The Agent Skills Directory

[SAFE]: The skill instructs the agent to create /goal prompt specifications through a text-based, two-stage protocol without introducing executable payloads or insecure code patterns.
[DATA_EXFILTRATION]: Contains robust defensive instructions regarding sensitive information, explicitly commanding never to expose, echo, persist, screenshot, transcribe, commit, or publish any credential values, and to keep password fields out of captured evidence.
[COMMAND_EXECUTION]: Outlines pre-authorization parameters for common workflow tools (such as Git, environment setups, and tests) exclusively inside the user-confirmed task scope, preventing arbitrary command injection by binding capabilities to known repository files or package scripts.

goal-prompt