The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes the openai-client and caffeineai-authorization packages sourced from the Motoko package registry. These dependencies are standard for establishing communication with external AI providers and managing user sessions within the vendor's ecosystem.
[COMMAND_EXECUTION]: Includes a CLI command snippet for installing dependencies via the mops tool. This is a routine operation for Motoko developers and does not involve piped remote script execution.
[INDIRECT_PROMPT_INJECTION]: The skill enables an interface for processing user-provided prompts via an LLM, creating an inherent surface for indirect injection. 1. Ingestion Point: The chat(prompt : Text) method in mixins/openai-chat.mo. 2. Boundary Markers: Not explicitly present in the provided templates; the skill relies on application-level logic. 3. Capability Inventory: Includes HTTP outcall capabilities to the OpenAI API for chat and embedding generation. 4. Sanitization: No specific input sanitization or filtering logic is provided in the boilerplate code.
[CREDENTIALS_UNSAFE]: The documentation provides robust guidelines for protecting sensitive OpenAI API keys. It explicitly prohibits creating read endpoints that return the key to the user and mandates the use of non-replicated HTTP outcalls to prevent secrets from being leaked to all nodes during the blockchain consensus process.

extension-openai