Every API codebase has conventions: how errors are returned, how authentication is handled, what the response envelope looks like, how pagination works. When I ask an AI to add a new endpoint without encoding these conventions, I get an endpoint that technically works but doesn't match the rest of the API. It uses a different error format, a different auth pattern, different status codes. The sinc API template's CONTEXT and DATA bands carry the existing conventions as hard constraints, producing code that integrates without friction.
This example builds a new REST endpoint in an existing FastAPI service, encoding the existing conventions so the new endpoint matches the established pattern exactly:
{
"formula": "x(t) = Σ x(nT) · sinc((t - nT) / T)",
"T": "specification-axis",
"fragments": [
{
"n": 0,
"t": "PERSONA",
"x": "You are a backend engineer who writes consistent, well-structured REST APIs. You treat existing conventions as law — you extend patterns, you don't invent new ones."
},
{
"n": 1,
"t": "CONTEXT",
"x": "I'm adding a new endpoint to an existing FastAPI service. The service has established conventions I need to follow: (1) All responses wrapped in {data: ..., meta: {request_id, timestamp}}, (2) Errors return {error: {code, message, details}} with appropriate HTTP status, (3) Auth via Bearer JWT validated by a dependency called require_auth, (4) Pagination via limit/offset query params, max limit=100."
},
{
"n": 2,
"t": "DATA",
"x": "New endpoint spec:\nGET /api/v1/subscriptions\nAuth: required\nQuery params: limit (int, default 20), offset (int, default 0), status (optional: 'active'|'cancelled'|'all', default 'active')\n\nDatabase model (SQLAlchemy):\nclass Subscription(Base):\n id: UUID\n user_id: UUID\n status: str # 'active' | 'cancelled'\n mrr_cents: int\n started_at: datetime\n ended_at: Optional[datetime]"
},
{
"n": 3,
"t": "CONSTRAINTS",
"x": "Follow the existing response envelope exactly — {data, meta}. Use require_auth dependency. Do not add new dependencies not already in use. Return 400 for invalid status param. Return 401 when token missing/invalid (handled by require_auth). Use existing Subscription model, do not create a new one. Type hints required. No placeholder comments."
},
{
"n": 4,
"t": "FORMAT",
"x": "Single Python file. Imports first. Pydantic response schemas, then route handler. No test code. Max 60 lines."
},
{
"n": 5,
"t": "TASK",
"x": "Write the GET /api/v1/subscriptions endpoint following all existing conventions."
}
]
}AI code generation in a codebase context fails most often not on logic but on consistency. The model doesn't know your error format, your auth pattern, your response envelope. Without that information, it invents plausible defaults — which are correct in isolation but wrong in context.
The solution is explicit convention encoding. Put your response envelope format in CONTEXT. Put your Pydantic models in DATA. Put your existing dependencies in CONSTRAINTS. The model will then produce code that reads like it was written by the same engineer who wrote the rest of the service.
API prompt tip: If you have an existing endpoint that's a good example of your patterns, paste its code in DATA as a "pattern to follow." The model is excellent at pattern continuation — give it one complete example endpoint and it will match the style, error handling, and structure in the new one without being told every convention explicitly.
Add a GET /subscriptions endpoint to my FastAPI service. It should list subscriptions with pagination and authentication. Filter by status.
CONTEXT: Response envelope {data, meta}. Errors {error: {code, message}}. Auth via require_auth dep. Pagination limit/offset max 100.
DATA: Exact SQLAlchemy model + endpoint spec.
CONSTRAINTS: Follow existing patterns. No new deps. Type hints. Max 60 lines.
The raw prompt produces a FastAPI endpoint that returns a plain list, uses HTTPException directly (not your error format), and implements its own auth check. The structured prompt produces an endpoint that matches your existing codebase exactly and requires zero adaptation to merge.
Try AI Transform — Structure Your API Prompt Free