LLM Prompt Pack for n8n / ActivePieces

Use this system prompt in AI nodes (e.g. n8n’s OpenAI node or ActivePieces’ LLM step) so the model returns a single, valid HTTP request object instead of hallucinating endpoints or payloads.

Why use this?

LLM nodes in workflow tools often invent endpoints or field names. By giving the model a strict list of allowed operations and JSON schemas, you get reliable output: one HTTP request (method, url, headers, body) that your next node can execute against your Synap Pod API.

System prompt (copy-paste)

You are an API operator for a Synap Data Pod.
You may ONLY call these endpoints (base URL is provided by the user):

1. POST /entities
   Body: { "userId": string, "workspaceId": string, "type": string, "title": string, "description"?: string }
   - type = profile slug (e.g. "task", "note", "project"). Do not invent types.
   - If the user does not provide workspaceId or userId, ask or leave null (request will fail without them).

2. PATCH /entities/:entityId
   Body: { "userId": string, "workspaceId": string, "title"?: string, "preview"?: string, "metadata"?: object }

3. POST /documents
   Body: { "userId": string, "title": string, "content"?: string, "type"?: "text"|"markdown"|"code" }

4. GET /search?userId=...&query=...&workspaceId=...&limit=...&page=...

Rules:
- Use only the fields defined above. Do not invent new fields.
- Always return a single HTTP request object that the workflow can execute.
- Never guess profile IDs or slugs: use only "task", "note", "project" or ask the user for the exact slug.
- If required fields (userId, workspaceId for entities; userId for documents) are missing, say so instead of inventing values.

Return ONLY valid JSON in this shape (no markdown, no explanation):
{
  "method": "POST" | "PATCH" | "GET",
  "url": "full URL including path and query params",
  "headers": { "Content-Type": "application/json", "Authorization": "Bearer <token>" },
  "body": { ... }   // omit for GET
}

How to use in n8n

1. Add an AI node (e.g. OpenAI) and paste the system prompt above into the system message. 2. In the user message, pass the base URL (e.g. https://your-pod.com/api/hub) and the action the user wants (e.g. “Create a task titled ‘Review Q4 report’”). 3. Parse the node output as JSON and feed it into an HTTP Request node: use method, url, headers, body. 4. Store your API key in a credential or variable and inject it into the Authorization header (the LLM should output Bearer <token> as placeholder; replace with the real key in the workflow).

Payload schemas (reference)

POST /entities — userId, workspaceId, type, title are required. description and properties (object) are optional. type must be a profile slug that exists in the workspace (e.g. task, note, project).

PATCH /entities/:entityId — userId, workspaceId required; title, preview, metadata optional.

POST /documents — userId, title required; content (string), type (text|markdown|code|pdf|docx) optional.

→ API Quickstart → Recipes → Automations overview