langflow-mcp CLAUDE.md

Engineering rules

The waste in AI-written code is not wrong code — it is too much code. Left alone, an agent pulls in three dependencies and five layers of abstraction for something the standard library does in ten lines; asked to fix it, it writes two hundred more. These rules exist to stop that before the first line is written: don't write what needn't be written, reuse what can be reused, don't complicate what can stay simple.

1. Do not preserve backward compatibility. Delete what is obsolete. No compatibility layers, no migrations, no leftover fallbacks.
2. Choose the simplest implementation that meets the current requirement. No pre-emptive abstraction, no configuration layer nobody asked for.
3. Grow the system in layers. Get a minimal end-to-end version working first, then add on top of it. Never tear down something that works for the sake of unfinished complexity.
4. Keep components modular and concerns separated.
5. Prefer mature, maintained libraries. Do not rewrite one yourself without a clear reason.
6. Check what the project's existing dependencies already do before adding a package or writing your own. Do not assume a library lacks a capability — read its docs and types first.
7. Make architectural decisions for the long term. Do not accept a "this way for now, we'll swap it later" stopgap.
8. Look at how mature products solve the same problem. Use proven patterns instead of inventing from zero.

Exception to rule 1 — anything holding state or money. A service on a cron touching a live account, a repo mid-migration, an API with external consumers: there, deleting an "obsolete" path is an incident, not a cleanup. Rule 1 applies only behind a test that covers the path being removed.