01 · Roasts
One-Shot Sprint Collector
oryzo-ai-replica (1 commit, ~1 hour), visual-lab (1 commit, <1 hour), joye-skills (4 commits in 14 minutes), zh-transcription-skill (1 commit same-day) — you have a gift for creating repos and immediately walking away. Your git log looks like a series of abandoned pilot episodes.
'eve' Is Not a Project
eve: 0 KB, 0 commits, 0 files, created and ghost-pushed on 2026-06-18. You named it, you claimed it, you left it. That repo is just a haunted namespace at this point.
143 PRs/Year But 0 Tests Across 10 Repos
You opened 143 pull requests this year — a frankly impressive number — yet only 1 of your 11 scored repos has any tests. You're aggressively contributing elsewhere while your own house is untested. CI for thee, not for me.
minimind-notes Carrying the Whole Portfolio
174 of your 234 total stars (74%) live in a single repo. Remove minimind-notes and you're at 60 stars across 51 repos. That's a portfolio with one load-bearing wall and a lot of decorative shelving.
Agent Skill Collector
You have skills/, joye-skills/, zh-transcription-skill, and memory-dashboard all orbiting AI agent tooling — but they're scattered across separate repos with no shared versioning or cross-linking. You're building an agent skill empire one disconnected SKILL.md at a time.
Built using
Zoral
Shadows one worker for a week, then takes over their job with zero extra setup. Behaves exactly like the original.
zoral.ai
02 · Category breakdown
- Impact25% weight58D
- Consistency20% weight65C
- Quality20% weight72B
- Depth15% weight65C
- Breadth10% weight72B
- Community10% weight50D
03 · Stats
365-day commit heatmap
198 active days
Language distribution
- Python31%
- JavaScript27%
- HTML23%
- TypeScript11%
- MDX3%
- Astro2%
- Other3%
04 · Numbers
Owned repos
non-fork
43
Commits
last 12 months
880
Followers
97
Joined GitHub
Sep 2023
05 · Top repos
joyehuang /
minimind-notes
Educational LLM training repository with modular teaching experiments (RMSNorm, RoPE, Attention), VitePress documentation site, learning notes system, and practical Python experiments. Well-structured but still building community adoption.
joyehuang /
blog
Astro 5 blog template with dual-language support, agent-team tournament infra, and interactive terminal shell. Fully typed, well-documented, CI/tests present. Personal portfolio project showing sustained multi-month work and structured architecture.
joyehuang /
skills
A unified collection of 20+ AI agent skills in the SKILL.md convention format, consolidating portable and personal procedural knowledge. Comprehensive README, clear architecture, proper documentation, but minimal external adoption signal (3 stars).
joyehuang /
report
Personal daily activity report generator integrating Hermes Agent and WakaTime data, with structured Python backend (4 lib modules), typed HTML rendering, and CI pipeline, but zero external visibility (0 stars/forks/watchers).
joyehuang /
memory-dashboard
Personal dashboard tracking agent memory & API usage (pi, Claude, herdr) with React 18 + TypeScript + ECharts. Typed, documented, structured, but early-stage (7 commits in ~2 hours, 0 stars).
joyehuang /
zh-transcription-skill
Specialized Claude Code skill for high-precision Chinese audio transcription using dual-engine approach (FunASR + whisper). Well-documented methodology but minimal codebase (1 script + glossary), 1 star, created and pushed same day (2026-06-22).
joyehuang /
joyehuang
Personal GitHub profile config repo with a polished README showcasing the owner's agent engineering work, projects, and social links. No substantive code, minimal project depth, but demonstrates active professional portfolio across related agent/LLM projects.
joyehuang /
joye-skills
Single-file HTML skill for terminal-style slide decks with live demo. Experimental, one-day dump (4 commits in 14 minutes), minimal scope but clear deliverable targeting Claude Code agents.
joyehuang /
visual-lab
NORTH® editorial studio site built with React, TypeScript, Vite, three.js particle galaxy, GSAP, and Framer Motion. Freshly initialized (1 commit, <1 hour old), no tests/CI, but demonstrates typed multi-component architecture with sophisticated WebGL and scroll-driven animation.
joyehuang /
oryzo-ai-replica
A single-day website replica project (21 MB HTML/assets dump) created via reverse-engineering another site. No original source code, one commit, no license, no tests. Replica-only output.
joyehuang /
eve
Empty scaffold repo with zero commits, no files, no documentation, and no development activity since creation on 2026-06-18.
06 · Timeline
- Sep 4, 2023Joined GitHub
- Nov 19, 2024Created joyehuang — Config files for my GitHub profile.
- Oct 23, 2025Created blog — joye's personal blog
- Nov 8, 2025Created minimind-notes — 🚀 [从零构建 LLM] 极简大模型训练原理与实践指南。包含 Transformer, Pretraining, SFT 核心代码与对照实验。 | A minimal, principle-first guide to understanding and building LLMs from scratch.
- Feb 13, 2026Created skills
- May 4, 2026Created report — Joye's daily agent activity reports
- Jun 18, 2026Created eve
- Jun 22, 2026Created zh-transcription-skill — 中文高精度本地转录 skill for Claude Code: FunASR + whisper 双引擎 + 正交术语正解 + 多agent还原
- Jun 22, 2026Created visual-lab — NORTH® — editorial studio site. React + TS + three.js particle galaxy + GSAP.
- Jun 22, 2026Created joye-skills — Agent skills by Joye Huang — terminal-style HTML slide decks. joyehuang.me
- Jun 29, 2026Created oryzo-ai-replica — Source-replay replica of the ORYZO AI website
- Aug 18, 2026Created memory-dashboard — Agent memory & usage dashboard (pi + herdr) — auto-updated daily
- Aug 18, 2026Most recent push to memory-dashboard
07 · Compare
08 · Rubric
How this score was produced
Overall = Σ (category × weight) + gentle top-end curve
Tier thresholds
▸ How the pipeline works
- 01Scrape.Pull every non-fork repo pushed in the last 90 days, plus your contribution calendar, followers, and language byte counts — straight from GitHub's REST & GraphQL APIs.
- 02Triage.A small model reads every repo's file tree + README and picks the 20 files per repo that actually reveal how you code.
- 03Grade each repo. All repos run in parallel through a fast scoring model that reads the picked files and rates each one independently on Impact, Quality, and Depth — with evidence citations.
- 04Aggregate. A larger reasoning model combines the per-repo scores with server-computed stats (heatmap, commit cadence, language entropy, follower count) to produce the 6-dimension profile score + roasts.
- 05Correct.Deterministic server-side checks enforce anchor-scale floors (e.g. a profile with 2,000+ public commits can't score 30 Consistency) and recompute the final verdict.
~90 seconds per profile, ~$0.25 in compute. Total of ~240 files read across your top-12 repos. One rating per GitHub account per day.
▸ Data sources & caveats
- Heatmap & commit totals: GitHub GraphQL
contributionsCollection— covers the last 365 days, includes private repos when the user has opted in (default). - Language %: byte totals across the top 30 owned non-fork repos.
- Curve: a small upward nudge centered on raw score ≈ 70, capping at 100. Prevents specialists from being unfairly penalised for narrow breadth.
- Anchor corrections: when server-measured signals (e.g. privateWorkLikely, multiRepoVolume, follower count) mandate a minimum category score, the aggregation step enforces it. These are signal-conditional, not identity-based floors.