The Agent Skills Directory

[PROMPT_INJECTION]: The skill employs high-pressure directive language and custom tag blocks such as <EXTREMELY-IMPORTANT> and <SUBAGENT-STOP> designed to override the agent's current task brief and safety protocols. The 'Prime Directive' explicitly commands the agent to prioritize this skill's logic before any other response or action, effectively hijacking the standard instruction flow.
[METADATA_POISONING]: The skill is titled 'GodMode', a term frequently associated with jailbreaking attempts and removing AI safety constraints. Furthermore, the reference files contain future-dated timestamps (October 2026) and citations of hypothetical software versions (e.g., agy 1.4.7, Claude Code 2.1.310), creating a false or deceptive context for the agent and user.
[COMMAND_EXECUTION]: The harness mapping documentation (references/*.md) provides templates for executing complex shell commands, including piping data between different AI CLI tools like Codex, Claude, and Antigravity. This encourages a pattern of external process execution that increases the attack surface for command injection if input parameters are not properly sanitized.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for downloading and installing additional plugin components from external GitHub repositories, specifically github.com/NoobyGains/godmode. While identified as vendor resources, these installation patterns represent the introduction of remote code into the agent's environment.

using-godmode