@conardli

kb-retriever navigates hierarchical knowledge bases via data_structure.md indexing, enforces pre-processing of pdf and excel files through reference learning, then executes multi-round grep and targeted reads to answer questions while avoiding full-file loads.

web-video-presentation transforms articles or scripts into screencast-ready react web presentations with optional synthetic audio. procedurally detailed but written in chinese, assumes external references, and lacks practical trigger phrases for agent invocation.

gpt-image-2 covers text-to-image generation and editing across three runtime modes (garden local, host-native delegation, advisor-only), with mode detection as prerequisite and 18 categorized prompt templates in references/.

Build or redesign polished browser-rendered visual artifacts with HTML/CSS/JavaScript/React: pages, dashboards, prototypes, slide decks, animations, UI…