Variables
How placeholders are substituted into instructions, and what happens to a missing key.
Variables keep the instructions stable while the specifics change. They are a plain key-value map on the routine, substituted into the instructions immediately before each run.
Syntax
Write a placeholder in the instructions:
Audit the repository {{repository}} for changes merged since last Monday.
Report in {{reportingCurrency}}.Then set the values on the routine:
{
"repository": "acme/web-app",
"reportingCurrency": "USD"
}Keys may contain letters, digits, _, ., and -. Whitespace inside the
braces is ignored, so {{ repository }} and {{repository}} are the same
placeholder.
Substitution rules
| The value is… | The placeholder becomes… |
|---|---|
| a string | the string, inserted as-is |
| a number | its string form |
| missing | an empty string — the run does not fail |
| an object, array, boolean, or null | an empty string; serialise it yourself if you need it in the prompt |
A typo fails quietly
{{ repositry }} is not an error — it becomes an empty string, and the model
receives an instruction with a hole in it. Read the rendered behaviour in the
first proof run rather than assuming the substitution worked.
Only the instructions are rendered. The routine name, description, and webhook configuration are not templated.
Naming
Use the same casing everywhere and prefer names that read as nouns in a
sentence: repository, siteUrl, reportingWindow, campaign,
reportingCurrency. Templates in the product follow this convention, and their
variables carry a label, a description, whether they are required, and an
example.
Changing a variable
Updating variables is a material configuration change: the routine returns to
draft and needs a fresh manual proof before it can be scheduled again.
This matters for routines whose variable rotates — a weekly blog topic, for example. Plan for a proof run each time, or drive the rotation from a variable that describes the rule ("this week's highest-traffic page") rather than the value itself.
What not to put in a variable
Never put credentials, tokens, or API keys in variables. They are substituted directly into the prompt, which means the model sees them and they can reach a tool call.
Credentials belong in connection headers, where they are encrypted at rest and resolved server-side at call time. See Connections.