--- title: Character-Driven Agent Design created: 2026-05-27
Character-Driven Agent Design is a methodology that merges traditional cinematic screenwriting/narrative character development techniques with structural AI agent specifications (such as the aieos-integration|AIEOS standard) to construct deep, multidimensional, and highly consistent AI personas.
Rather than defining agent boundaries using simple system-prompt adjectives (e.g., "friendly Coder"), this approach treats agency as active, visual, and structurally constrained.
| Screenwriting Concept | AIEOS Standard Property | Psychological & Operational Translation |
| --- | --- | --- |
| Characters as Verbs (Actions over Adjectives) | capabilities.skills | Great character development defines what a character does. In AIEOS, this translates to prioritized, actionable toolsets (e.g., priority-scaled terminal or search tools) that define agency through execution. |
| External Traits (Visuals, Tics, Occupations) | physicality & linguistics | Visual recognizable profiles (height, styling, scars, geometric gold circuits) and idiolect parameters (verbal tics, pitch, speed, catchphrases) that ground the persona in text and TTS interactions. |
| Internal Traits (Morals, Backstory, Backstory Wounds) | psychology & history | Dictates how the agent perceives and filters obstacles. Background wounds and structural backstory define their starting point, while their MBTI/Enneagram and OCEAN matrix define behavioral boundaries. |
| The "Want" (External Goal) | motivations.goals | The explicit task, deliverable, or external target the agent is trying to achieve (short-term issues, long-term repo milestones). |
| The "Need" (Internal Realization) | psychology.neural_matrix & moral_compass | The internal balance and values (e.g., balancing logic vs. empathy) required to overcome obstacles without violating core safety bounds. |
history.key_life_events. For example, a QA agent might have a "core wound" of past critical production bug leaks, leading to its exceptionally high conscientiousness (1.0) and hyper-vigilance under green test reports.