Context Handling

Agent Actions builds prompt context differently depending on execution mode.

Mode Differences

Mode	Context Structure
Online	Input data under source namespace
Batch	Input data at root level

Online Mode

{
  "source": {
    "page_content": "Document text...",
    "url": "https://example.com"
  },
  "seed": { "exam_syllabus": {...} }
}

Analyze: {{ source.page_content }}

Batch Mode

{
  "page_content": "Document text...",
  "url": "https://example.com",
  "seed": { "exam_syllabus": {...} }
}

Analyze: {{ page_content }}

warning

The same template can fail in different modes. {{ source.field }} works in online but fails in batch.

Mode-Agnostic Templates

{% if source is defined %}
  Content: {{ source.page_content }}
{% else %}
  Content: {{ page_content }}
{% endif %}

Context Variables

Variable	Description	Available In
source	Input record (wrapped)	Online only
seed	Static seed data	Both modes
{action_name}	Previous action output	Both modes
Root fields	Input record fields	Batch only

Context Scope

Control data visibility with context_scope:

actions:
  - name: my_action
    context_scope:
      observe:
        - seed.exam_syllabus
        - prev_action.result
      drop:
        - source.sensitive_field

Best Practices

1. Choose one mode per workflow - Don't switch modes mid-workflow
2. Match template to mode - Use source.* for online, root fields for batch
3. Use seed data for static content - Works identically in both modes
4. Analyze schemas - Run agac schema -a my_workflow to check field dependencies

See Also

• Run Modes - Batch vs online execution
• Context Scope - Field visibility configuration