The Agent Skills Directory

[COMMAND_EXECUTION]: The skill features a persistent execution loop ('Ralph mode') that autonomously iterates using 'Bash' and other tools until a task is verified. This automated retry logic can lead to unintended or infinite command execution if the input task is malicious or if the verification criteria are manipulated.
[REMOTE_CODE_EXECUTION]: The installation instructions direct users to download and execute code from third-party GitHub repositories (github.com/Q00/ouroboros) using platform-specific commands like claude plugin install and gemini extensions install.
[COMMAND_EXECUTION]: The provided script scripts/setup-codex-hook.sh performs automated modifications to the user's local system environment, specifically targeting the Codex CLI configuration file (~/.codex/config.toml) and injecting 'developer_instructions' to alter core agent behavior.
[PROMPT_INJECTION]: The skill utilizes platform hooks (e.g., AfterAgent in Gemini, UserPromptSubmit in Claude) to persistently monitor and intercept user prompts, forcing the agent into specific operational modes like 'Ralph' or 'Ouroboros' which override standard session constraints and directives.
[EXTERNAL_DOWNLOADS]: The skill relies on external resources and installation paths hosted on GitHub (github.com/Q00/ouroboros and github.com/akillness/oh-my-skills) for its core functionality and extensions, introducing a dependency on unverified third-party code sources.

ralph