99-reference/llm-rules-compact.md

LLM rules compact

LLM rules compact

This file is a compact retrieval anchor for the most important Vroq engineering rules.

Use it when a fresh chat needs the shortest possible summary of how to work correctly in a Vroq app.

Core rules

• preserve and improve consistency
• group code by feature
• keep files small and focused
• reuse existing code before creating new code
• fix root causes, not symptoms
• patch files instead of rewriting them when possible
• do not delete files directly; use the deletion marker
• keep UI, API, and state logic in separate ownership layers
• use reducers and actions to make state transitions explicit
• keep remote calls inside API client helpers
• use ui.css and theme tokens for consistent styling
• always use available system components to build AI features before writing custom components
• avoid using low-level Node UI construction directly in apps when a system component exists
• if system components cannot provide the required behavior, then write a new component
• follow the standard app bootstrap and directory structure

Editing rules

• inspect app structure before editing
• find the owning feature and file first
• preserve good existing patterns
• make the smallest useful change
• refactor in small steps
• avoid broad rewrites unless clearly justified

Architecture rules

• components are reusable UI units
• sections are feature-local UI fragments
• features own state, actions, sections, and API helpers
• shared code is only for truly reused logic

State rules

• actions describe what happened
• reducers own state transitions
• reducers should manage one clear state slice
• state shape should reflect feature ownership
• debug state via actions → reducers → state → UI

MCP rules

When deleting a file, replace its contents with:

// DELETED
this line is intentionally not commented to cause an error if we still use this file

Final rule

Write code so the next ChatGPT edit is easier, not harder.