README does not say a subagent definition can trigger advise without the user asking #31

Raised by @GauranshMathur in a Claude Code session, after #26.

What is wrong

Since #26, advise treats a subagent definition that tells it to run as the user asking. The skill cannot tell who wrote that file, and inside a subagent it cannot ask the user. A definition from a cloned repository or another plugin can therefore start a Codex run, which sends the brief and the files Codex reads to OpenAI, when the person at the keyboard did not ask for one.

The README's Data handling section still says advise sends data "only when you ask for a second opinion". That is the same concern that put the plugin's review on hold, so it has to be disclosed plainly.

Decision made

Say it plainly in the README's Data handling section. The skill's behaviour does not change.