The Agent Skills Directory

[DATA_EXFILTRATION]: Information disclosure of local filesystem paths. The debug-tricks.md file describes diagnostic tools (get_project_metadata and get_logs) that return the absolute project root path and the absolute path to the development log file on the host system.
[COMMAND_EXECUTION]: Local service interaction via command-line tools. debug-tricks.md provides examples for using curl to send JSON-RPC requests to a local development server to invoke diagnostic tools.
[PROMPT_INJECTION]: Indirect prompt injection vulnerability surface. The skill describes tools for reading development server logs and runtime error stacks (debug-tricks.md). Because logs and error messages often incorporate data from external HTTP requests (e.g., headers, query parameters), they provide a path for untrusted data to enter the agent's context.
Ingestion points: debug-tricks.md (server logs via get_logs, build/runtime errors via get_errors).
Boundary markers: Absent; the instructions do not specify the use of delimiters when the agent processes these data sources.
Capability inventory: The skill references the ability to perform network requests (curl) and read file paths provided by tools.
Sanitization: Absent; no validation or filtering of log content is suggested before analysis.
[EXTERNAL_DOWNLOADS]: Use of standard ecosystem utilities and well-known services. The skill references npx commands for running Next.js codemods (@next/codemod) and deployment tools (create-sst). It also includes instructions for loading third-party scripts from established services such as Google Tag Manager, Google Analytics, and YouTube, which are treated as safe sources.

next-best-practices