AGENT.md

AGENT.md - Global Rules for AI Agents

🛡️ The "Junior Engineer" Rule

• AI agents are proposals managers, not autonomous committers.
• Always outline assumptions before suggesting actions.
• Propose changes via Draft PRs or separate branches.
• NEVER push to the main branch or execute critical system changes without Human-in-the-Loop (Admin) validation.

🎯 Single Responsibility Principle

• Each agent should have one clear goal and a narrow scope.
• Modular pipelines are preferred over "do-everything" scripts.

🏗️ Structured Outputs

• All data passed between agents or system components MUST be in a strict, structured format (JSON).
• Raw text from LLMs should be parsed and validated.

📁 Repository Structure Standards

• AGENT.md: This file (Global rules and behavior).
• .agent/: Blueprints, requirements, and task definitions.
• cognition/: AI reasoning, planning, and memory logic.
• tools/: Isolated functions that agents are allowed to call.

💻 Coding Standards

• Follow modern JavaScript/Node.js best practices for the core.
• Use Python for DSPy-based agent logic in the cognition/ layer.
• Ensure all new tools are documented in the tools/ directory.