The Agent Skills Directory

[PROMPT_INJECTION]: The skill provides code templates that ingest untrusted data from form inputs and API responses, creating a potential surface for indirect prompt injection.
Ingestion points: Templates in server-action-form.tsx and route-handler-api.ts process data from formData and request.json().
Boundary markers: Not included in the template logic.
Capability inventory: The templates demonstrate capabilities for network operations (fetch), database mutations, and local file system writes (fs.writeFile).
Sanitization: templates/app-router-async-params.tsx contains an example using dangerouslySetInnerHTML, which is a vector for injection if the source content is untrusted.
[COMMAND_EXECUTION]: The skill includes a utility script (scripts/check-versions.sh) designed to run locally to verify software versions and project structure.
Evidence: The script executes node -p, find, and grep to analyze the project environment.
[PROMPT_INJECTION]: The skill metadata contains deceptive authorship claims.
Evidence: SKILL.md identifies the author as 'Claude Skills Maintainers', which conflicts with the provided developer context of 'spiosifidis'.

nextjs