add-model-descriptions

The add-model-descriptions skill is a specialized automation agent designed for developers maintaining the Chat UI interface. It streamlines the maintenance lifecycle of LLM model configuration by automating the discovery and documentation process for new models released via the HuggingFace Inference router. When new models become available at the router endpoint, this agent ensures the application's configuration files (prod.yaml and dev.yaml) remain current, accurate, and consistently formatted without manual intervention.

The process begins by querying the HuggingFace router API to identify newly added models, comparing these against the existing MODELS array in the repository's configuration. For every identified delta, the agent performs targeted web research to synthesize technical insights—including parameter counts, MoE architecture, and specific capabilities such as vision, coding, or agentic tool use. It then generates descriptive metadata following strict stylistic guidelines to ensure uniformity within the UI. This includes specific patterns for FP8 variants, regional models, and multimodal architectures, followed by an automated commit to the repository.

• Automated discovery of LLM additions using the HuggingFace router v1/models API endpoint.

• Comparative analysis between current production state and new router offerings to prevent configuration drift.

• Intelligent content generation for model metadata following a 8-12 word descriptive pattern focused on architecture and core capabilities.

• Direct modification of chart/env/prod.yaml and chart/env/dev.yaml configuration files.

• Automated Git operations for committing updates, reducing manual boilerplate task overhead for SvelteKit maintainers.

• Ensure the environment is properly authenticated with a valid HuggingFace API token to prevent fetch failures.

• The agent is optimized for models using the OpenAI-compatible protocol, which is the standard requirement for Chat UI model integration.

• Descriptions should be written as sentence fragments without periods, adhering to established style conventions for professional presentation.

• Always prioritize model architecture (e.g., MoE, dense) and primary use cases (e.g., coding, vision-language) in the generated descriptions.

• This skill should be triggered whenever the team releases new model versions to the router or upon the announcement of new model availability.