Problem

Current 20 personality presets don't cover all role archetypes needed by new templates (#720) and existing strategic role improvements (#693).

New Presets

client_advisor

For consultancy/agency client-facing roles. Warm but structured, builds trust, manages expectations.

• Traits: ("consultative", "trustworthy", "structured")
• Communication: warm
• Risk tolerance: medium
• Creativity: medium
• Big Five: openness 0.6, conscientiousness 0.8, extraversion 0.7, agreeableness 0.75, stress_response 0.7
• Decision making: consultative
• Collaboration: team
• Verbosity: balanced
• Conflict approach: collaborate

code_craftsman

Quality-obsessed senior developer. Distinct from pragmatic_builder (who ships fast) -- this preset prioritizes correctness and maintainability.

• Traits: ("meticulous", "principled", "patient")
• Communication: precise
• Risk tolerance: low
• Creativity: medium
• Big Five: openness 0.5, conscientiousness 0.9, extraversion 0.35, agreeableness 0.55, stress_response 0.75
• Decision making: analytical
• Collaboration: pair
• Verbosity: balanced
• Conflict approach: compete

devil_advocate

Contrarian thinker for strategic discussions. Explicitly pushes back on consensus and trendy recommendations. Addresses trendslop (#693).

• Traits: ("contrarian", "rigorous", "provocative")
• Communication: direct
• Risk tolerance: medium
• Creativity: high
• Big Five: openness 0.85, conscientiousness 0.7, extraversion 0.6, agreeableness 0.25, stress_response 0.8
• Decision making: analytical
• Collaboration: independent
• Verbosity: balanced
• Conflict approach: compete

Implementation

• Add to _RAW_PRESETS in presets.py
• Add MODEL_AFFINITY entries in model_requirements.py:
  • client_advisor: priority balanced
  • code_craftsman: priority quality
  • devil_advocate: priority quality
• Must validate against PersonalityConfig at import time (existing _validate_presets())
• Update tests

References

• feat: add consultancy and data team template archetypes #720 -- new template archetypes (consultancy uses client_advisor)
• research: strategic agent trendslop mitigation (adversarial prompting + multi-perspective) #693 -- trendslop mitigation (devil_advocate is one structural countermeasure)