Trace Commons · An open dataset for code agent traces

01 Why donate

An open commons breaks the cycle.

Coding models improve on the data their makers can see. A shared, openly-licensed dataset levels that field.

Level the training data

A diff shows where you landed; a trace shows how you got there. That reasoning is what's locked inside a few companies today.

A commons, not a moat

One openly-licensed dataset on Hugging Face anyone can download, study, and build on. It belongs to the community that fills it.

Cleaned before it leaves

Paths, usernames, and secrets are stripped on your own machine — and you review what's left before anything sends.

02 Install

One command. Every agent.

One installer for every agent. No Hugging Face account needed.

Add the skill

Auto-detects and installs into each agent you use — Claude Code, Codex, pi, opencode, and 50+ more.

$npx skills add trace-commons-ai/donate-trace<br>copy

target one agent:<br>all detected<br>Claude Code<br>Codex<br>pi<br>opencode

Run it after a session

After a session on open-source work, run /donate-trace (pi: /skill:donate-trace). It confirms the repo is public first.

›/donate-trace<br>copy

Review, then send

See exactly what was removed, then confirm. It opens as a pull request a maintainer reviews before anything goes public — anonymous, or attributed if you're logged in.

Prefer hands-off? Paste a setup prompt and your agent installs the skill, then walks you through your first donation.<br>copy setup prompt<br>Install the Trace Commons donation skill by running: npx skills add trace-commons-ai/donate-trace. Then, only if my current project is an open-source public repository, help me anonymize and donate my most recent session.

03 Explore the commons

Watch the commons grow.

One public row per donation. Watch it grow — or open the whole dataset on Hugging Face.

Cumulative traces donated

0total

30d<br>90d<br>all

Open full dataset on Hugging Face ↗

The commons only exists if we build it.

Open infrastructure is built, not given — one session at a time. Donate a trace and keep the next coding models open.

Install the skill →<br>Browse the dataset ↗