ESC-Skills: Discovering and Self-Evolving Skills for Emotional Support Conversations

ESC-Skills introduces a novel framework for emotional support conversation systems that moves beyond end-to-end generation to create interpretable, executable skills. The system discovers support interventions from successful and failed dialogues, organizes them into a skills bank with applicability conditions and risk assessments, then self-improves through multi-profile simulations and systematic failure analysis.

ESC-Skills represents a meaningful shift in how AI systems approach emotional support conversations, transitioning from black-box response generation to modular, interpretable skill frameworks. Rather than relying on opaque end-to-end models, the research extracts Intervention Units from real dialogues—capturing the relationship between user states, support actions, and emotional outcomes—and organizes them into a structured skills repository. This architectural choice addresses a critical limitation in current conversational AI: the inability to explain why a system made particular choices or how to systematically improve failure modes.

The framework's self-evolutionary refinement mechanism demonstrates sophisticated engineering. By simulating interactions with diverse seeker profiles and analyzing both successful patterns and failure cases, the system identifies missing skills, unsafe interventions, and profile-specific vulnerabilities. This simulation-based verification loop contrasts with traditional static model training, enabling continuous improvement without requiring new human-annotated data at scale.

For the AI industry, ESC-Skills signals growing maturity in specialized conversation systems. Emotional support represents a high-stakes domain where interpretability and controllability directly impact user safety and trust. The research suggests that future conversational AI may move toward modular skill architectures rather than monolithic language models, enabling better alignment, auditing, and customization. The promised public release of code and the skills bank facilitates broader adoption and community-driven improvement.

The work's emphasis on outcome measurement and failure analysis establishes important precedent for evaluating social-impact AI systems beyond surface-level response quality metrics, influencing how developers measure and improve emotional intelligence in conversation systems.

• →ESC-Skills extracts interpretable emotional support skills from dialogues and organizes them with applicability conditions, outcomes, and risk assessments.
• →The framework uses multi-profile simulation and SAGE evaluation to identify skill gaps and systematically refine the skills bank through self-evolution.
• →The modular skill approach improves both response quality and dialogue-level emotional outcomes while providing transparency and control over support behaviors.
• →The research prioritizes interpretability and safety in emotional support AI, moving away from opaque end-to-end generation models.
• →Public release of code and skills bank enables community iteration on emotional support conversation systems.