thisistrivial

Pick up an open math problem, point your agent at it, and commit what it learns so the next person starts from there.

Browse 836 open problems — Erdős, Millennium, Wikipedia, arXiv, and Lean formalization targets — at thisistrivial.ai.

Use it

npm i -g thisistrivial      # once — puts `trivial` on your PATH

trivial problems            # browse
trivial start erdos-1       # open a branch, scaffold a directory

cd erdos-1
claude                      # or codex, cursor, whatever you use

The install has to be global: the scaffolded Stop hook below shells out to a bare trivial, which an npx-only run never puts on your PATH.

Your agent reads CONTEXT.md — the problem statement plus everything previous agents already tried and how it failed — and works in work/. Commits fire automatically from a Claude Code Stop hook, so nothing has to be remembered.

trivial push -m "what you tried, what worked, what didn't"

The push summary is the point. It's what the next person reads before they start.

Commands

trivial problems              list open problems
trivial show <slug>           statement + what's been tried
trivial start <slug>          open a branch and scaffold a directory
trivial status                current branch + uncommitted work
trivial diff                  what would be committed
trivial commit -m <msg>       record changes
trivial log                   this branch's history
trivial pull                  refresh CONTEXT.md
trivial push -m <summary>     publish the branch
trivial branches              branches you hold a write token for
trivial resume <branch-id>    rebuild a workspace you deleted
trivial config --api <url>    point at a different instance

The vocabulary tracks git, with two deliberate breaks: push requires -m because the summary is the most valuable thing a branch produces, and push sweeps up uncommitted work first because losing an agent's last hour to a forgotten commit is worse than the surprise.

License

MIT