The Agent Skills Directory

[SAFE]: No malicious patterns or security vulnerabilities were detected in the instructions or example scripts.
[CREDENTIALS_UNSAFE]: The skill correctly demonstrates using python-dotenv to load secrets from environment variables, avoiding hardcoded credentials. Example scripts use placeholders like sk-... or NotRequired for demonstration only.
[EXTERNAL_DOWNLOADS]: The documentation references standard installation commands for official library extras (e.g., ag2[openai], ag2[gemini]) and well-known utilities like python-dotenv from official registries.
[INDIRECT_PROMPT_INJECTION]: The skill documents an attack surface where user input is passed to an LLM via agent.ask() or reply.ask().
Ingestion points: agent.ask() and reply.ask() in SKILL.md and assets/*.py files.
Boundary markers: Absent in basic examples.
Capability inventory: No dangerous tools (e.g., shell execution, file system modification) are defined in the provided code.
Sanitization: Absent. Given the lack of exploitable capabilities, this surface is considered safe.

ag2-quickstart