The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides templates for building assistants that use the file_search and code_interpreter tools to process external data, creating a potential surface for indirect prompt injection.
Ingestion points: External data enters the agent context through file uploads (openai.files.create) associated with threads or vector stores in templates/file-search-assistant.ts and templates/code-interpreter-assistant.ts.
Boundary markers: The skill relies on natural language instructions to guide behavior but does not implement strict technical delimiters (e.g., dedicated XML tags or boundary strings) to separate untrusted data from instructions.
Capability inventory: Assistants created with these templates possess capabilities for Python code execution (code_interpreter) and semantic search across large document sets (file_search).
Sanitization: The provided examples do not demonstrate explicit sanitization, filtering, or validation of the content within uploaded files before they are processed by the assistant tools.
[EXTERNAL_DOWNLOADS]: The skill references the openai Node.js package as a primary dependency, which is managed by a trusted organization.
Version inconsistency: The documentation and templates specify version 6.7.0 of the openai package; however, as of the current date, this does not correspond to an official stable release of the OpenAI Node.js SDK (typically in the 4.x range), which may lead to installation issues or dependency confusion.
[COMMAND_EXECUTION]: The skill includes a local helper script (scripts/check-versions.sh) designed to verify the development environment and package versions using standard commands such as npm list, grep, and sed.

openai-assistants