The Agent Skills Directory

[PROMPT_INJECTION]: The instructions command the agent to be "FULLY AUTONOMOUS" and explicitly forbid asking the user questions. This pattern is designed to bypass human-in-the-loop safety checks and confirmation prompts, potentially leading to unauthorized or destructive file modifications.\n- [COMMAND_EXECUTION]: The skill triggers the execution of project-specific build commands, test suites, and Git push operations. These actions interact directly with the local environment and remote repositories, carrying risks if the project configuration is compromised.\n- [INDIRECT_PROMPT_INJECTION]: The skill performs a "Full UI Codebase Scan" to map frameworks and components. Ingesting this untrusted code as input for the design process without explicit sanitization creates an attack surface where malicious comments or data in the project files could influence the agent's behavior.\n- [DYNAMIC_EXECUTION]: The skill identifies and executes build and test scripts at runtime based on the project's configuration (e.g., package.json or Makefiles), which involves running dynamically determined commands derived from the local file system.

design-overhaul