📝 Prompt Placeholder Conventions

This document defines the canonical conventions for writing prompt files in the Orchable system. Consistency across all templates is required for reliable orchestration.

---

1. Variable Delimiters

System Variables

Variables injected by the orchestrator/n8n at runtime use %%variable_name%% syntax.

Variable	Description
%%input_data%%	Primary input payload (JSON/TSV)
%%batch_number%%	Sequence number in high-volume batches
%%target_question_count%%	Total items to generate in this run

Content Placeholders

Internal templates (e.g., for Drag & Drop questions) use {{placeholder_name}}. These are NOT system variables and are NOT replaced by the orchestrator.

✅ System variable:   %%input_data%%
✅ Content blank:      The plant takes {{chat_1}} from air.

---

2. Global Section Order

All prompt files must follow this canonical sequence to ensure model predictability:

1. # Header: Title and metadata blockquote.
2. ## MISSION: Role definition and core constraints.
3. ## INPUT DATA: The %%input_data%% injection point.
4. ## REFERENCE: Domain-specific enums or knowledge.
5. ## INSTRUCTIONS: Numbered execution steps.
6. ## VALIDATION: Internal self-check list.
7. ## OUTPUT FORMAT: Strictly valid JSON schema.
8. END OF PROMPT footer marker.

---

3. Output Format Standard

Every prompt must return a root JSON object with an output_data array.

{
  "output_data": [
    {
      "id": 1,
      "result": "..."
    }
  ]
}

Last Updated: 2026-02-25