Skillry ↗: visual-style Agent Skills for websites, decks, images and video
- MRR
- - Verified 2026-10-18
- Total revenue
- -
- Active subscriptions
- -
Verified by TrustMRR
In one line: A library of visual-style Skills for AI agents (websites, slides, images, video), with free Skills and a paid membership for the full library.
Timeline
- Founded date listed on TrustMRR
- First month of revenue: $27 (TrustMRR)
Compared with similar products
Sources: TrustMRR: Skillry; TrustMRR: Uizze.
TrustMRR figures are monthly revenue as shown on 2026-10-23 (read-only sync from the payment provider).
Moves Skillry made
No public information yet
How to copy it
Pick an AI output people find generic, collect specific complaints, then see who asks "how did you make this"
Pick two kinds of output AI often makes in your niche (pages, slides, short videos, images), search the communities where your target users gather for complaints that something "looks AI generated", and choose the one most different people complain about with specific issues (color, layout, fonts). Then publish 1-2 samples that avoid those issues and note who comes asking how you made it, or for the file or prompt. What the founder did: many friends asked how his product's promo animation was made, so he packaged it as a Skill (X post by @yihui_indie).Expand ▾Collapse ▴
Pick two kinds of output AI often makes in your niche (pages, slides, short videos, images), search the communities where your target users gather for complaints that something "looks AI generated", and choose the one most different people complain about with specific issues (color, layout, fonts). Then publish 1-2 samples that avoid those issues and note who comes asking how you made it, or for the file or prompt. What the founder did: many friends asked how his product's promo animation was made, so he packaged it as a Skill (X post by @yihui_indie ↗).
- The chosen output type with complaint quotes (each with a working link, from different people); screenshots of people who asked how you made the samples or asked for the files.
- The people asking how you did it are all different and not friends you asked to cheer.
I want to build a visual-style Agent Skill so AI coding tools stop making "<your chosen output type>" that obviously look AI made. Two sets of material: A) complaint quotes I collected in <the communities where your target users gather>, each with handle and link; B) messages from people who asked how I made my samples or asked for files or prompts, each with handle. Please: 1) group A by specific problem (color, fonts, layout, motion, etc.), with how many quotes and how many distinct accounts each, removing self-promotion; 2) pick the 1-2 most complained-about problems that a fixed set of rules can solve; 3) answer separately: is there enough evidence people dislike it, and is there enough evidence people want it (count only distinct people in B); 4) judge only from my material and say plainly if it is not enough. Material: <paste>
Step 1: read install counts for similar visual Skills on the cards in the Skill list and note whether each is free or members-only; free ones get more installs, so don't read them as paid demand.
Step 1: check 12-month search interest for "claude skills", "agent skills" and "claude code"; go on only if at least one term is rising, and save a screenshot. Then type these terms into Google search and look at the autocomplete suggestions to see which styles people search for.
Define a style that doesn't clash with existing work, write it as a Skill and add sample work
For the problem from step 1, define a concrete style: a few hex colors (one accent), 1-2 fonts and a few layout bans aimed at the complaints. Give it a name that explains itself, write it as a SKILL.md in your AI coding tool's skills folder, check existing Skill directories first so it doesn't clash, then use it to make 2-3 sample pieces and take full-page screenshots. What the founder did: he wrote fixed visual styles as Skills, and every Skill in his directory shows real output (skillry.dev).Expand ▾Collapse ▴
For the problem from step 1, define a concrete style: a few hex colors (one accent), 1-2 fonts and a few layout bans aimed at the complaints. Give it a name that explains itself, write it as a SKILL.md in your AI coding tool's skills folder, check existing Skill directories first so it doesn't clash, then use it to make 2-3 sample pieces and take full-page screenshots. What the founder did: he wrote fixed visual styles as Skills, and every Skill in his directory shows real output (skillry.dev ↗).
- A SKILL.md with name and description frontmatter and explicit rules for details like mobile, navigation and price units; a comparison with existing similar Skills; full-page screenshots of 2-3 pieces made with the Skill.
- In a new session, invoking it produces work clearly in the same style as the samples.
Help me write a visual style as an Agent Skill (at .claude/skills/<style name>/SKILL.md). Output type: "<your chosen output type>"; problem to solve: "<the problem from step 1>"; style name: "<name>"; colors: <3-5 hex values, mark the single accent>; fonts: <1-2 font names and weights>; layout bans: <3 rules>. Write the SKILL.md: 1) frontmatter has only name and description, and description says what the user will say when this Skill applies; 2) the body is rules an agent can follow one by one, each concrete down to a value or a checkable practice, no "make it look nice"; 3) end with a pre-delivery checklist. Before writing, list everything I haven't decided as questions and wait for my answers; don't guess.
Step 2: go through Skills of the same type before defining your style, so you don't duplicate one.
Step 2: following Anthropic's official docs, put SKILL.md in .claude/skills/<name>/ and start a message with /<name> to make the sample pieces with it.
Step 2: optionally has Claude turn your style rules into a SKILL.md with a few test tasks; the prompt above works without it.
