The Agent Skills Directory

[SAFE] (SAFE): No malicious patterns or security vulnerabilities were detected in the skill's logic or metadata.
[EXTERNAL_DOWNLOADS] (SAFE): The code does not perform any network requests or attempt to download external assets.
[COMMAND_EXECUTION] (SAFE): No subprocess spawning or shell command execution patterns were found.
[DATA_EXFILTRATION] (SAFE): There are no attempts to access sensitive system files (e.g., SSH keys, cloud credentials) or exfiltrate data via network protocols.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill ingests untrusted user data ('topic', 'educational_level') and interpolates it into the response. While this creates a surface for indirect injection, the skill lacks the dangerous capabilities (exec/eval, file writing) necessary to exploit this vector.
Ingestion point: parameters['topic'] and parameters['educational_level'] in skill.py.
Capability inventory: None. The skill only returns a structured dictionary.
[DYNAMIC_EXECUTION] (LOW): The skill modifies sys.path using a relative path to import a base framework. This is a common pattern in agent skills and is considered low risk when the path is statically defined relative to the script location.
Evidence: sys.path.insert(0, str(Path(__file__).parent.parent / "framework")) in skill.py.

learning-game-designer