Developer experience engineer at LynxFlow — I write the integration code, the docs that explain it, and the workshops where people type it out themselves.
Integration examples.
Every endpoint we ship gets a runnable client in at least Python, JavaScript, and
plain curl. Not pseudocode, not a fragment in a code fence — a file you can
clone and execute. If an example goes stale, that is a bug and it gets a fix
like any other.
Developer workshops. Two-to-three hour sessions built around a working environment and a sequence of small exercises. Most of the effort goes into the setup script: a room where a third of the people cannot install dependencies has already lost its first hour.
Documentation. Reference material generated from OpenAPI descriptions, and the hand-written guides around it — quickstarts, error catalogs, migration notes. I follow the Diátaxis split (tutorial / how-to / reference / explanation) because mixing those four modes in one page is how docs become unreadable.
Error messages and status codes. An unhelpful 400 costs more support time than a missing page of docs. I spend a surprising amount of my week arguing about response bodies.
api-examples
The same REST workflow — list, fetch, create — implemented three times: Python
with requests, Node with the built-in fetch, and a POSIX shell script over
curl. Runs against a public demo API, so there is no key to obtain first.
workshop-kit The template I start every workshop from: an agenda with realistic timings, a preflight script that verifies the room's toolchain before anyone sits down, and three exercise stubs that build on each other.
bookmarks A reading list I actually maintain — the specs, books, and tools I keep sending to people who ask where to start with API design, DX, or technical writing.
changelog-template
A Keep a Changelog template plus a Python script that reads git log and drafts
a changelog entry from conventional-commit messages. The first pass is mechanical;
the writing that matters is still yours.
sdk-snippets The same API call — create an item — written in curl, Python stdlib, and Node fetch. No dependencies, no pseudocode. Clone and run. If a snippet stops working, that is a bug, same as any other code.
- Examples are tested, or they are deleted. A broken sample teaches the wrong thing twice: once about the API, once about whether we can be trusted.
- I prefer the standard library where it is good enough. Node has
fetchnow; most quickstarts do not need a dependency tree. - Workshop exercises get written backwards, from the thing I want people to be able to do at the end.
- Docs bugs are filed in the same tracker as code bugs, with the same triage.
Reachable through GitHub — issues and discussions on any of the repositories above, or the profile inbox. That is the only channel I check reliably, so it is the only one listed here.
Opinions here are my own. Code in these repositories is MIT licensed unless a repository states otherwise.
My daily briefing: t.me/Lx_groups (Chinese, source links on every item). Check the archive without Telegram: t.me/s/Lx_groups. Where I send people for depth: blog.lynxflow.co · the channel's landing page.