Code Quality & Rules

Relevant source files

The following files were used as context for generating this wiki page:

• CLAUDE.md
• README.md
• package.json
• lib/ai/prompts.ts
• lib/db/schema.test.ts
• tests/e2e/api.test.ts

Code quality in the LV.AI project is governed by strict compliance with construction industry standards (GAEB, VOB, ÖNORM, SIA), technical linting and testing requirements, and specific AI interaction protocols. These rules ensure that the autonomous AVA-Copilot generates legally sound and technically accurate construction documents for the D/A/CH region.

Project Core Rules

The LV.AI project operates under five primary mandates that dictate architectural and functional requirements.

• GAEB-Konformität: All Bill of Quantities (LV) data must follow the GAEB DA XML structure (X81–X86).
• D/A/CH-Compliance: The system performs automatic adjustments based on the target country's legal framework (VOB for Germany, ÖNORM for Austria, SIA for Switzerland).
• Projekt = Chat: Every chat session represents a specific construction project containing associated metadata.
• Dual Memory: Persistence is managed through a combination of project-scope and global-scope memory via mem0.
• Deutsche Fachsprache: All user interface elements and AI responses utilize professional German construction terminology.

Sources: CLAUDE.md:29-34, README.md:14-22

Verification Procedures (LV.AI-Prüfverfahren)

Before completing any task or code change, developers must follow a mandatory 19-point verification process. The last five points are specific to the LV.AI domain:

Step	Rule	Target Component
15	GAEB-Konformität	lib/gaeb/, preisspiegel-actions, bieter-actions
16	mem0-Scope Integrity	Memory layer, mem0-mcp
17	Legal Correctness	lv-queries, preisspiegel-actions, bieter-client
18	Data Integrity	lv-queries-offers
19	Infrastructure	Hostinger connectors, DNS, VPS

Sources: CLAUDE.md:68-75

AI Interaction & Prompt Rules

The system utilizes modularized prompts to maintain consistency and optimize token usage. Prompts are split into stable parts (roles, trades, language) and dynamic parts (request hints, project context).

Prompt Constraints

The AI system follows strict operational rules when interacting with artifacts and documents:

1. Single Tool Execution: The AI must call only one tool per response and stop immediately after.
2. No Content Duplication: After creating or editing an artifact, the AI does not output the content in the chat.
3. German Language: Titles must be 2-5 German words without prefixes or formatting.

Sources: lib/ai/prompts.ts:34-43, lib/ai/prompts.ts:168-181

AI Architecture Flow

This diagram illustrates how request hints and project context are integrated into the system prompt generation.

Technical Standards & Testing

The project employs a modern tech stack with specific tools for linting, fixing, and testing to maintain code health.

Development Scripts

Key scripts defined in package.json enforce code quality:

• pnpm check: Runs ultracite check for linting.
• pnpm fix: Runs ultracite fix for automatic formatting.
• pnpm test: Executes unit tests via vitest.
• pnpm test:e2e: Executes end-to-end tests via playwright.

Sources: package.json:8-23

Testing Coverage

The testing suite is divided into unit/integration tests and E2E tests to verify API behavior and UI state.

Database Schema Constraints

Data integrity is maintained through specific Enums and table structures that reflect construction industry requirements.

Legal & Format Enums

The database schema enforces valid data entry for regional compliance and file formats:

• landEnum: Restricted to ['DE', 'AT', 'CH'].
• gaebFormatEnum: Supports formats X81 through X89.
• projektStatusEnum: Covers phases entwurf, ausschreibung, vergabe, ausfuehrung, abrechnung, and abgeschlossen.

Sources: lib/db/schema.test.ts:37-65

Core Data Models

Table	Description	Key Fields
Project	Root entity for construction projects	id, land, status
LvDocument	Represents a GAEB document	id, projectId, gaebFormat
LvPosition	Individual line items in an LV	oz, menge, einheit, kurztext
Offer	Bidder submissions	id, bieterStatus

Sources: lib/db/schema.test.ts:98-119

Conclusion

Code quality in LV.AI is a synthesis of technical linting and deep domain-specific compliance. By enforcing GAEB standards and regional legal frameworks through both the database schema and the AI prompting logic, the system ensures that autonomous operations remain within the boundaries of construction industry regulations. Accuracy is further verified by a mandatory 19-point check before any deployment.