The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The master installation script (scripts/install.sh) downloads a shell script from an untrusted external domain and executes it via a pipe to bash.
Evidence: curl -fsSL https://plannotator.ai/install.sh | bash in scripts/install.sh.
[EXTERNAL_DOWNLOADS]: The skill fetches code and components from various non-trusted external sources during installation and setup.
Evidence: Downloads from https://plannotator.ai/install.sh and refers to plugins at https://github.com/Yeachan-Heo/oh-my-claudecode.
Installation of global NPM packages: agent-browser, playwriter, and opencode-ai.
[PRIVILEGE_ESCALATION]: Setup scripts modify global user configuration files for multiple AI agent platforms to inject executable hooks and change environment settings.
Evidence: scripts/setup-claude.sh modifies ~/.claude/settings.json to inject a command-type hook for plannotator.
Evidence: scripts/setup-codex.sh modifies ~/.codex/config.toml to inject a notify hook executing a Python script.
Evidence: scripts/setup-gemini.sh modifies ~/.gemini/settings.json to add an AfterAgent command hook.
Evidence: scripts/setup-opencode.sh modifies ~/.config/opencode/opencode.json to register third-party plugins.
[PERSISTENCE]: The injection of hooks into agent configuration files creates a persistent execution mechanism where the skill's logic (or the downloaded plannotator binary) is triggered automatically during normal agent operations (e.g., turn completion or plan exiting).
Evidence: The ExitPlanMode hook in Claude and AfterAgent hook in Gemini ensure the logic runs repeatedly.
[COMMAND_EXECUTION]: Extensive use of shell scripts to manage local processes, background tasks, and network port monitoring.
Evidence: scripts/plannotator-plan-loop.sh uses background processes and probes local ports using /dev/tcp/127.0.0.1/ to manage the planning lifecycle.

omu