What does your Skill do?
shop-creator-kit turns a seller's NAS folder of product photos and videos into organised, per-product media sets plus ready-to-post captions and narration scripts for short-form shop videos.
Target user / problem
Small TikTok Shop sellers who film on their phone and post several videos a day. Their media piles up unsorted, they reuse clips by accident, and writing a fresh caption and script for every product eats the time they should spend selling.
Solution
A Skill that runs on the seller's AI NAS: it inventories new media, tags each clip by product and scene, tracks which clips were already used, and drafts one caption and one narration script per product from the seller's own price list — never repeating a clip, never mixing products in a video.
How it works
Drop media into a NAS folder; the Skill tags it and shows a per-product shelf; the seller picks a product and a slot; the Skill returns three unused clips, a caption and a narration script; the seller approves, and the clips are marked used.
What makes it different
Generic AI video tools generate new footage and ignore the seller's real stock; this Skill organises and reuses the seller's own footage, enforces one-product-per-video and no-clip-reuse, and drafts from the seller's real price list instead of inventing prices.
NAS capabilities used
ug-filesystem (ls, find) inventories new media at ingest; ug-filemgr (mediainfo, tag-add-file) tags each clip by product; ug-photo (search-screen, vision-vlm) and ug-video (search) match clips to products; ug-document (qa) reads the price list at drafting; ug-filesystem (echo) records used clips in the ledger at approval. Every step reads only the seller's authorised folders.
Architecture
Markdown-only skill package following the Uliya Skill Development Standard (capability catalog version 2.2.0). A planner delegates to three experts — a tagger, a scriptwriter, and a checker. Media work runs locally; no credentials or personal data leave the NAS.
What is built / what is not
Built: the SKILL.md workflow naming real catalog actions, the tagger, scriptwriter and checker prompts, and the used-clip ledger design, tested locally on sample media. Not built: a run against real NAS hardware. Needs NAS testing: real NAS file events, indexing speed on a full media folder, and the approval flow on UGOS.
Testing
Core 1: ingest 30 mixed clips for three products and check every clip is tagged to the right product. Core 2: draft three videos and check no clip repeats and every price matches the price list. Edge: a clip showing two products at once must be flagged, not silently assigned to one.
Risks / limitations
Tagging quality depends on footage clarity; the Skill drafts text but does not render video; NAS capability names and limits come from the official Uliya Skill Kit 0.7.7 and capability catalog version 2.2.0, which we have not yet run against real hardware. No third-party AI service is required.
Built With
- markdown
- ugos