Run the same task without and with the Skill, capture the difference and blind-test it
Run one task prompt twice: A without the Skill, B with it, each in a fresh session, with no hand edits, and put full-page screenshots side by side; repeat with a different product. Remove the A/B labels, shuffle left and right, and privately send them to a few people not involved, asking only "which looks less AI made?". This comparison is the main material for your preview page and promo. What the founder did: every Skill on his preview page shows real output (skillry.dev).Expand ▾Collapse ▴
Run one task prompt twice: A without the Skill, B with it, each in a fresh session, with no hand edits, and put full-page screenshots side by side; repeat with a different product. Remove the A/B labels, shuffle left and right, and privately send them to a few people not involved, asking only "which looks less AI made?". This comparison is the main material for your preview page and promo. What the founder did: every Skill on his preview page shows real output (skillry.dev ↗).
- At least 2 A/B pairs (same prompt, fresh sessions, no hand edits, different products) and the raw blind-test replies for each.
- Most blind testers pick B; if not, revise SKILL.md from the comparison, rerun B and test again.
Below are screenshots of two outputs from the same task prompt "<prompt>": one without a Skill (A) and one with my "<Skill name>" (B), each from a fresh session with no hand edits, plus my SKILL.md. Please: 1) compare color, fonts, layout and copy item by item, especially whether the problems complained about in step 1 remain in A and B; 2) point out where B didn't follow SKILL.md or has issues, map each to the rule to change, and write the revised rule text; 3) describe only what is visible in the screenshots and don't judge which looks better. Screenshots: <upload> SKILL.md: <paste>
Step 3: optionally use its evals, which run each test task in a clean subagent with and without the Skill so you can compare the two outputs side by side.
Build a preview page: free Skills open, paid Skills behind a membership
Build a page where each Skill shows its real output from step 3 so people see the effect before installing. Offer one free Skill that installs without a card and lock the rest behind a membership, using a site template and payment tool you know. What the founder did: free Skills plus a monthly or yearly membership (TrustMRR); the preview site was built on the TanStarter template and deployed on Cloudflare (X post by @yihui_indie).Expand ▾Collapse ▴
Build a page where each Skill shows its real output from step 3 so people see the effect before installing. Offer one free Skill that installs without a card and lock the rest behind a membership, using a site template and payment tool you know. What the founder did: free Skills plus a monthly or yearly membership (TrustMRR ↗); the preview site was built on the TanStarter template and deployed on Cloudflare (X post by @yihui_indie ↗).
- The preview page has a public URL, at least 1 free Skill installs directly, and paid Skills show a preview but need payment to install.
- A stranger installs a free Skill or buys a membership (check analytics and the payment dashboard).
I'm building a preview page for visual-style Agent Skills that give AI coding tools a distinct style for "<your chosen output type>". Below are each Skill's name, SKILL.md, real without/with comparison screenshots, and which are free or paid. Write the page structure and copy: 1) a one-line hero saying output stops looking generic once installed; 2) one card per Skill, main image from real output with the Skill, clicking opens the A/B comparison; 3) free Skills marked "free, no card" with install paths for Claude Code (.claude/skills/), Codex (.agents/skills/) and Cursor (.cursor/skills/); paid Skills show only the preview and a join button; 4) use only my screenshots and content, inventing no results or install counts. Material: <paste>
Step 4: starts the preview page from this TanStack Start boilerplate (auth, payments, blog, SEO pages), which the author says Skillry was built on.
Step 4: hosts the preview page; the author picked TanStarter for its Cloudflare cost advantage, and TrustMRR lists Cloudflare in Skillry's backend.
Step 4: takes the membership payments (monthly or yearly); TrustMRR lists Stripe as Skillry's payment provider.
Make a promo with the Skill itself and see which channel brings paying users
Use an AI coding tool with a video editing plugin to turn the step 3 comparison into a 30-40 second promo and post it in the channels where your target users are. Tag preview page links by channel and compare installs and payments per channel, not just views. What the founder did: he used Claude Code driving the ChatCut plugin to make the promo (X post by @yihui_indie).Expand ▾Collapse ▴
Use an AI coding tool with a video editing plugin to turn the step 3 comparison into a 30-40 second promo and post it in the channels where your target users are. Tag preview page links by channel and compare installs and payments per channel, not just views. What the founder did: he used Claude Code driving the ChatCut plugin to make the promo (X post by @yihui_indie ↗).
- The promo is posted in at least 2 channels and the preview page tracks visits and installs or payments per channel.
- Per-channel data shows at least one channel bringing installs or payments, not only views.
Make a 30-40 second promo for my visual-style Skill preview page (<URL>): first the generic version made without the Skill, then real output with "<Skill name>" installed, then how to install the free Skill and where to join. Every on-screen line must come from that page, and footage is only my comparison screenshots and recordings. Export one 16:9 and one 9:16 and tell me the file paths. Material: <upload>
Step 5: runs a "URL in, promo film out" Skill that drives ChatCut, which the author says is how Skillry's promo animation was made.
Step 5: as a Claude Code plugin, it generates and edits the promo on an editable timeline (motion, captions, voiceover, export); the author used it for Skillry's promo.
Sources
Next case
Lancer: AI agent for Upwork proposals
Business model: Vertical AI agent SaaS subscription
Data source: Starter Story, checked 2026-10-15
