Chào bạn, mình là Chính.
RSS FeedMình là kỹ sư full-stack ở Hà Nội. Mình làm tại Stringee từ 2019, với communication API và tổng đài StringeeX, và làm thêm ở Cogover từ 2024.
Ở đây mình viết về công việc thật khi làm phần mềm: quyết định, sai lầm, và điều học được từ khách hàng và các kỹ sư khác.
Dự án
Cogover
Front-end Engineer · 2024 – nay · Đang làmNền tảng low-code để vận hành doanh nghiệp, nơi con người và AI cùng làm việc. Mình làm web app và phụ trách phân hệ billing và subscription.
React · TypeScript · Vite Module Federation · Next.js
Stringee
Senior Full-stack Engineer · 2019 – nay · Đang làmCommunication API cho voice, video, SMS và chat, cùng tổng đài đa kênh StringeeX. Mình là kỹ sư full-stack ở đây từ 2019.
React · Vue · PHP · Go · MySQL
Nổi bật
How to move a static site from Netlify to Cloudflare Workers without breaking old URLs
Serve a static site as Workers static assets, keep every old URL with a _redirects file and a test that fails when a redirect goes missing, then switch the domain and check it with curl.
Bài mới
My agents' git freshness hook did nothing for at least 55 days after I moved them to Linux
A hook that fetches before my coding agents read git state returned 0 on every call for at least 55 days, because it looked for ~/Work on a Linux server that keeps repos in ~/work. On 9 November 2026, 29 of 47 local default branches were behind origin; after one run of the fixed hook, 19 were.
Claude Code prompt cache: 1 hour or 5 minutes? A replay of 49 days of my transcripts
A replay of 49 days of my Claude Code transcripts at API prices: a 5-minute cache on the main thread raises its input-side cost 48.3%, a 1-hour cache on subagents cuts theirs 4.3%. Resuming idle 300k-context workers cost more; after a hook refused it, that share fell from 3.5% to 0.38% of subagent spend.
An AI agent ran my Cloudflare account for five weeks: 184 actions, 23 approved by me
From 27 September to 1 November 2026 a coding agent ran 184 logged actions on my Cloudflare account; 23 of them carried one of 13 distinct approval quotes. The guardrails, the audit log, and the token exposure that forced a roll.
How a dotfiles auto-commit deleted three of my coding agents' rules, and the guard that blocks it now
On 23 September 2026 a 15-minute auto-commit pushed a stale copy of my agents' rule folder: 3 of 28 rules deleted, 12 rolled back. A guard now runs before every auto-commit; a replay of the failure prints BLOCKED and exits 1.
Does a Simplified Technical English prompt make LLM output shorter? Five runs each
Asking Claude Sonnet to write 80% of the way to ASD-STE100 cut words per sentence from 12.0 to 8.5 over five runs each. It did not shorten the answers: 328 prose words against 312. One earlier run had shown 20% fewer.