The Agent Skills Directory

Indirect Prompt Injection Surface: The skill analyzes the project's source tree using search tools and tails development server logs to identify prefetching insights. These ingestion points represent a surface where untrusted data (like code comments or log output) could theoretically influence agent behavior.
Ingestion points: Project source code audited via rg, next dev terminal logs, and next build route tables.
Boundary markers: No explicit delimiters or instructions are used to isolate external data from the agent's core logic.
Capability inventory: The skill executes shell commands (via npx, pnpm, and next), performs file system writes through codemods, and modifies configuration files.
Sanitization: External content from logs and source files is not sanitized before being processed by the agent.
External Resource Fetching: The skill downloads and integrates supplementary functionality from external repositories and package registries.
Evidence: Instructions include npx skills add https://github.com/vercel/next.js/tree/canary/skills/next-dev-loop to fetch additional agent capabilities.
Context: These resources originate from the framework's official maintenance channels and are intended for development loop integration.
Automated Code Modification: Utilizes first-party codemods to modify the application's source code programmatically.
Evidence: Commands such as npx @next/codemod@canary remove-partial-prefetch ./app are used to remove configuration exports across the app/ directory.
Context: While this is a standard developer workflow for framework migrations, it involves remote code execution through npx and automated bulk file edits.
Local Command Execution: The skill instructs the agent to execute various local CLI tools to audit the codebase and manage the build lifecycle.
Evidence: Uses rg (ripgrep) for auditing Link components and coordinates project-specific build (next build) and test commands via the project's package manager.

next-partial-prefetching-adoption