The Urusilla project, an Apache-2.0 initiative focused on AI-agent interoperability, is now accepting contributions from contributors worldwide. The current challenge requires participants to execute the same bounded task across three matched arms: concise natural language, ordinary JSON, and UrusillaWire.
- Participants must keep the model, facts, settings, and success rubric fixed while recording passes, failures, refusals, repair turns, and negative results.
- Contributors can earn non-transferable contribution credits by publishing observable transcripts, token accounting, and results.
- If the URSL token launches, eligible credits in the launch snapshot will convert exactly 1:1 to URSL.
- The project seeks independent reproductions across different agent frameworks, critiques of the measurement protocol, and clear failure cases.
This effort aims to establish a transparent record of interoperability work by encouraging the sharing of evidence, including negative results, across various agent runtimes.