The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to read and follow project-specific files like AGENTS.md, README.md, and skin.json, explicitly stating to "trust them over this skill." This creates a vulnerability where a malicious repository could provide conflicting or harmful instructions that override the agent's safe behavior.
Ingestion points: AGENTS.md, README.md, skin.json, and external URLs accessed via the browser for design reference.
Boundary markers: The skill does not define specific delimiters or warnings to treat these external files as untrusted data rather than instructions.
Capability inventory: Shell execution (npx, npm), file system modification (design/spec.md), web browsing, and project deployment.
Sanitization: No sanitization or validation is applied to the content of the project files or external websites before the agent processes them.
[COMMAND_EXECUTION]: The skill triggers shell commands including npx @videojs/cli skin create to scaffold projects. It also executes local project scripts such as typecheck, build, and deploy (targeting Vercel, Netlify, or Cloudflare). While the deploy process involves user-interaction for sign-in, the execution of arbitrary scripts defined in a project's package.json presents a risk if the project is malicious.
[EXTERNAL_DOWNLOADS]: The skill uses npx to fetch and execute the @videojs/cli package and references documentation hosted on videojs.org. These resources are managed by the official vendor videojs and are consistent with the skill's stated purpose.

videojs-skin