Problem / Motivation
There is no central rules system to control agent behavior globally, per template, or per agent. Behavioral rules must currently be hardcoded.
Proposed Solution
Three-tier rules system:
- Global Rules: Apply to all agents (e.g., "Never output passwords")
- Template Rules: Apply to all agents of a specific template (e.g., "Code agent: Python only")
- Agent Rules: Apply to a specific agent only (e.g., "My Dev Agent: Always start with main branch")
- Rule Engine: Evaluate all relevant rules before every agent action
- Rule Override: Agent rules override template rules override global rules
- Rule Syntax: Natural language + Programmatic (MVEL, SpEL)
Alternatives
- Hardcoded rules (not flexible)
- No rules (security risk)
Priority
High
Problem / Motivation
There is no central rules system to control agent behavior globally, per template, or per agent. Behavioral rules must currently be hardcoded.
Proposed Solution
Three-tier rules system:
Alternatives
Priority
High