The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references the installation of the ag2 library and its provider-specific extras (e.g., ag2[openai], ag2[anthropic]) via standard package managers. These are official vendor resources required for the framework's operation.
[COMMAND_EXECUTION]: Includes a Python one-liner intended for environment sanity checks. This command is used to confirm the active Python executable and the installed version of the framework.
[INDIRECT_PROMPT_INJECTION]: The skill evaluates user mentions of AG2 features to recommend specific sub-skills. This ingestion of user intent is a standard part of its routing functionality and does not involve unsafe data interpolation into sensitive operations.
[CREDENTIALS_UNSAFE]: Mentions the requirement for various API keys (e.g., OPENAI_API_KEY) but follows security best practices by recommending that they be loaded from a .env file rather than being hardcoded.

ag2-overview