Skip to main content

Agent: System Context

Alongside the assigned prompt and your input, the Agent node can weave in the operator's system context — the global environment (brand voice, audience, tone, compliance) covered in the operator docs. It's optional, and you control whether and how much of it comes along.

Turning it on​

On the Agent node, enable include system context. When it's on, the node reads the system_context from the system_snapshot and folds those entries into the request as a system message — so the model sees the brand and tone guidance before the instruction and the work item. If system context is empty, that message is simply omitted.

All of it, or a selection​

By default, enabling system context includes all enabled entries. If you'd rather include only some, list the entry keys you want (comma-separated); leave it empty to include everything. That lets a step pull in just the Visual Style Guide for an illustrator, say, without the unrelated editorial entries — though including everything is perfectly fine, and is the common choice.

When a named entry is missing​

If you name specific entries, one of them might not exist in the run's snapshot (the operator renamed or disabled it). You choose how the node reacts, via missing-context behaviour:

  • ignore — skip the missing entry silently and carry on;
  • ignore and report — skip it, but log a warning so the gap is visible in the execution log;
  • fail — treat it as an error: log it, terminate the run, and stop.

Pick based on how essential that context is. If a step genuinely can't produce good output without a particular entry, fail makes the dependency explicit; if the context is nice-to-have, ignore (or ignore and report) keeps the run going. When you include all entries rather than a named list, there's nothing to miss — so this setting only comes into play with an explicit selection.