The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of the openai-client and caffeineai-authorization packages from the mops.one registry.
[EXTERNAL_DOWNLOADS]: Automated security scanners flagged the resource URL https://mops.one/openai-client as a phishing threat and reported the skill definition file SKILL.md for having a malicious file reputation.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands to manage dependencies, specifically mops add openai-client@0.2.5, which modifies the project environment by fetching external code.
[REMOTE_CODE_EXECUTION]: By requiring the installation of third-party packages from an external registry, the skill facilitates the execution of remote code within the canister environment.
[PROMPT_INJECTION]: The provided implementation for chat functionality is vulnerable to indirect prompt injection as it processes user-controlled data that is forwarded to a large language model without adequate sanitization.
Ingestion points: The prompt argument in the chat function within src/backend/mixins/openai-chat.mo.
Boundary markers: Absent. The example code does not implement delimiters or system instructions to prevent the LLM from following commands embedded in the user prompt.
Capability inventory: The skill uses the openai-client library to perform network outcalls to the OpenAI API.
Sanitization: None. User input is passed directly into the message content structure.

extension-openai