The Agent Skills Directory

[SAFE]: The skill implements a delegation pattern using the Task tool to forward user queries to specialized sub-agents. It includes explicit mapping logic to validate model names (e.g., codex, gemini) before assigning a sub-agent type, which prevents the execution of unauthorized or arbitrary agents.
[SAFE]: No instances of prompt injection, hardcoded credentials, or data exfiltration attempts were found. The skill does not request network access or interact with the local file system.
[SAFE]: The skill processes user input from $ARGUMENTS, creating a standard surface for indirect prompt injection common to agent-to-agent communication. However, the scope of action is limited to delegation, and the skill instructions explicitly require a recognized model name to proceed.

consult