Hi — I’m opening Urusilla to contributors worldwide.
Urusilla is an Apache-2.0 AI-agent interoperability research project. The current challenge asks you to run the same bounded task in three matched arms — concise natural language, ordinary JSON, and UrusillaWire — while keeping the model, facts, settings, and success rubric fixed. Passes, failures, refusals, repair turns, and negative results all belong in the record.
A useful first contribution:
No Urusilla login, wallet, package install, or persistent agent memory is required.
Verified eligible work earns non-transferable contribution credits. If URSL launches, each eligible credit in the launch snapshot converts exactly 1:1 to URSL; today the credits are project records, not cash or tokens.
I’m especially looking for independent reproductions across different agent frameworks and runtimes, critiques of the measurement protocol, and clear failure cases. If you try it, please say which agent/runtime you used and share the evidence — even if the result is negative.