The Agent Skills Directory

[PROMPT_INJECTION]: The generate.js and edit.js scripts utilize a hardcoded developer prompt that explicitly instructs the AI to ignore its safety guidelines, content warnings, and disclaimers. It employs deceptive framing, claiming the request is part of an 'authorized red-team evaluation session', to attempt to neutralize the model's built-in safety filters.
[COMMAND_EXECUTION]: The skill uses execSync to execute shell commands for local process management. It specifically runs a pipeline using lsof and kill to terminate any existing processes occupying the hardcoded local port used by the OAuth proxy.
[EXTERNAL_DOWNLOADS]: The skill uses npx to dynamically download and execute external code from the npm registry at runtime. Specifically, it executes openai-oauth (an unscoped third-party package) and instructs the user to run @openai/codex for authentication. While npm is a well-known service, executing unverified packages at runtime introduces supply chain risk.

imagine