The engineering jobs on this site, built for you — with tests that prove the work

Updated 13 September 2026

The guides here teach you how to brief someone else. This page is the other option: four jobs I build myself, at a fixed price, where the delivery includes a test suite and evidence that the test suite actually catches breakage. That second part is the whole difference. Anyone can hand you passing tests; passing tests are worthless until something has tried to break them.

Key takeaways

• Every delivery includes source you own, a test suite, and proof the suite catches breakage.
• Four fixed-price jobs: store data automation $280, custom MCP $600, price and stock monitor $220, named-defect page-speed fix $750.
• A job that cannot be enumerated goes hourly at $45, per the routing test.

What every delivery includes

• Source you own outright. Plain Python or TypeScript in your own repository. No hosted dependency on me, no licence, no per-seat fee, nothing to renew.
• A test suite that runs on your machine. One command, no test framework to install, exit code 0 or 1.
• Proof the tests work. I inject deliberate defects into the finished code one at a time and show you that each one turns the suite red. A test that stays green while the code is broken is decoration, and you should not pay for decoration.
• A README that names the failure modes. What happens on a rate limit, an expired token, a renamed field — written down, not discovered by you in production.

Proof, before you send anything

You do not have to take the above on trust. github.com/hello532/shop-mcp is public: a single-file Shopify catalogue and stock server for Claude, standard library only, no dependencies to audit. The install, the four tools, and the self-test are on the shop-mcp page. It is also shipped, not just written — published on PyPI and listed in the official MCP Registry, which validated that the io.github.hello532 namespace is mine before it would accept the entry. That listing answers for itself — one record, status: active — rather than being a badge I drew.

• 200 assertions in self_test.py, runnable with python3 shop_mcp.py --self-test.
• 23 deliberate defects injected by mutation_test.sh — a wrong pagination cursor, a swapped comparison, a dropped error branch — and all 23 turn the suite red. That is the number that means something.
• Zero third-party packages, so the whole thing is readable in one sitting by whoever you ask to check it.
• It installs and runs as a stranger gets it. uvx --from shop-mcp shop-mcp --self-test needs no clone and no credentials, and reports 183 assertions rather than 200: manifest.json (5 assertions), llms-install.md (11) and README.md (1) are deliberately not shipped into site-packages, and every one of those checks skips rather than fails when its file is absent. The count moves, the verdict does not. That difference is measured and written in the README instead of being rounded away.
• What that repository does not prove. It has never been run against a live Shopify store. There is no credential in it and no recorded API session. The queries are written to the documented schema and every path around them is tested against a transport double — but the doubles are my model of Shopify's behaviour, not Shopify. The $280 package is delivery of that same discipline against your store, not a claim that the public file has already done a live round trip.

1. Store data automation — $280

Three working days. The jobs that are tedious by hand and dangerous in bulk, done against the Shopify Admin GraphQL API rather than the CSV importer.

• Bulk edits across the catalogue: prices, tags, collections, metafields, variant options.
• Inventory or price sync from a spreadsheet, supplier feed, or another store, on a schedule.
• Scheduled exports: stock levels, order lines, or anything else you currently copy by hand.
• A dry-run mode that prints every change before it writes, because a bulk mistake on a live catalogue is expensive and silent.

A second public build, for this package specifically. shop-mcp above is read-only and has never touched a live store, so it is evidence of method rather than of this job. github.com/hello532/gpt-ads-feed is the scheduled half, running: a real Shopify Admin GraphQL client against API version 2026-04, not the CSV importer, emitting a daily full product snapshot.

• 277 assertions across three layers — function level, nine end-to-end scenarios, and the missing-dependency and missing-credential branches — in one command, bash tests/run_all.sh, with no network and no credentials.
• Three guards that refuse to publish rather than overwrite live data with a bad snapshot: an empty result, a row count under the configured floor, and a rejected-row ratio over its threshold. Each exits 3 and uploads nothing. With full-snapshot semantics one bad write delists an entire catalogue, which is why the failure mode is a refusal and not a warning.
• --dry-run writes the files locally and touches neither the upload nor the saved state, so the run you inspect is the run that would have shipped.
• Tokens and shop domains are redacted from logs and notifications, with an assertion covering it, because an alert channel is the usual place a credential escapes.

