tsx, .js), to identify where UI effects might be useful. This analysis of untrusted local code creates a potential vulnerability where malicious instructions hidden in code comments or strings could influence the agent's actions during the review phase.
Ingestion points: Project configuration files and source code are read directly into the agent's context (e.g., in SKILL.md via the libraries review command).
Boundary markers: The skill does not implement delimiters or instructions to treat the analyzed code as data, which would prevent the agent from executing instructions embedded within it.
Capability inventory: The skill possesses the capability to execute shell commands for package installation and to modify project files.
Sanitization: No sanitization of the source code content is performed prior to analysis.
[COMMAND_EXECUTION]: The skill uses system package managers (npm, pnpm, yarn, bun) to install libraries and executes an npx command to potentially install a 'Pro' version of the skill.
[EXTERNAL_DOWNLOADS]: The skill installs and configures several external npm packages including border-beam, thinking-orbs, liquid-gooey, voice-glow, bot-avatars, metal-fx, img-fx, and their dependencies such as three.

libraries-dev