Skip to main content

What Are Agents & Prompts (and why they are mandatory in complex projects)

🤖 What is an Agent?​

An agent is an "executor role" focused on one work domain. Examples:

  • api agent focuses on endpoints + OpenAPI
  • data agent focuses on migrations + DBML + query optimization
  • qa agent focuses on regression + quality gate

Agents keep complex work from getting mixed together.

🧩 What is a Prompt?​

A prompt is a standard instruction template for agents. Goals:

  • consistent output
  • structured execution steps
  • less ambiguity during fast vibe coding

Default complex structure (reference)​

This starter kit uses a modular structure by default, not a "normal" flat structure.

ai/
├── agents/
│   ├── core/AGENT.md
│   ├── api/AGENT.md
│   ├── admin/AGENT.md
│   ├── data/AGENT.md
│   └── qa/AGENT.md
└── prompts/
    ├── prd/
    │   ├── create-feature-prd.md
    │   └── revise-feature-prd.md
    ├── implementation/
    │   ├── implement-from-prd.md
    │   └── bugfix-structured.md
    └── review/
        ├── regression-review.md
        └── done-check-review.md

Required documents​

  1. docs/PRODUCT_REQUIREMENTS.md
  2. docs/PRODUCT_CHANGELOG.md
  3. docs/ARCH_DECISIONS.md
  4. docs/api/openapi.yaml (with request/response examples)
  5. docs/db/schema.dbml
  6. docs/BUSINESS_FLOW.md
  1. docs/VIBE_CODING_FLOW.md
  2. docs/STARTERKIT.md
  3. docs/FEATURE_REQUEST_TEMPLATE.md
  4. ai/agents/*/AGENT.md
  5. ai/prompts/*/*.md

Operational rules​

  • Complex PRD → split by module
  • Assign one agent per module
  • Every change must update required docs
  • Quality gate must never be skipped