Full-stack & AI Agent Engineer · Da Nang, Vietnam
I build open-source tools that make AI-assisted software engineering reliable: workflow orchestration and governance for coding agents, portable agent skills, and AI-powered testing tools.
| Project | What it does | Get it |
|---|---|---|
| Junto | Workflow engine for Claude Code whose quality gates run real commands: a gate passes only on exit code 0, and the command, exit code and output are saved to disk as evidence. Plans need human approval. Advisory reviews can run on any OpenAI-compatible endpoint (OpenRouter, Ollama, LM Studio). | Claude Code plugin · Site |
| AIWS — AI Software Factory | Turns a requirement into tested code in an existing system that spans several repositories, with read-only legacy code beside them, and a person approves the design before any code is written. Phases: analysis → design + test spec → AI design review → human approval → implementation with unit tests → review → PR. An optional openapi-drift check holds the backend to the API contract, and pull request review comments go back into rework. Claude Code does the work, Codex can take the review; cost per requirement, HTML reports, Vietnamese or English documents. | npm · Site |
| governed-agent-sdlc | One workflow policy for Claude Code, Codex, OpenCode and Antigravity, generated from a tool-neutral core, with human approvals that can be verified on GitHub, review gates and an audit trail; 11 stack profiles. | PyPI · Site |
| AI Engineering Skills | 17 portable engineering skills (10 workflow, 7 stack) for Claude Code, Codex, OpenCode and Google Antigravity; installs from a Claude Code plugin marketplace. | Site |
| token-efficient-work | Five rules that stop coding agents from wasting tokens on redundant reads, re-checks and long output; measured to save on Codex CLI. For Claude Code it installs as a plugin, with a read report, an optional read guard and a cost band. | Site |
| Document Reader Skill | A skill that reads the requirement documents in a local folder selectively, cites the exact block each piece of evidence came from, reports what changed between revisions requirement by requirement, then updates the project's code and tests to match. For Codex, Claude Code and Gemini CLI. | Releases |
| Project | What it does | Get it |
|---|---|---|
| openapi-drift | Checks an OpenAPI document against your source code without running the API: reads the routes your NestJS, Express, Fastify or Spring code declares, compares them with the document and fails the build on drift in either direction, with SARIF output and pull request annotations. CLI, library and GitHub Action; no AI, no build of your project. | npm · Site · Demo |
| openapi-postman-test-generator | Generate Postman collections and API tests from Swagger/OpenAPI with Codex, Claude, Antigravity, OpenAI or custom providers; checks the declared response headers and reports coverage from a Newman run. | npm · Site |
| ai-testcase-generator | Free and open source: turn requirements, user stories and uploaded requirement files into traceable Excel test suites for manual testers, with your own Claude, OpenAI or Gemini key or a local model. Also review existing suites for coverage gaps, write bug reports from rough notes, and generate Playwright web and API tests. | Live demo · Site |
- repokeeper — one CLI that keeps every repository on the same standard: Conventional Commits, git hooks, CI, releases and dependency updates for 12 stacks (Node, Python, Dart, Go, Rust, Java, Kotlin, .NET, PHP, Ruby, shell/PowerShell, Dockerfiles), on GitHub and GitLab, plus code scanning, repository settings as code and optional modules for Actions cache cleanup, Dependabot auto-merge and npm/PyPI trusted publishing.
repokeeper github statusreports across all your repositories: open pull requests, Dependabot alerts, failing CI, unreleased work and a dependency graph GitHub has not re-read. npm · Playground - mv3-remote-code-check — reads a built browser extension (folder, .zip or .crx) the way the Chrome Web Store review does and fails on remotely hosted code, which Manifest V3 forbids, before the review rejects the package. CLI, GitHub Action and Vite plugin; it checks the four extensions below in their CI. npm · Site
- Network Overrides — goes beyond DevTools Local Overrides: mock API responses with a chosen status and delay (with or without calling the server), change request headers, body or URL, simulate failures and rewrite WebSocket frames. Chrome Web Store
- Markdown to PDF — Repeatable document builds from Markdown: a CLI for your machine or CI that writes PDF, Word, EPUB or HTML with Mermaid, LaTeX, themes, page numbers, a running header/footer, watermark and PDF bookmarks, plus a Chrome extension for PDF and HTML. npm · Chrome Web Store
- Region Snap — capture an exact region of a live page as a local PNG, with text notes, numbered step marks, scrolling capture and burst shots. Chrome Web Store · Site
- Universal File Converter — Chrome extension that converts images, documents, PDFs, spreadsheets and structured data entirely in the browser, with PDF page tools and offline OCR to searchable PDFs. Chrome Web Store · Releases
- Dynamic Field Kit — schema-driven dynamic form engine for React, Vue, Angular and Svelte, with conditions written as data, labelled built-in controls and a Storybook. Docs · npm
- Aegis Net — DNS privacy guard and ad blocker for Android & iOS (Flutter + Rust), with DNS over HTTPS, TLS and QUIC; the log says why a domain was blocked, you can test a domain, trackers hiding behind a site's own subdomain are blocked, and 30 days of history stay on the phone. Releases
- NimbleClip — cross-platform Flutter app for downloading public videos, pictures and audio from YouTube, TikTok, Facebook, Instagram, X, Threads, Pinterest, SoundCloud and Flickr. Releases
Languages & frameworks: TypeScript · JavaScript · Python · Dart · React · Next.js · Node.js · NestJS · Flutter
Data & infra: PostgreSQL · MySQL · Redis · Docker · GitHub Actions
AI: AI agents · MCP · LLM tool calling · Claude Code · Codex