Skip to content

Latest commit

 

History

History
35 lines (27 loc) · 2.67 KB

File metadata and controls

35 lines (27 loc) · 2.67 KB

GEMINI.md - Engineering Mandates for Prompt Refiner

CRITICAL: For localized project context and module-specific instructions, always refer to the context.md hierarchy.

This file defines the engineering mandates and standards for the Gemini Prompt Refiner project. These rules are foundational and take precedence over general workflows.

1. Enhanced Project Detection & Contextual Awareness

  • ID: enhanced-detection
  • Category: Architecture
  • Description: Extend src/detectors/project-scout.ts to identify a wider range of modern frameworks and tools (e.g., Tailwind CSS, Prisma, Drizzle, etc.). Ensure detection logic handles diverse project structures like monorepos. The goal is to provide highly accurate, context-aware prompt refinements.

2. Standardized MCP Tool Response Format

  • ID: standardized-responses
  • Category: Architecture
  • Description: All MCP tools in src/core/server.ts MUST return a consistent, structured response. Responses should clearly separate the primary tool output from additional context or suggested actions. This ensures reliable interoperability with all MCP-compliant clients.

3. Comprehensive Automated Testing Strategy

  • ID: automated-testing
  • Category: Quality Assurance
  • Description: Maintain a comprehensive automated test suite using Vitest. Prioritize unit tests for logic in src/linters/ and src/refiners/, and integration tests for the MCP server handlers. All new features or bug fixes MUST include corresponding tests to ensure long-term stability and behavioral correctness.

4. Strict Type Safety & SOLID Principles

  • ID: type-safety-solid
  • Category: Development Standards
  • Description: Adhere to strict TypeScript typing; avoid any unless absolutely necessary. Rigorously apply SOLID and SRP (Single Responsibility Principle) across the codebase. Each module, class, and function should have one clear responsibility, facilitating maintainability and extensibility.

5. Gapless Traceability & Real-time Audit

  • ID: gapless-traceability
  • Category: Architecture
  • Description: All repository changes (commits) and agent outcomes (executions) MUST be logged to the Event Store and emitted to the Command Center Dashboard in real-time. This ensures a complete, verifiable audit trail from prompt intent to final code artifact.

6. Optimized Background Autonomy

  • ID: background-autonomy-opt
  • Category: Performance
  • Description: Background tasks (ingestion, correlation, lesson extraction) MUST run in parallel using Promise.allSettled to prevent slow LLM sampling/timeouts from blocking critical history ingestion.