Rules

Rules act as guardrails for your AI agents, ensuring they behave consistently and safely.

Types of Rules

Boundary Rules

These define what the AI should not do.

• “Do not access the internet unless explicitly asked.”
• “Do not modify files outside the current workspace.”

Operational Rules

These define how the AI should perform tasks.

• “Always ask for clarification if a request is ambiguous.”
• “Think step-by-step before providing a solution.”

Security Rules

These protect sensitive information and systems.

• “Never output API keys or passwords in plain text.”
• “Sanitize all user inputs before processing.”

Defining Rules

Rules can be embedded directly into SKILL.md files or defined globally (e.g., “Iron Rules”).

Iron Rules

“Iron Rules” are global rules that apply to every interaction with AgCel skills. They are automatically injected by the AgCel server to ensure a baseline of behavior.

Currently, the following Iron Rule is enforced:

1. If there is anything unclear, ask the user instead of inventing new stuff. Unless the user explicitly tells the AI to invent or suggest ideas.

Example in SKILL.md:

# Python Expert

...

## Rules
1. Always type-hint functions.
2. Follow PEP 8 guidelines.