The Agent Skills Directory

[SAFE]: No executable scripts (Python, JavaScript, Shell, etc.) are included in the skill. The files consist entirely of Markdown documentation and TOML/Markdown configuration templates.
[SAFE]: The skill provides detailed routing policies and instructions for agents to delegate tasks to subagents. The instructions include security-conscious practices, such as treating subagent input as external and requiring authorization for implementation changes.
[SAFE]: Numerous external links are provided for reference; all verified links point to official documentation (Anthropic, OpenAI, Claude, Cursor, OpenCode) or reputable academic repositories (arXiv).
[SAFE]: The installation procedures described are transparent, targeting standard configuration directories for the respective AI tools, and explicitly state they should only be performed under a user configuration grant.
[SAFE]: Prompt instructions for subagents use standard task-framing language and do not contain safety bypasses, behavior overrides, or extraction attempts.

model-routing