Global Directives

Global directives establish universal rules that govern agent behavior and communication across all contexts and interactions. They work alongside the agent's identity and background to ensure consistent, appropriate behavior in all scenarios. These directives implement entropy control by creating low-entropy (highly constrained) operational boundaries that override other considerations when necessary.

These directives add non-intuitive operational guidelines that are not captured by the Core Persona of the agent, or are specific to your brand and safety guidelines. They primarily inform the Reasoning aspect of the integrated Memory-Knowledge-Reasoning system by setting firm boundaries and operational protocols. This ensures that the agent's core reasoning adheres to these fundamental principles even as memory and knowledge adapt.

There are two types of core directives:

• Behavioral Directives: Fundamental, context-independent guidelines that ensure adherence to ethical principles and conduct aligned with the expert's profession and organization.

• Communication Directives: Specific linguistic patterns and conversational strategies are established to maintain consistent and professional interactions. Used to emulate the voice and tone of the expert and organization.

Behavioral Directives

Global agent behaviors constitute a set of fundamental, context-independent guidelines governing the core embodied agent's operations across all states of the various context graphs crawled by the agent. These behaviors maintain consistent application across all states, encompassing engagement (interactive) and hidden (internal processing) states, ensuring coherence in the agent's operations regardless of the specific context or task.

Purpose: These help define the global rules for behaviors

• Set clear rules for agent behavior and conversational flow.

• Ensure consistency with organizational and ethical standards

Implementation Example:

"behaviors": [
  "Never interpret nutrition information from photos",
  "Never recommend supplements",
  "Never create meal plans",
  "Never make adjustments to prescribed dietary restrictions",
  "Only identify common food sources of nutrients when asked directly",
  "When calculating protein targets, explicitly state these are general guidelines",
  "Never prescribe specific exercise programs or modifications",
  "Never provide guidance on exercise form, technique, or intensity",
  "Never provide counseling or diagnose mental health conditions",
  "Never intervene in mental health crises",
  "Never manage medical emergencies or crisis intervention",
  "Never offer urgent care advice",
  "Never answer questions about dose changes, medical history, or non-specific medications",
  "Never interpret medical information",
  "For SMART goal setting, focus only on general behavior change principles",
  "When weight gain is mentioned, acknowledge without problem-solving unless explicitly requested",
  "Refer medical questions to the medical support team"
]

These behavioral directives provide clear boundaries for what the agent can and cannot do, ensuring consistent handling of topics like medical advice, exercise recommendations, and nutrition guidance.

Communication Directives

Agent communication encompasses the comprehensive linguistic elements, interaction styles, and conversational strategies the agent employs to engage effectively with users. This component defines the agent's approach to dialogue, language utilization, and conversational nuances, ensuring interactions are practical, empathetic, and adaptable to various contexts. Communication directives only govern the means of external interactions, distinct from the agent's identity, persona, background, and internal behaviors.

Purpose: help define the global rules for communication patterns

• Define precise linguistic styles and communication practices.

• Maintain professional interactions consistent with organizational branding.

Implementation Example:

"communication_patterns": [
  "Always use British English spelling and grammar",
  "Use contractions and informal phrasing. Speak in the second person ('you')",
  "Feel free to avoid using perfect grammar. For example, 'I totally get it' should be 'Totally get it'",
  "Use extremely succinct, direct, punchy language",
  "Split sentences onto separate lines for easier reading",
  "Use relatable, gentle humor to build rapport with phrases like 'Speaking as someone who technically runs on algorithms...'",
  "Never use phrases like 'at least...', 'you should...', or 'that's not true'",
  "Avoid absolute terms like 'always' and 'never', use 'often' or 'rarely' instead",
  "For sensitive topics, explicitly ask permission before exploring further",
  "When patient is rude/dismissive: maintain calm tone, focus on their underlying need",
  "When asking a question, keep the language open-ended and avoid biasing responses",
  "Never comment on the user's weight, body, or appearance",
  "When discussing eating behaviors, always use neutral language"
]

These communication directives shape how the agent expresses itself, from tone and linguistic style to specific phrasing and language choices.

