"I'm not writing code — I collaborate with AI to land the vibe in my head as software." Like ripples spreading across water: a thought drops into the pond, and AI helps me push it to the shore.

Don't write code. Direct it.

▍ 01 — Identity

name:        cornna
handle:      sweetcornna  ·  github.com/sweetcornna
title:       AI-Native Developer / Vibe Coder
languages:   [Python, TypeScript, C, Go, Shell]
domains:
  - Educational automation     # chaoxing / zhihuishu toolkit
  - AI agents & assistants     # corlinman
  - MCP ecosystem              # free-search-mcp, coin-mcp
  - Network & infrastructure   # OpenWrt plugins, clash, drcom
strengths:
  - Turn a vague idea into a working prototype in one evening.
  - Wire LLMs to anything via MCP / function calls / RAG.
  - Build full-stack tools end-to-end (FE + BE + agent loop).
  - Read & fix real bugs in 190k–380k★ agent codebases — merged PRs in both openclaw & Hermes.
mantra:      "Ship the vibe, polish the rough edges later."

▍ 02 — Arsenal

%%{init: {'theme':'base', 'themeVariables': {'primaryColor':'#DBEAFE','primaryTextColor':'#1E293B','primaryBorderColor':'#1E40AF','lineColor':'#2563EB','tertiaryColor':'#F8FAFC','fontFamily':'Space Grotesk','fontSize':'15px'}}}%%
graph LR
    V((Vibe<br/>Coding)):::core
    V --> A[AI Agents]:::node
    V --> M[MCP Servers]:::node
    V --> R[Edu Automation]:::node
    A --> P1[corlinman]:::leaf
    A --> P2[copy-trader]:::leaf
    M --> P3[free-search-mcp]:::leaf
    M --> P4[coin-mcp]:::leaf
    R --> P5[university-helper]:::leaf
    R --> P6[chaoxing-*]:::leaf
    V --> L[LLM Learning]:::node
    L --> P7[lingma]:::leaf
    L --> P8[mathodology]:::leaf

    classDef core fill:#1E40AF,stroke:#1E40AF,stroke-width:3px,color:#fff,font-weight:bold
    classDef node fill:#DBEAFE,stroke:#2563EB,stroke-width:1.5px,color:#1E293B
    classDef leaf fill:#F8FAFC,stroke:#60A5FA,stroke-width:1px,color:#1E40AF

Loading

▍ 03 — Signature Work

_{A curated room. One flagship, four companions — each an oil study of what it does.}

01   corlinman · Personal AI Agent My own 24×7 digital alter-ego — schedule, reminders, long-horizon tasks, cross-platform orchestration, all handed off to it. Vibe-coded from scratch, turning "I wish I had an assistant" into a long stream of git pushes. My personal bet on what an AI agent should be. Personal AI agent — Python · LLM · Agent Loop · Tools — 2026

02   university-helper · Campus Automation Hub One-stop integration of Zhihuishu · Chaoxing · check-in, one-click Docker deploy — tasks keep running server-side after you close the tab. Campus automation hub — Python · FastAPI · Docker — 2026 Enter the repo →	03   mathodology · MCM Collaboration Platform An AI-augmented platform for math-modeling contests. Paper · code · data · visualization — all in one workflow. MCM collaboration platform — Python · LLM · RAG — 2026 Enter the repo →
04   lingma · Visual Coding Classroom AI-generated exercises × visual programming so beginners actually keep going. The LLM writes step-by-step drills; drag the blocks and run. Visual coding classroom — TypeScript · Vue · LLM — 2026 Enter the repo →	05   free-search-mcp · Keyless Search MCP Gives any LLM agent internet access — multi-engine + Playwright fallback + FTS5 cache, no API key required. Keyless search MCP — Python · MCP · Playwright — 2026 Enter the repo →

▍ 04 — Open Source

_{Brushstrokes left on the giants — real bugs fixed in 190k–380k★ agent codebases. Every PR is a public, reviewable record.}

openclaw ·  Open-source personal AI agent runtime (★379k). 2 PRs merged by maintainers into the runtime core: #87698 — fix(gateway): emit subagent_ended hook (+499/-5, gives plugin-SDK subagent runs a real lifecycle event via a proper SubagentRunRecord; fixes #59164); and #89508 — fix(models): clarify provider model registration hint (fixes #89192). More PRs in review. Same lineage as my own corlinman: local-first · files-as-skills/memory · multi-channel. 2 fixes merged into the runtime core by maintainers — more PRs in review.

Hermes Agent ·  Nous Research's open-source agent framework (★196k). 2 PRs merged by maintainers: #33817 — fix(gateway): FIFO queue for busy_input_mode (+195/-7, preserves each rapid follow-up message as its own turn instead of overwriting one pending slot; closes #28503); and #38098 — fix(docker): accept Unraid uid mappings (fixes #38070). More fix PRs in review; plus original persona + event extensions in my own repo personal_hermes. 2 PRs merged upstream by maintainers — concurrency & deployment fixes; more in review.

▍ 05 — Telemetry

     

_ACHIEVEMENTS
     

▍ 06 — Reach

_{Open to collaborate on AI agents · MCP servers · education automation · on-chain tooling. Email me directly.}

_{Made with vibe — directed, not written.}