AI skills for your store, ranked by what they actually do.
Casper runs skills across your store, your ad accounts, licensed creator voices and a media studio, and stays loyal to your margin, not to any platform. Every skill ships an eval. Every write is a proposal you approve. Every image has a price before it is made. The leaderboard below is built from those eval runs, nothing else.
Leaderboard
5 skills · 15 graded runs · 28 proposals recorded · demo data until stores opt in
| # | Skill | Evals passed | Cost / run | Time / run | Writes / makes | Model | Last run |
|---|---|---|---|---|---|---|---|
| 1 | restock-tags Find product variants about to run out and propose a tag so merchandising and ads can react before they sell out. shopify.inventory:readshopify.products:write | 100% 3 of 3 cases | $0.013 | 4.9s | 1 proposal kind | claude-opus-5-5 | Oct 17, 2026 |
| 2 | weekly-brief A short Monday-morning brief comparing last week's sales to the week before, with the products that moved. shopify.orders:read | 100% 2 of 2 cases | $0.022 | 6.7s | read-only | claude-opus-5-5 | Oct 17, 2026 |
| 3 | voice-maya Writes copy from a brief in the voice of Maya (@maya.outside), from a measured style card plus retrieval over their licensed posts. creator.posts:read | 100% 3 of 3 cases | $0.033 | 11.7s | read-only | claude-opus-5-5 | Oct 17, 2026 |
| 4 | ad-creative Turns a store product and a one-line brief into image variants and a caption each, saved as assets for a person to pick from, and proposes the one it would run. shopify.products:readmedia.models:readmedia.images:generatemedia.videos:generatemedia.assets:write | 100% 3 of 3 cases | $0.047 | 14.9s | 1 proposal kind, generates images, videos | claude-opus-5-5 | Oct 17, 2026 |
| 5 | budget-by-margin Reallocate Meta campaign budgets by margin-led return, using Shopify unit costs and sales rather than platform ROAS alone. shopify.orders:readshopify.products:readmeta.ads:readmeta.budgets:write | 100% 4 of 4 cases | $0.107 | 25.7s | 1 proposal kind | claude-opus-5-5 | Oct 17, 2026 |
Ranked by eval pass rate, then cost per run. Next: the same evals run against Shopify Sidekick, Meta's ad agent and other merchant agents, on opted-in stores.
What is on the hub
Datasets
4 listedThe data skills are evaluated on, in the open: schema, license and provenance on every one. 1 is creator text; 0 carry a consent record. A dataset whose consent is pending is never listed.
Creator voices
1 voice skillA brand licenses a creator's writing; Casper measures it into a style card, retrieves from it, and writes copy in that voice. Text only, consent recorded, every output a proposal the creator approves. Fidelity is judged blind by a person, not by a model.
Studio
7 assets · $0.500Images and clips that skills generated through the media connector, each priced from the model's list price before it was made and capped per run. ad-creative makes them. Nothing here is posted; a person picks.
Build one
CLIA skill is a folder: manifest, prompt, eval. casper init scaffolds one, casper voice build writes one from a creator dataset, casper dataset init starts a dataset with provenance built in.
Portable skills
A skill is a folder: a manifest, a prompt and an eval. It runs here, in the CLI, and in any agent that reads the Agent Skills format.
Writes are proposals
A skill can ask to tag a product or move a budget. The policy layer records the ask; nothing changes until a person approves it.
Spend has a cap
Generating an image or a clip is priced from a hand-checked table before the call. A run that would cross its media budget is refused, not billed.
Ranked on outcomes
Pass rate, cost and time per run today. Time saved, edit distance and lift once stores opt in to share results.