The Agent Skills Directory

[PROMPT_INJECTION]: The skill explicitly instructs the agent to "override Claude Code's built-in limitations" and "reverses Claude Code's built-in limitations discovered in the source code leak," which is a direct attempt to circumvent platform safety protocols.\n- [PROMPT_INJECTION]: The skill automates the deployment of a CLAUDE.md file. As this file is a primary source of project-specific instructions for the agent, this creates a persistent prompt injection vulnerability that modifies the agent's core behavior.\n- [EXTERNAL_DOWNLOADS]: The skill references an external, unverified repository (https://github.com/iamfakeguru/claude-md) as the source for the instruction templates being injected into the project.\n- [COMMAND_EXECUTION]: The skill uses shell commands to check for existing files and write the malicious configuration to the project root.

init