For each

Step · Stable

Loop over a list, running the body once per item (sequential).

Four invariants to know:

1. Items source is upstream — the foreach itself does not make a list; it iterates one that already exists in context (e.g. {{ memory_find.items }}).
2. The body is one path with a loop variable — inside the body, {{ <loop_var> }} (default: {{ item }}) is the current item.
3. The body must back-edge to the foreach — without it the foreach runs once and stops. This is the ONLY legitimate cycle in the graph (the validator allows this one back-edge and rejects all others).
4. After-edge runs once — whatever connects via the “after” label out of the foreach runs exactly once, with the foreach’s label equal to a list of each iteration’s results.

Execution is sequential (no concurrency). limit hard-caps the number of iterations. A human gate (Ask actor) inside the body pauses per item; resume rotates to the next.

When to use

Use when an upstream step produces a list (e.g. List Contacts returns values: [...], List from memory returns neighbour entities) and you need to run the same body — enrich, notify, update — for every item. The loop variable (default: item) is the current array element; inside the body, primitives bind directly as {{ item }} (numbers, strings) while objects expose fields via {{ item.field }}. Nested access works the same way: {{ item.address.city }} for nested objects and {{ item.roles[0].title }} for nested arrays. The after-edge runs once when all items are processed.

When not to use

For a single conditional split use Branch or Switch — they are far cheaper when there is only one item to route. For parallel fan-out (multiple concurrent asks), use fire-and-continue Ask Actor with a downstream Checkpoint; For each is strictly sequential.

Inputs

items

string · required

A {{ reference }} to a list to iterate.

Example: {{ memory_find.items }}

loop_var

string · optional

The name each item binds to inside the body (default: item).

Example: item

limit

int · optional

Optional hard cap on how many items to process.

Example: 5

---

Auto-generated from the skill registry (load_skills()). Do not edit by hand.