Xinwei Xiong (熊鑫伟, aka cubxxw / BEAR, b. 2001) builds AI products with a focus on agent evaluation and reliable execution. He contributed to OpenIM's engineering and community workflows, and writes about projects and monthly reflections at cubxxw.com. Travel remains part of his life.
2001 born · Wuhan CS student · 2021 first blog/GitHub · 2022 learning-note open source · 2023 OpenIM internship & this blog · 2024 graduation, first trip abroad, quit corporate, nomad · 2025 treks & inner work · 2026 Tibet/AI identity, solo AI products, new job Aug 10.
X/Twitter @cubxxw account created (public digital footprint begins on that platform)
Late high school — no reliable public digital traces
Entered university (CS, class of 2020, Wuhan) — sparse public traces
Joined CSDN
GitHub account cubxxw created; first repo nsddd (campus parking mini-program)
First blog post on nsddd.top: “hello! 欢迎来到我的博客”
Note-taking open-source burst: 菜鸟成长手册 (awesome-cs-cloudnative-blockchain, later 1.4k stars), LeetCode, myblog
Juejin profile; Bilibili 程序员知识分享计划 videos
First commit to labring/sealos
OpenIM first internship; #1 contributor leaderboard; designed & hosted community biweekly (双周会); projects: sealer, horizon, kubernetes, k8sgpt, openkf
Current Hugo blog repo + 《我的第一个博客》 — “你的价值取决于你创造了什么，而不是你拥有什么”
2023 annual summary; NYE Victoria Harbour self-marking: loyalty to adventure & open source
First time abroad — Kuala Lumpur (second semester of senior year)
Graduation → Japanese/foreign company (Shanghai) → left the job for travel & building; RemoteAI-Nomads guide (74 stars)
Digital nomad year: Jeju, Qingdao, Vietnam, Mt. Fuji, Kansai; monthly thought notes; metacognition review “自我解释不等于自我理解”
Sichuan–Tibet road (G318) → Lhasa residency ~1 month; AI identity essay “如果 AI 能代替我思考，那‘我’在做什么？”
2025–2026 annual review 《我在游荡，但我没有迷路》; Cambodia Angkor field notes
Laos & Cambodia border month; disconnection experiment in May
GEO series + solo AI products including DayPage and Solo Compass; 55 Chinese posts / 733 GitHub contributions that month
Started a new job after a ~2-year career gap: “不会让恐惧替我做决定”
IMStage live; StripSearch research agent; still writing monthly thought notes
Published paired Personal Agent product and technical studies, separating source-code observations from proposed task and authorization models
Public monthly notes connect user stories to evaluation cases, failure analysis and regression; research methods remain distinct from measured production outcomes
I created my GitHub account and began publishing my notes.More of the story
In June 2021 I created my GitHub account cubxxw, with a campus parking mini-program (nsddd) as its first repository. On November 15 I published the first post on nsddd.top, “hello!”, and started writing in public.
map[I registered on CSDN that January and began posting there in November. Writing and open source were the same act from the start:turning my own notes into something other people could open and reuse.]
I joined OpenIM as an intern, contributed code, and hosted community meetings.More of the story
In 2023 I joined OpenIM as an intern, reached the top of its contributor leaderboard, and designed and hosted the community biweekly. On September 14 this blog opened, and I wrote: “Your value is what you create, not what you own.”
That year I also contributed to sealer, horizon, kubernetes, k8sgpt, and openkf. On December 31 I spent New Year's Eve at Victoria Harbour in Hong Kong, marking my loyalty to adventure and open source.
My first trip abroad, graduation, a job, then leaving to travel and build.More of the story
On February 25, 2024, in my final university semester, I flew to Kuala Lumpur, my first trip abroad. After graduation I joined a Japanese foreign company, then left that job the same year to travel and build.
After leaving the company I compiled the RemoteAI-Nomads guide and began a life of travelling while building. Graduation, the first flight, and the resignation were all decisions I made myself that year.
My first high-altitude trek. From February, I began writing monthly notes.More of the story
In January 2025 I finished my first high-altitude trek, the Annapurna Circuit in Nepal, and cried twice: “I didn't cry from the pain; I cried because I was still walking.” From February I began monthly thought notes, a standing review of my own judgment.
That year I moved between Jeju, Qingdao, Vietnam, Mount Fuji, and Kansai, keeping the monthly notes and metacognition reviews as fixed practice. “Self-explanation is not self-understanding” was the conclusion I kept.
I kept building and writing. On August 10, after roughly two years away, I started a new job.More of the story
Through 2026 I independently built AI products including DayPage and Solo Compass, and wrote about identity in the AI age. On August 10, after roughly a two-year career gap, I started a new job. My August thought notes follow the ongoing work on agent harnesses, products, and self-knowledge.
In February I took the Sichuan–Tibet road to Lhasa and stayed about a month; in March I finished the annual review “Wandering & Growing”. Products, writing, and the new job all continue.
In September and October I focused on agent outcomes. What does the user want to accomplish, what counts as success, and how do we diagnose a failure? Those questions began connecting to evaluation and regression cases.
I also continued working on DayPage, StripSearch and agent-kit, and studying how personal assistants preserve task progress, handle authorization and recover execution. In public records I distinguish implemented code from proposals that still need validation.
CUBXXW / READING PATHSWant to read my writing?Open four curated reading paths
Now that you know the person, meet the writing. Everything on this blog follows four threads: AI agents, engineering & open source, growth & thinking, and travel & observation. Not sure where to begin? Here is the map drawn for your first visit.
cubxxw.com is Xinwei Xiong's open archive of practice and thought. Its writing grows from real projects, engineering choices, and lived experience, covering AI workflows, agent engineering, open-source practice, systems thinking, personal growth, philosophy, and life. Xinwei has contributed to projects including OpenIM, OpenKF, and Sealos, and openly documents the methods, failures, and judgment formed through practice.
Author
Xinwei Xiong (cubxxw)
Identity
Practitioner · Writer · Open-source contributor
Threads
AI systems / Engineering methods / Real projects / Growth & philosophy
Footprint
9 countries · 35+ cities · 400+ km hiked
Product lab
Products are where ideas meet reality.
A public record of the tools I build, what they test, and where the evidence still runs out.
An explorable map of products and writing. Click a node to open real posts and projects without breaking reading flow.
Current test
Talent Signal
gettalentsignal.comIn testing
In testing
Know who needs you now, and why.
Relationship intelligence for independent recruiters, turning meaningful conversations into reviewable momentum, source-linked evidence, and one clear next move.
What this cycle tests
Can source-linked candidate momentum and one smallest next move reduce reconstruction between conversations while preserving recruiter judgment?
BoundaryThe current foundation uses synthetic data to test product language, review states, and the cross-platform loop. It is not yet a production candidate-data system.
Each product keeps its current state, live path, and source visible. No finished-product theater.
LiveDeep readers · Original-language reading and language learning
Leafhour
Stay with a book.
A personal reading room with AI by your side. Resume where you left off, move between the original, translations and your own notes, and open a reading conversation when a passage needs more attention.
Start with a premise. Generate dialogue, people and photos, edit each detail, reuse a template, or make a series of stories in one project. Export the result as a PNG.
Open the map. Find what is truly worth doing nearby.
A living map for solo travelers that turns nearby places into concrete, story-rich experiences and uses AI to narrow the field without making the decision.
See which agent crossed the line and which project is burning tokens.
A local agent-governance tool that gates risky changes before push, deduplicates session usage by message ID, and ranks projects and sessions by token use and estimated cost.
