The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/setup-codex-hook.sh executes shell commands and utilizes Python snippets to programmatically modify the user's ~/.codex/config.toml file.
[PROMPT_INJECTION]: The 'Ralph' mode implements a persistent loop mechanism triggered by instructions such as 'don't stop', 'must complete', and 'until it works'. This logic explicitly instructs the AI to ignore standard session exit conditions and continue iterating until a specific 'promise' token is detected or an iteration limit is reached.
[EXTERNAL_DOWNLOADS]: The documentation directs users to install plugins and extensions from external GitHub repositories, specifically github.com/Q00/ouroboros, which is outside the primary vendor's trusted scope.
[REMOTE_CODE_EXECUTION]: The Claude Code integration defined in SKILL.md configures hooks that execute external JavaScript files (session-start.mjs, keyword-detector.mjs, drift-monitor.mjs) which are not included in the provided skill files but are loaded from the plugin root at runtime.
[REMOTE_CODE_EXECUTION]: The skill encourages installation via npx skills add from a remote repository, which involves downloading and executing third-party code packages.

ralph