Skip to content

Generating scripts with AI➧

The API description in OpenAPI format is machine-readable, so it's convenient to hand it to an AI assistant (Claude, ChatGPT, etc.) and have it generate a ready script for your task — without reading the whole specification by hand.

Step 1. Get the API description➧

Download the specification from your server:

curl -s https://demo.wildcore.tools/swagger/openapi.yaml -o openapi.yaml

The file is large (dozens of methods). If your assistant has a context-size limit, cut out only the section you need: open Swagger, find the methods you want (groups are described on the API groups & sections page) and copy just those.

Step 2. Give the assistant context➧

For the script to work right away, always state three things in your prompt:

  1. Base URL — methods live under the https://<your-server>/api/v1 prefix.
  2. Authentication — the X-Auth-Key: <token> header (see Authentication).
  3. Response format — payload is in the data field, wrapped in {statusCode, meta, data}.

Step 3. State the task➧

Example prompt to the AI:

Here is a fragment of the Wildcore DMS OpenAPI specification (attaching openapi.yaml). Base URL is https://wildcore.example.com/api/v1, auth is the X-Auth-Key header. Write a Python script that: 1) fetches all devices (GET /device); 2) for each OLT-type device, gets the list of unregistered ONUs; 3) prints the result as CSV. Read the token from the WC_KEY environment variable, add error handling and pagination.

The AI returns a ready script; if needed, ask it to rewrite in another language (bash/PHP/JS) or add logic.

Tips for a good result➧

  • Narrow the context. Pass only the methods the task needs — the answer will be more accurate and the risk of "inventing" non-existent fields lower.
  • Ask it to stick to the schema. Add: "use only fields present in the specification".
  • Ask for parameterization. Token and address via environment variables, not hard-coded.
  • Iterate on errors. If the script returns an error, paste the server response (the field with the code and text) back to the assistant and ask it to fix it.
  • Test on safe data. First run the script against a test/demo server (demo.wildcore.tools, admin/admin), and only then against production.

Security➧

  • Never paste a real token into the AI conversation — use a placeholder; keep the real key in an environment variable or a secret store.
  • For integrations, issue a dedicated user with the required rights and a key with a limited lifetime — that makes access easy to revoke.
  • Read the generated code before running it, especially methods that change data (POST/PUT/DELETE).