The Agent Skills Directory

[COMMAND_EXECUTION]: The skill manages the execution of external CLI tools (codex, agy) using the shell. It implements safeguards against command injection by mandating that untrusted content (like code diffs or PR descriptions) be passed via stdin or temporary files with delimiters, rather than being interpolated into the command string.
[DATA_EXFILTRATION]: Context is shared with external AI services (OpenAI and Google). This behavior is the primary purpose of the skill and targets well-known providers. The skill instructions specify that authentication is handled by the CLIs themselves and no secrets are stored or managed by the skill.
[PROMPT_INJECTION]: The skill defines a clear trust boundary by instructing the agent to treat output from delegated models as peer opinion or data, rather than authoritative instructions. This reduces the risk of indirect prompt injection where a secondary model could influence the primary agent's behavior.

dispatch