whitelonng/mancode

What it solves

AI coding agents often suffer from "over-engineering," where they ignore existing project patterns, skip planning, or create redundant components. mancode acts as a local workflow layer that forces agents to align with your project's existing design system, coding standards, and requirements before writing code.

How it works

mancode integrates with your existing AI coding tools (like Claude Code, Cursor, or GitHub Copilot) by adding a structured governance layer. It manages tasks through specific "modes" (like solo for quick tasks or /man for complex, planned work) and maintains a local record of requirements, plans, and verification evidence. It ensures that any code generated is checked against your project's actual UI components, dependencies, and design tokens, preventing the agent from reinventing the wheel.

Who it’s for

It is designed for developers and teams using AI coding agents who want to maintain high engineering standards, ensure consistency across multiple sessions, and prevent sensitive data (like credentials or local paths) from being accidentally shared or leaked.

Highlights

• Structured Modes: Choose from various workflows like solo for daily tasks or /man for rigorous, plan-based development.
• Privacy Protection: Automatically identifies and redacts sensitive information like credentials or personal data before sharing logs or configurations.
• Context Preservation: Keeps track of project decisions, task history, and requirements across different chat sessions and AI platforms.
• Project Health Checks: Includes tools to scan for stale dependencies, unused code, and inconsistent design patterns.
• Evidence-Based Delivery: Ensures that tasks are only marked complete after meeting defined acceptance criteria and verification steps.