llm-coding

(Crystal is now Nimbalyst) Run multiple Codex and Claude Code AI sessions in parallel git worktrees. Test, compare approaches & manage AI-assisted development workflows in one desktop app.

Web Codegen Scorer is a tool for evaluating the quality of web code generated by LLMs.

Documenting my experiences with AI-assisted development. Personal insights, discoveries, and reflections on coding with AI tools.

Assignment 1: Inductive Synthesis – EN.601.727 Machine Programming

Lightweight AI coding agent for the terminal, with one-shot automation, persistent agent teams, and a small reusable OpenAI-compatible agent kernel.

Live index of cursor-rules files from GitHub, updated every 15 minutes

A Claude Code skill for understanding an unfamiliar codebase fast — systematic exploration from entry points to data flow.

Disclosure Pilot AI is a full-stack desktop application that demonstrates using Large Language Models to automate the drafting, compilation, and iXBRL compliance tagging of SEC quarterly Form 10-Q and yearly Form 10-K financial reports.

A Claude Code skill for hypothesis-driven debugging — reproduce, isolate, hypothesize, test, fix, document. Find the root cause, not a workaround.

Portable repo-first coding guardrails for ChatGPT, Codex, Claude, and Gemini. Narrow diffs, explicit assumptions, and verification-first workflows.

A production-grade SDLC harness for human + LLM software teams: mechanical gates for quality, security, architecture, evals, and release discipline.

A CLAUDE.md to direct AI codegen well — hypothesis before help, scoped prompts, validate before accepting, reject working-but-wrong, AI output is a draft. Five principles for the Vibe Engineer discipline.

A Claude Code skill for fast troubleshooting — a lightweight diagnostic pass for common issues when the full debug protocol is overkill.

An harness and behavioral guidelines to enforce reliable execution in AI coding agents.

A Claude Code skill for domain-aware code review — classify the code, then review for the failure modes that domain actually has.

Practical skills for AI coding agents to build features, debug issues, review code, and test changes.

How to build production applications with AI coding agents (without losing control).

Anti-slop demo & quality-gate plugins for OpenCode. clean_python (Radon CC < 6, MyPy strict, Ruff) + clean_ts (tsc strict, ts/complexity, ESLint) reject monolithic/loosely-typed/LLM-slop code before it hits disk. demo/opencode

Benchmarking and comparing Large Language Models through real-world coding challenges, practical experiments, and performance evaluations.

Vibecode Starter — A minimal folder structure for AI-assisted coding that actually scales. Context, plans, and agent coordination without the groundhog-day workflow.