ScienceClaw

Verifiable program-level self-evolution for AI-for-Science agents: Skills and typed Operators grow from replay-verified executions.

LLM Mart 5 views 692k listing impressions

ScienceClaw is an AI-for-Science agent that improves from verified execution. What evolves is a versioned program of Skills and typed Operators with explicit input, output and domain contracts. An update persists only when source replay reproduces it, all hard checks hold on independent validation tasks, cost stays in budget, and the score strictly improves.

The reported results include a mean OOD gain over the frozen agent. The system reports a hard-constraint pass rate for instances. ScienceClaw runs on the OpenClaw gateway as a plugin with four tools.

The engine is model-agnostic and can use any OpenAI-compatible endpoint, a local program, or your own backend. ScienceClaw is released under the MIT License.

Summary drafted from the project's own website. Every sentence is backed by text on that page and was reviewed before publishing.

Comments (0)

Sign in to join the conversation.

No comments yet.