The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it analyzes untrusted source code and configuration files from a Next.js repository while maintaining capabilities to write files and execute build tools.
Ingestion points: The reconnaissance scripts (scripts/recon.py, scripts/probe-capabilities.py) and various detection commands (references/detect/*.md) read source code, package manifests, and configurations into the agent context.
Capability inventory: The skill can create directory structures (scripts/scaffold-plan.py), modify code files through fix recipes, install npm/pnpm packages, and execute build/deployment commands (next build, vercel build).
Boundary markers and Sanitization: The skill implements robust safety protocols (references/workflow/safety-rails.md) and sanitization logic (e.g., Unicode escaping in JSON-LD scripts) to mitigate the risks associated with processing external content. It explicitly requires human review for 'one-way door' migrations.
[COMMAND_EXECUTION]: The skill relies extensively on shell commands for reconnaissance and implementation of fixes.
Auditor Commands: Uses ripgrep (rg) and find to scan the filesystem for specific patterns, API calls, and configurations. These are strictly read-only and targeted at optimization discovery.
Fixer Commands: Uses npm install, npx (for codemods and analyzers), and the vercel CLI for deployment diagnostics. These are standard development tools used for their intended purpose in a controlled optimization workflow.

optimize-nextjs-fluidity