Why SKILL.md might be a better abstraction than function calling for agent tooling

We have SenseNova Office Skills(repo: github.com/OpenSenseNova/SenseNova-Skills ) for example: Skill Directory Structure: skills/ ├── sn-image-base/ # Tier 0 - Low-level tools (T2I, recognition, text optimize) │ └── SKILL.md ├── sn-infographic/ # Tier 1 - Auto prompt scoring, 87 layouts, 66 styles │ └── SKILL.md ├── sn-image-imitate/ # Tier 1 - Style imitation from reference images ├── sn-image-resume/ # Tier 1 - Resume image generation ├── sn-ppt-generate/ # PPT with template system ├── sn-excel/ #