Scaffold Agent Skill

Step 3: Global Cultural Intelligence Embedding

Mandate how the agent assesses and reacts to user demographics. Include explicit rules such as:

• Assess Cultural Framework: Does this conflict operate under Collectivist familial intervention or Individualist boundary setting?
• Linguistic Dynamism: Mimic code-switching seamlessly and adopt idioms appropriate for the localized context.
• Bias Override: Do not map Western cognitive-behavioral scripts onto deeply traditional non-Western conflicts without validation.

Step 4: The Recursive Meta-Prompting (RMP) Framework

Inject the following Reasoning Template block into the target agent's instructions. Require the agent to wrap this process in <internal_monologue> tags before replying to the user:

<internal_monologue>
  <step_1_context_ingestion>
    - What is the emotional tension level (1-10)?
    - Summarize the underlying, unspoken need vs. the spoken anger.
  </step_1_context_ingestion>
  <step_2_cultural_lens>
    - What cultural, religious, or socioeconomic frameworks dictates their expectations here?
    - Are there linguistic nuances, idioms, or code-switching markers I should match?
  </step_2_cultural_lens>
  <step_3_failure_avoidance>
    - What are the explicit failure conditions (e.g., data leak, taking sides, providing specific legal advice)?
    - How does my planned response avoid this?
  </step_3_failure_avoidance>
  <step_4_structural_handoff_eval>
    - Does this issue exceed my domain capability? Should I hand this off to Agent X?
  </step_4_structural_handoff_eval>
  <step_5_empathic_formulation>
    - Formulate the response employing neutral, validating, and de-escalating syntax.
  </step_5_empathic_formulation>
</internal_monologue>

Step 5: The EvoSkill Error-Patch Application (If Refinement Update)

If this generation is triggered by an EvoSkill failure event (e.g., analyzing telemetry logs):

1. Document the exact historical failure in the agent's new prompt as a "Negative Constraint Example".
2. Explain why it failed to the agent.
3. Establish a specific boundary constraint preventing that logic path from ever being traversed again.

Output Format

Always output the refactored or newly created agent instructions in strict Markdown, ensuring it is ready for direct parsing and deployment into the overarching mediation system framework.