Docs › Loops
Instructions
Your own words for every loop's agent, and a look at everything a loop loads.
Where Settings › Loops › Instructions
1 min read
How it works
- Open Settings, then Loops, and write in Custom instructions. They are appended to the system prompt of every loop, whichever agent runs it.
- They always come after the agent's own prompt and never replace it, so the agent keeps what it runs on.
- In a loop, Loop context, at the foot of the message box after the mode, opens a sheet of what the loop loads as it starts.
- Its rows include Origami's prompt, your Custom instructions, the agent's own prompt, instruction files, memory, skills, subagents, MCP servers and hooks. Each says whose it is, Built-in, User or Project, and how many tokens it takes.
- Press a row that has a file to open it in Files.
Good to know
- A loop loads what its agent would load in a terminal: its own settings, skills and servers.
- Tasks a loop hands to another agent are given your instructions too.
- GitHub Copilot does not say what else it loads, and the sheet says so.