The Agent Skills Directory

Indirect Prompt Injection Surface: The skill retrieves external migration guides and documentation from public URLs, alongside parsing local project configuration files. If the fetched documentation or files were to contain unexpected instructions, they could potentially alter the agent's behavior.
Ingestion points: Reads the local package.json file and fetches documentation from nextjs.org URLs in SKILL.md.
Boundary markers: Explicit boundary delimiters or isolation constraints for external instructions are absent.
Capability inventory: Executes shell commands (npx, npm install) as documented in SKILL.md.
Sanitization: No formal filtering or structural validation is applied to incoming content before processing.
Command Execution: The skill involves running shell commands to perform package upgrades and execute codemods. This is a standard pattern for upgrade utilities, though it represents an operational area to monitor.
Evidence: Execution of npx @next/codemod@latest and npm install next@latest within SKILL.md.

next-upgrade