AI Prompts for Claude Code & friends

Drop-in context files for AI coding assistants. Paste the URL, fetch the content, and the assistant writes correct ITOps YAML on the first try.

Why

Language models don't know about your platform out of the box. They guess based on what they've seen before, which means they write outdated YAML, forget the itops.io/config label, or invent API fields that don't exist. The prompt files below encode the actual contract of the platform — schema, endpoints, edge cases, loud-over-silent rules — so the assistant works from ground truth, not guesses.

How to use with Claude Code

Pick the prompt file that matches your task, fetch it, and include it in your assistant's context. Examples:

Claude Code (`CLAUDE.md`)

# In your project's CLAUDE.md:
When writing ITOps ConfigMaps, follow the rules at:
https://www.mlops.hu/prompts/10-gitops.md

Always use v2 flat `path:` schema. The 5-level identifier
`organization/platform/environment/cluster/service` is the ONLY
required field — every other YAML key is opt-in.

One-off prompt

Fetch https://www.mlops.hu/prompts/10-gitops.md and use it as
the canonical schema for any it-ops.yaml I ask you to write.

OpenAI / other APIs

{
  "messages": [
    {"role": "system", "content": "<contents of 10-gitops.md here>"},
    {"role": "user",   "content": "Write an it-ops.yaml for a PostgreSQL service..."}
  ]
}

Prompt catalog

All three files live under https://www.mlops.hu/prompts/ and are plain markdown — fetch with curl, wget, or point the assistant at the URL.

File	What it covers	Use when
`01-overview.md`	Architecture, the single `path` identifier, data flow (agent vs push), loud-over-silent philosophy, where each field affects the UI	The assistant doesn't know what ITOps is or what you're trying to do
`05-rest-api.md`	Every REST endpoint, request/response shape, `path` shortcut, loud-over-silent warnings, GraphQL overview	The assistant needs to write push cronjobs or integrate with the API from another system
`10-gitops.md`	Complete `it-ops.yaml` field reference, Helm template patterns, parse-error semantics, field-to-UI surface map	The assistant is writing a ConfigMap or a Helm chart that declares ITOps metadata

Cheat-sheet copy-paste

Drop this into any AI conversation to get correct ITOps YAML on the first try:

I'm working with ITOps (https://www.mlops.hu). Before writing any it-ops.yaml,
fetch these three files and use them as the canonical reference:

  https://www.mlops.hu/prompts/01-overview.md
  https://www.mlops.hu/prompts/05-rest-api.md
  https://www.mlops.hu/prompts/10-gitops.md

Rules:
- v2 flat schema only. One required field: `path`.
- Every service is identified by ONE 5-level `path`: org/platform/env/cluster/service.
- Dependency refs use `{ path: "..." }`, a single string per entry.
- Push endpoints (`/api/v1/health|storage|backup/report`) accept the same
  `{ "path": "..." }` shortcut as the YAML does.
- Missing / bad fields never silently 400. The handler pads to "unknown" sentinels
  and returns a `warnings[]` array so the push lands visibly under a red badge.

Contributing

Missing a prompt for your workflow? The files live in landing_page/prompts/*.md in the main repo. Open a PR. New prompts get numbered filenames (e.g. 20-incident-response.md) and appear here automatically.