What it does not prove. It reads and exports; the bulk-write half of this package is not demonstrated by it. Its parquet output, SFTP upload, and one third-party API call have been through their guard branches only — no real bytes, no real connection — and its README says so in those words rather than leaving you to find out. The $280 is this discipline applied to your store, including the writes, not a claim that either public repository has already done that for you.

2. Custom MCP server — $600

Five working days. Your own data or internal API, exposed to Claude or any MCP client, so you can ask questions of it in plain language instead of building a dashboard nobody opens.

• Read tools over whatever you already have: a store, a database, a spreadsheet, a REST API, log files.
• Write tools only where you ask for them, each one confirming what it is about to change before it changes it.
• Runs locally or on your own server. Your credentials never leave your machine, and no data passes through anything of mine.
• The same self-test-plus-mutation evidence as shop-mcp, on your server.

3. Price and stock monitor — $220

Three working days. Scheduled checks on the pages you care about, with alerts when something moves.

• Competitor or supplier pages: price, stock state, title, variants — whichever fields matter.
• Alerts by email or webhook on change, not a daily digest you stop reading after a week.
• A machine-readable history, so you can look back at what a price did rather than just what it is.
• A self-check that tells you when the scraper broke rather than reporting "no change" forever. Silent failure is the normal way these die.

4. Named-defect page-speed fix — $750

Five working days. Not "make my site faster". A fixed list of defects I have already measured on one page of yours, repaired and re-measured on that same URL. If you got an email from me, the list in it is this scope.

• What is in scope is the list, and the list exists before you pay. Server response time to first byte, HTML compression and size, redirect chains, missing viewport meta, render-blocking scripts in <head>, CSS inlined instead of cached, images declaring no width and height, below-the-fold images loading immediately, missing srcset, above-the-fold image weight, missing alt text and meta description. Whichever of those your page actually has.
• Before and after, same URL, same method. The numbers you are quoted are numbers you can reproduce yourself with curl -I and view-source before I touch anything, and again when I hand back. Take the fastest of several tries rather than one request, because a single request can hit a warm cache and tell you nothing.
• A theme diff you own, not access to a dashboard of mine. Reversible, section by section.

What this is not. Not a redesign, not a theme migration, and not a decision about which of your apps to remove — those change what the page is, and none of them can be enumerated before work starts. If your slowest problem turns out to be an app you rely on or a plan your host will not upgrade, you get told that instead of getting billed for a workaround. The measurement is of the public page as a stranger sees it, so nothing behind checkout or a login is included.

Why this one can be fixed-price at all. Because the defect list is closed before the quote, which is the same test the hourly rule below applies. "Make this faster" is open-ended and gets the hourly shape; "fix these eleven measured things on this URL" does not.

Anything else — $45 an hour

The guide on fixed price or hourly routes a job by one test: can the deliverable be enumerated before work starts. The three packages above are fixed price precisely because they pass that test. A job that does not pass it — "make this faster", "find out why this breaks", an open-ended migration — gets the hourly shape instead, because forcing an open scope into a fixed price is where disputes come from. I follow my own rule here rather than quoting you a fixed number I would have to defend later.

How to start

I am Cun Dong. A bank transfer for any of the above reads CUN DONG as the recipient, which is the name my payment provider verified against a passport before the account could receive anything — so the name on this page, the name signing the email, and the name in your banking app are one person. Email coolun337@gmail.com with what you are trying to do. Useful in a first message, though none of it is required:

• The outcome, not the implementation — "I re-key supplier stock into Shopify every morning" beats "I need a sync script".
• Which platform and where the data lives now.
• Whether anything is time-bound.

You get back a scope in writing — the deliverable, the price, the turnaround — before any money moves. If the job is a bad fit, or you would be better off buying a $20 app, I will say so; that is cheaper for both of us than a refund.

Or hire through Contra, with the money held in escrow

Email is the direct route and costs neither of us a platform fee. If you would rather not wire money to a stranger on the strength of a repository, the first package is also listed on Contra, which holds the payment until you accept the delivery: Shopify Admin API automation, $280, three days. Same scope, same price, same evidence — the escrow is the only difference, and it protects you rather than me.

FAQ