Skip to content
View DevCrazyCube's full-sized avatar
👾
Thinking in ultracode
👾
Thinking in ultracode

Block or report DevCrazyCube

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
DevCrazyCube/README.md
Lines of code, commits, public repos, followers and stars of this account's public repos, with the share of each language, recounted every week Joyren Terken. I build by directing AI coding agents, and I check what they make. Beside the words, my StudySync character holds up a magnifying glass.



studysync.nl · LinkedIn · @dev.joyren · @studysyncnl



I'm building StudySync: it reads your course in Moodle or Canvas and turns it into flashcards, quizzes and cheatsheets, week by week. It is pre-launch.

I specialize in agent pipelines and orchestration: getting AI coding agents to build, review and ship real software, and knowing when their output can be trusted.

Almost everything here was built by directing AI coding agents, mostly Claude Code. I don't hand-write most of the code anymore. My part is the direction and the checking:

  • Orchestration: splitting work across parallel agents in separate worktrees, with one job per agent and a clear brief for each.
  • Review pipelines: multi-agent audits and critique rounds , where separate agents verify findings before anything is changed.
  • Guardrails: project rules, skills and memory files that agents read before they work, plus tests that fail when a rule is broken, such as an untrue marketing claim or a leaked answer key.
  • Long-running jobs: render and build pipelines on rented GPUs that resume after a failure, shut themselves down and log their cost.
  • Verification: type checks, tests and my own eyes on the result before it ships.

Every README says what was run and what was not, so you can tell what is verified from what is only built.



StudySync

Four StudySync screens on a phone: a week's path, a missed answer with the right one shown, a flashcard and an opened chest

The product I'm building. A student connects Moodle or Canvas once, and every week of the course becomes a short path of practice: flashcards with spaced repetition, a quiz, a cheatsheet, and a class quiz on the big screen. Pre-launch, so no customers yet. The source is private; the explainer with screenshots of the current app is public.

StudySync Live: the class quiz lobby on the big screen with a join code and the players as characters, and two phones: You are Steady Heron, and Correct, plus 846

Live, the class quiz: the teacher's screen shows the question, phones are the buzzers, and every player gets a made-up name and character. The picture shows the new Live design, which I'm building now.

Claude skills

Seven small skills for Claude Code. Each is one folder: the instructions, a script for the part a machine can check, and tests for that script.

# install one: clone it into your skills folder
git clone https://github.com/DevCrazyCube/claude-skill-cold-read ~/.claude/skills/cold-read
Skill Use it when Checked by
cold-read You wrote a README, pitch or brief and can no longer read it as a stranger. A fresh agent gets only what the audience gets and says what it understood Two example reads in the repo
evidence-first-research A research log must not turn guesses into findings. Every claim carries a label: measured, sourced or assumed A linter, 28 tests
agent-progress-log Several agents and people work in one repo and the progress file keeps getting overwritten. One fragment per piece of work, one generated log 9 tests
film-ear-eye A film is "done" and nobody measured it. Loudness and silences; cuts, camera speed and holds A self-test on a generated film, 17 checks
art-director-round The numbers pass and it still looks unfinished. A separate critic judges stills of every shot, full size and phone width 11 tests, one example round
short-reel-method A short comedy reel needs a point. One truth, one turn, and a one-line pitch that is checked before anything is drawn 25 tests
comfyui-job-pipeline A frontend talks to ComfyUI node ids directly. This puts ComfyUI behind one boundary 51 tests against a mock server, not against ComfyUI


Projects

The mc-trend-analysis operator dashboard
mc-trend-analysis
Links new token launches to news and search narratives and scores potential against risk. It does not trade. The picture shows the key-free demo on made-up data.
VitalMap: a world map coloured by births in the year 2000
VitalMap
World births and deaths, 1950 to 2100, on an interactive D3 map. Flask, D3, GSAP.
The Piper pipeline dashboard
piper
A data pipeline with encryption and audit built in. A four-person school project; the security pillar is my part. The picture shows two public test datasets.
An animated Sokoban solution of 69 moves
dynamic-sokoban-pathfinding
A Sokoban solver in pure Python: BFS over moves, A* over pushes. The animation is the solver's own SVG export.
Also What it is
trading-research-lab Evidence-first trading research: a measurement harness and forward paper tests. Not a trading bot, and nothing in it is shown to make money after costs
Flappybird-Reinforcement-learning A PPO agent that learned Flappy Bird from five numbers. The game underneath is someone else's and is credited there

Pinned Loading

  1. studysync-showcase studysync-showcase Public

    What StudySync is, in pictures: your Moodle or Canvas course turned into flashcards, quizzes and a week path. Explainer only, no source code.