PenguinHarness 0.2.13 advances agentic recursive self-improvement

While "agents building agents" often borders on hype, establishing sandboxed harnesses and decoupled evaluation environments is the critical missing infrastructure needed to make recursive self-improvement viable in real-world software engineering.

• –**Focus on Infrastructure over Novel Algorithms:** PenguinHarness shifts focus from prompt engineering to harness infrastructure, providing closed-loop feedback and multi-agent evaluation to mitigate reward hacking.
• –**Dramatic Cost Reduction:** By standardizing agent generation and self-evaluation, the system claims end-to-end agent construction for fractions of a cent, lowering the barrier to automated agent development.
• –**Open-Source Pedigree:** Coming from the creators of LlamaFactory, the project benefits from extensive model compatibility across 1,000+ LLMs and self-hosted, local-first deployment options.