I explore AI testing, agent tooling, and infrastructure by forking repositories, running them in real environments, and documenting what breaks. My method: small PRs, visible assumptions, and commit-level proof instead of summary promises.
Recent focus areas:
- AI-driven test automation — investigating LLM-based test generation, bug reproduction workflows, and CI/CD integration patterns across platforms like WHartTest, AITestPlatform, and qaitest.
- Agent-native software — studying CLI-Anything, MCP server implementations, and composa 8000 ble tooling that makes existing software controllable by agents without bespoke integrations.
- Infrastructure reality checks — validating Docker-based deployments, proxy configurations, and cross-platform compatibility issues on Linux and macOS environments.
Here's the thing about being a fork collector: the interesting stuff hides in the annotations. These are the ones worth opening.
| 名称 | 说明 |
|---|---|
| sketchbook | 拾光册 — a private, page-flipping photo album built as a static site. One of the few things here that isn't a fork. |
| qa-toolkit-docker | Common test-engineer tooling (Streamlit + Docker) packaged for offline deployment. |
| Horizon | AI news radar generating daily English/Chinese briefings — actively maintained. |
| WHartTest · AITestPlatform · qaitest · CLI-Anything | The AI-testing ecosystem — forked, run, and annotated. |
- Keep it small — PRs should be reviewable without context loss. One clear change beats ten unrelated edits.
- Document the why, not just the what — assumptions decay faster than code. If a decision isn't obvious six months later, it wasn't captured properly.
- Study forks like study notes — I annotate, run, break, and rebuild. Passive reading doesn't surface the hidden dependencies.
- Tools must be composable — no monoliths, no magic globals. If it can't pipe input or expose a stable interface, it's technical debt.
- Prefer clarity over cleverness — if it can't be explained in a paragraph, it's too complex for production.
Languages: Python · JavaScript · TypeScript · Shell
Platforms: Linux · macOS · Docker · Node.js · Git
Testing focus: Playwright · MCP servers · LLM-driven test generation · CI/CD pipelines
- GitHub: @Fargolee · Building in public since 2018
This profile is a workspace, not a showcase. If a fork looks abandoned, it's probably annotated locally.