Bots
Memory, skills and sub-agents
What a bot carries between conversations, the commands you can teach it, and how it splits big work.
Memory
A bot sees its notes at the start of every conversation. It writes them itself with remember (one fact per note: a decision you made, a preference, a convention it found) and you can add, edit or forget any under What it remembers on its page. Notes never hold secrets, and a bot's memory is its own: the collections bot does not read the receptionist's.
Skills
A skill is a reusable instruction with a command id, a one-line description and a prompt; $ARGUMENTS stands for what you type after it. Make them under Account → Skills (New skill, or Import), and run one in the composer with /command. Team admins add team skills that every member and every team bot can use.
Sub-agents and named agents
For work whose details would swamp one conversation (a broad search, a long investigation, a slow test suite) a bot dispatches a sub-agent with a fresh context that returns only a written summary. A project can also define named agents with their own standing brief and a subset of tools; mention one with @name, or a bot picks the one whose role fits. All of them run on the same account and project, under the same permission mode.