The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of the openai-client and caffeineai-authorization packages from the Mops registry (mops.one). Automated scans identified a potential phishing risk associated with the mops.one/openai-client URL, which serves as the primary source for the OpenAI integration library.
[INDIRECT_PROMPT_INJECTION]: The skill implements a chat functionality that ingests untrusted user input and processes it through an LLM, creating a potential attack surface for indirect prompt injection.
Ingestion points: The prompt parameter in the chat function within src/backend/mixins/openai-chat.mo receives raw text directly from the caller.
Boundary markers: None identified. The input is passed directly to the model as a user message without delimiters or instructions to ignore embedded commands.
Capability inventory: The skill enables network communication with OpenAI's API (api.openai.com) and manages canister-side secrets for API keys, providing a path for data transit if an injection is successful.
Sanitization: No sanitization or escaping of the user-provided prompt is performed before it is interpolated into the JSON request body.

extension-openai