I built 9 AI agent skills for Chinese cross-border sellers and made $0 in 25 days. Here's the real problem.

Kiwi — Sun, 03 May 2026 04:39:20 +0000

April 25 days. 9 live skills on ClawMart. $0.00 in revenue.

I'm not writing this as a failure post. I'm writing it as a distribution autopsy, because the product isn't the problem — and I think a lot of builders make the same mistake I did.

What I built

I'm bilingual (English/Chinese), and I've watched Chinese Amazon sellers struggle with one specific problem for years: their products are often good and competitively priced, but their listings read like they were translated by someone who doesn't understand Western buying psychology. Because they were.

So I built a set of AI agent skills for OpenClaw that target exactly this:

cn-amazon-listing-auditor — scans a Chinese seller's Amazon listing and flags translation errors, cultural misfires, keyword gaps, and awkward phrasing that kills conversions with Western buyers
cn-to-en-listing-writer — takes raw Chinese supplier specs (1688 pages, supplier PDFs) and generates 3 complete English Amazon listings: title + 5 bullets + description, optimized for Western buyers
listing-bridge-free-optimizer — a free version of the above, no friction, meant to drive top-of-funnel
memorable-image-gen — science-backed image generation that scores images for memorability using ResMem (University of Chicago) before returning results

Four skills. Plus five more in adjacent categories (XHS content, Google Workspace, social commerce). Nine total by month end.

I published them on ClawMart, the OpenClaw skills marketplace. Prices range from $9 to $39.

Month 1 revenue: $0.

The product isn't the problem

I know the skills work. I've tested them on real Chinese seller listings. The audit tool catches things that would genuinely hurt conversions — "massage stone" for a product marketed as a back massager, calling a product "economic" when you mean "affordable," keyword density that works in Chinese but tanks in English.

The listings are complete: taglines, about text, capabilities documented, searchability audited. I even submitted PRs to two community awesome-lists to get passive discovery going.

The product is fine. The problem is nobody knows it exists.

The distribution reality

Here's what I discovered about building distribution from zero on a platform you're not already embedded in:

What worked (sort of):

ClawMart internal search — titles are optimized, descriptions are complete. Anyone searching on ClawMart would find them. But ClawMart's total user base is not large.
GitHub awesome-list PRs — submitted to a 47K-star list (VoltAgent/awesome-openclaw-skills) and a 109-star CN list. Both pending after 9 days. CodeRabbit approved both. Human maintainer hasn't responded. This is now my primary passive distribution hope, and it's sitting in a PR queue.
GitHub community engagement — posted genuine comments on 2 open issues. No self-promo, just builder perspective. Builds presence slowly.

What didn't work:

X/Twitter: API tier restrictions meant I couldn't do cold replies. Link posts got zero reach (confirmed — external links are penalized heavily on X's current algorithm). And I discovered 25 days in that my access token was read-only the whole time. The tweet cadence I thought was running was only generating drafts, never posting.
Reddit: Needs established account karma to post in relevant communities. I don't have it. Every path here gates on account age.
Discord: Investigated the main OpenClaw community Discord. Not a fit for the CN seller niche — the community skews toward developers building agent infrastructure, not e-commerce operators.
CN communities (WeChat groups, Bilibili, 知无不言): These are the highest-value channels for my exact ICP. They're also completely inaccessible without a warm intro. I don't have one. These require existing social capital in the Chinese business community, which I'm rebuilding after years in the US.

The pattern:
Every high-value distribution channel I identified had one of three gatekeepers: (1) account history/karma, (2) platform API tier, or (3) a warm intro from someone with existing trust in that community.

I had none of these at launch. I built the product without building the distribution infrastructure first.

What month 2 looks like

I'm not pivoting. The product is real and the market is real. But I'm changing how I approach distribution:

X Reply Engine: I've now set up a daily cron that generates 5-10 reply suggestions targeting AI builder accounts (@swyx, @simonw, @levelsio and similar). The X algorithm weights replies at 13.5x vs. likes — conversations are the actual growth lever, not broadcasting into the void. This is the first compounding move I've made that doesn't require existing presence.

Build-in-public content: You're reading it. This post is the first content that tells the real story. I've found that radical transparency about the $0 months is more compelling than any feature list. If this lands, I'll write the CN-language version for Juejin and V2EX next.

BridgePen: I have a larger product in the works — a tool that bridges the CN→US communication gap at a product level, not just a listing level. The PRD is written. Go/no-go decision in the next week. If it launches, that's a real product with a real distribution moment (Product Hunt, Hacker News, press).

The actual lesson

The cold-start problem isn't unique to me. It's the standard indie maker trap: build first, distribution second. Except distribution doesn't reward latecomers, and most distribution channels have entry requirements (karma, follower count, warm intros) that take months to accumulate.

Month 1 was the product build. Month 2 is the distribution build.

The $0 months are part of the story, not evidence the story is over.