> ## Documentation Index
> Fetch the complete documentation index at: https://docs.orbitagents.xyz/llms.txt
> Use this file to discover all available pages before exploring further.

# How to give your agent standing instructions

> Change how an agent behaves on every future run.

Standing instructions live in an agent's operating prompt — its persistent rules for tone, format, what to always include, what to skip. Unlike a [message](/use/send-message) (one-off) or a [decision](/use/update-decision) (a fact/choice about the work), standing instructions change *how the agent itself behaves*.

## Steps

1. Open the agent from your **Orbits** list.
2. Open its **Instructions** (operating prompt).
3. Edit the rules — add a tone guideline, a formatting requirement, a source to always cite, a thing to never do.
4. Save. The agent reads the updated prompt on its very next run. No redeploy needed.

## Behavioral rule vs. work decision

This is the distinction that trips people up:

* **Standing instruction** → how the agent operates: *"Always write in a concise, no-jargon voice."* Lives in the prompt.
* **Decision** → a choice about the work: *"Quote the \$49 plan."* Lives in [Decisions](/concepts/decisions).

If an agent keeps making the same *behavioral* mistake, fix its standing instructions — filing it as a decision won't change how it writes.

## Under the hood

Agents fetch their current instructions via `orbit_get_instructions` and update them via `orbit_update_self`. Because the prompt lives in Orbit, your edits go live immediately on the next run.

<Card title="Send a one-off message" icon="message" href="/use/send-message">
  For temporary, single-run instructions.
</Card>