Welcome to the Accelist Standards repository!
This guide outlines our organizational paradigm shift: Vibe Coding. Rather than typing line-by-line, developers now act as strategic architects and directors. By partnering with AI tools like Cursor, Replit, Bolt, or v0, we can vastly increase velocity and focus our human intellect on core business logic.
But remember: AI is a powerful assistant, not a senior full-stack developer. To succeed, you must master the "language" of interacting with it.
To prevent hallucinations, ensure robust security, and enforce project standards, strictly follow these five steps in your workflow:
Before instructing the AI, establish the "laws" of the codebase by providing specific foundational documents. If using IDEs like Cursor, attach these files as rules globally or in the current session.
Required Standards:
fe-standard.md- Frontend (React/Next.js) patterns & conventions.be-standard.md- Backend (FastAPI/Node) structures & logic guidelines.security-standard.md- Mandatory security rules (Auth, OWASP, Data policies).
Refuse the temptation to vaguely ask the AI to "build an app." Build out a Project Requirements Document (PRD) first. Your PRD (saved as .md) should clearly dictate:
- Project Name & Brief Description
- Detailed list of Features and UI Components
- Expected Application Architecture (e.g., Tech Stack, libraries)
- Desired Project Folders/Directory Structure
Tip: A well-written PRD.md provides rich context, minimizing frustration and "thinking budget" costs.
Construct each request as a well-engineered prompt to get precise, reliable results.
- A (Aim): State the specific, core objective directly.
- C (Context): Provide background context (Why and Where).
- C (Character): Assign a persona (e.g., "Act as a senior React developer").
- E (Example): Give a snippet or desired output format.
- LI (Limitation): Set boundaries (what not to do or change).
- ST (Style): Demand specific coding styles, commenting, or patterns.
Never blindly trust AI results. Review code before accepting changes. If fixing a bug, don't just say "fix this." Provide rich context:
- The exact error message/console warning.
- The problematic line of code.
- The expected outcome.
Tip: Leverage Model Checking Platforms (MCPs) like TestSprite or Playwright to automate frontend testing and validation.
AI context windows degrade over time, leading to forgotten rules. Protect yourself:
- Commit your stable work before launching a massive AI refactor.
- If the AI breaks the app or hallucinates severely, revert using Git, clear the AI workspace context, and try again with a cleaner prompt.
While Vibe Coding accelerates development, it carries serious risks if you ignore the guidelines:
- Hallucination: AI will confidently invent fictional facts or libraries. Mitigation: Verify and test thoroughly. Clear context if it spirals.
- Data Leakage: Pasting proprietary API keys, trade secrets, or client data into web-based AI tools is strictly forbidden. Mitigation: Only use enterprise-managed AI tools with zero-data-retention guarantees.
- Introduction of Security Vulnerabilities: AI trains on insecure code and may reproduce SQL injections or XSS. Mitigation: Human code review is non-negotiable. Leverage
security-standard.mdrules. - Skill Atrophy: Over-relying on AI for fundamental logic harms your long-term problem-solving abilities. Mitigation: Understand the code it writes.
- Overfitting (Pattern Fixation): The AI might refuse to use efficient solutions, sticking to what it was trained on. Mitigation: Ask for diverse approaches.
Happy Vibe Coding! Architect boldly, prompt carefully, and review strictly.