The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes a 'User Off-Ramp' feature that allows users to bypass domain-specific skills using phrases like 'this time without skill'. While these resemble override commands, they are implemented as legitimate functional controls for the agent's routing logic rather than attempts to bypass underlying model safety filters.
[DATA_EXFILTRATION]: There are no commands or instructions that access sensitive files, environment variables, or perform network requests to external domains.
[COMMAND_EXECUTION]: The skill does not utilize shell commands, subprocesses, or any form of local system interaction. It is a text-only orchestration prompt.
[REMOTE_CODE_EXECUTION]: The skill does not download, reference, or execute any external scripts or packages.
[OBFUSCATION]: The content is written in clear Markdown and plain text. No encoded strings (Base64), hidden characters, or homoglyphs were detected.
[DYNAMIC_EXECUTION]: The skill does not generate or compile code at runtime. It relies on standard prompt interpolation for its routing logic.

thinking-router