The Agent Skills Directory

Remote Code Execution & External Downloads (MEDIUM): The skill utilizes npx @next/codemod@latest to fetch and execute transformation scripts directly from the npm registry. While this is the official method for Next.js migrations, executing code from a public registry at runtime carries inherent risk of supply chain attacks.
Command Execution (LOW): The skill executes multiple shell commands including npm install, npm run build, and npm run dev. These operations modify the project environment and execute local scripts based on the state of the codebase.
Indirect Prompt Injection (LOW): The skill exhibits an indirect prompt injection surface by fetching external documentation from nextjs.org to determine migration steps.
Ingestion points: Instructions 2 and 3 involve fetching data from external URLs via WebFetch.
Boundary markers: Absent. There are no instructions to ignore embedded commands or malicious patterns in the fetched text.
Capability inventory: The skill has the ability to run shell commands (npx, npm), install packages, and execute build/dev scripts.
Sanitization: Absent. The agent is expected to interpret the documentation and apply it directly to the command line without validation filters.
Unverifiable Dependencies (MEDIUM): The skill explicitly instructs to install latest versions of several packages (next, react, react-dom). Using unpinned versions can lead to non-deterministic environments and potential exposure to malicious package releases.

next-upgrade