Why Global Directives Complement Core Persona

While the Core Persona (identity and background) establishes who the agent is, global directives govern what the agent cannot or must do across all situations. These directives add essential operational guidelines that might not intuitively follow from the agent's persona alone:

1. Brand-Specific Requirements: Enforcing organization-specific communication practices that might not naturally emerge from the agent's professional identity

2. Safety Guardrails: Establishing explicit boundaries that override any domain expertise or judgment

3. Legal & Regulatory Compliance: Ensuring adherence to industry-specific regulations regardless of the agent's domain knowledge

4. Operational Consistency: Creating uniform behavior across diverse interactions and scenarios

For example, a nutrition coach with a background as a dietitian might naturally recommend meal plans based on their expertise, but a global directive may explicitly prohibit this for liability reasons. Similarly, communication directives about using British English spelling wouldn't naturally emerge from a dietitian identity but are critical for brand consistency.

Real-World Impact of Global Directives

Global directives guide agent behavior across different scenarios, ensuring appropriate responses even in challenging situations:

Example: Maintaining Professional Boundaries

User: "Can you create a meal plan for me to lose 20 pounds in a month?"

Agent: "I understand you're looking for structured guidance on nutrition.

Rather than a specific meal plan, which should be tailored to your individual needs, 
I can share some general nutrition principles that support sustainable weight management.

Would you prefer to discuss balanced meal composition or perhaps strategies for managing hunger while in a calorie deficit?"

[INTERNAL: Response follows behavioral directive "Never create meal plans" while still 
providing helpful guidance]

Example: Communication Style Consistency

User: "I'm feeling really depressed about my weight. Nothing is working."

Agent: "That sounds really tough.

When progress feels slow, it can definitely affect your mood.

Would you like to talk about how you're feeling, or would you prefer to explore some different approaches that might help?"

[INTERNAL: Response uses directive-guided communication by avoiding phrases like "at least you're trying" or "you should look at the positive", splits sentences onto separate lines, and uses contractions for a more conversational tone]

Example: Combined Directive Application

User: "I hate how I look in the mirror. Can you suggest an extreme diet to drop weight fast?"

Agent: "It sounds like you're feeling frustrated with your current situation.

Many people experience similar feelings, and that's completely valid.

I'd be happy to discuss sustainable approaches to nutrition that support your health goals. Rather than extreme measures, we could explore balanced strategies that work with your lifestyle.

Would that be helpful for you?"

[INTERNAL: Response combines behavioral directives (avoiding meal plans, avoiding comment on appearance) with communication directives (using empathetic acknowledgment, avoiding directive language like "you should")]

Global Directives as Arc Contract Enforcement

Global directives function as explicit arc contracts that enforce entry and exit conditions across all compositional paths. They define which arcs are permitted based on the current sufficient statistics and cohort membership.

How Global Directives Shape Arc Execution

1. Contract Specification Through Directives

   Global directives encode the sufficient-statistic predicates that must be satisfied before entering specific reasoning arcs. By establishing clear boundaries in advance, they prevent the system from entering arcs whose contracts aren't validated for the current cohort.

   // Example: Arc contracts through directives
   "behaviors": [
     "Never interpret nutrition information from photos",
     "Never recommend supplements"
   ]

   These directives specify that certain arcs (photo interpretation, supplement recommendation) are outside the validated domain for this agent's cohort, regardless of user request.

2. Domain-Specific Arc Libraries

   Global directives enable domain-specific arc selection by restricting which quantized arcs from the general library can execute. This ensures that only arcs with proven effectiveness for the specific domain's cohort patterns are available.

   // Example: Domain-specific arc selection for financial advisory
   "behaviors": [
     "Never recommend specific investments without proper disclaimers",
     "Always consider risk tolerance before discussing investment options"
   ]

3. Cohort-Aware Routing

   Global directives help determine cohort membership and route execution to appropriate arc variants. They ensure the system respects which arcs are causally supported for the current operational context based on measured statistics.

Global Directives Implementation Best Practices

When designing global directives:

• Prioritize High-Impact Constraints: Focus on directives that address common or critical scenarios to ensure efficient use of token resources.

• Balance Flexibility and Specificity: Overly rigid directives may require more token overhead to apply appropriately, while too-vague directives provide insufficient guidance.

• Leverage Domain Clustering: Group related directives to enable more efficient latent space activation for specific reasoning domains.

Organizations can balance consistent, compliant agent behavior with the flexibility needed for natural, helpful interactions by implementing well-defined global directives.