You're running AI agents that send email, move money, touch files. The moment an agent can act, a bad instruction becomes a bad action — with no undo.
Most tools log what happened after the fact. Chloros is a gate outside your agent that checks each action before it fires: scan skills before you install, hold irreversible actions (money, sends, deletes) for your approval, rewind or kill when needed.
Built for people running their own agents — not the enterprise. Free to try.
