Rules Engine

The Rules Engine stores and resolves hierarchical rules in Cortex. Rules influence agent behavior by injecting context into prompts — from global conventions down to file-specific patterns.

Rule scopes

Rules are organized into five scopes, from broadest to most specific:

Higher priority scopes override lower ones when rules conflict.

Rule structure

typescriptCopy
{
  id: string;              // Unique identifier
  content: string;         // The rule text
  scope: RuleScope;        // global | project | agent | skill | file
  scopeTarget?: string;    // Name or pattern for scoped rules
  priority: number;        // Numeric priority (higher = more specific)
  source: RuleSource;      // "learned" | "manual" | "imported"
  confidence: number;      // 0.0–1.0
  enabled: boolean;
  createdAt: string;       // ISO timestamp
  learnedFrom?: string;    // Session key (for learned rules)
}

Hierarchical resolution

When building a prompt, the engine resolves applicable rules:

1. Collects all enabled rules from global → project → target scope
2. Filters by scopeTarget match (if applicable)
3. Sorts by priority (descending), then creation date (descending)
4. Injects into the prompt as a <rules> block

xmlCopy
<rules>
- Always use TypeScript strict mode
- Use pnpm, not npm
- Files in src/api/ must have JSDoc
</rules>

Learning rules

The engine can propose rules from observed patterns. Proposed rules start disabled with low confidence:

bashCopy
mayros rules learn    # Analyze session → propose rules

Proposed rules can be confirmed or rejected:

bashCopy
mayros rules list                    # Shows proposed (disabled) + active rules
mayros rules confirm <ruleId>        # Enable + set confidence to 0.8
mayros rules reject <ruleId>         # Delete proposed rule

Cortex storage

{ns}:rule:{scope}:{id} → {ns}:rule:content → "Always use strict mode"
{ns}:rule:{scope}:{id} → {ns}:rule:scope → "global"
{ns}:rule:{scope}:{id} → {ns}:rule:priority → "0"
{ns}:rule:{scope}:{id} → {ns}:rule:source → "manual"
{ns}:rule:{scope}:{id} → {ns}:rule:confidence → "1.0"
{ns}:rule:{scope}:{id} → {ns}:rule:enabled → "true"

CLI

bashCopy
mayros rules list [--scope global]         # List rules
mayros rules add "use strict mode" --scope global
mayros rules add "focus on security" --scope agent --target reviewer
mayros rules remove <ruleId>
mayros rules learn                         # Propose rules from session
mayros rules status                        # Show engine status

Related

• rules CLI — command reference
• Project Memory — conventions and decisions
• Agent Memory — per-agent persistent memory