Converting 10 coding harnesses including Claude Code into RL environments, cutting tool calls 31%
vllm_project · x · 2026-10-07
vLLM project highlighted Clement's open-source work: converting 10 harnesses—including Claude Code and Codex—into RL environments connected to vLLM via a proxy for flexible reward design. The same model showed gains across multiple harnesses, including a 31% reduction in tool calls. Fully open-source and reproducible.
More from coding & agent
- Creator's 10-minute AI skill-made video hits 170K views overnight — AlchainHust · 2026-10-07
- Organizer runs Tokyo dev forum with 5 AI agent workflows, caps attendance at 150 — soleio · 2026-10-07
- Consultant ditches custom voice agent build, deploys Grok bot with Cursor teams — alexcovo_eth · 2026-10-07
- Open-Source sopc: Manage Prompts Across Many Voice Agents Like Code — BadAthMOFO · 2026-10-07
- Benchmark: OpenAI Decisions API costs 2x more, 5-10% worse than Jev — xeophon · 2026-10-07
- Devs are cloning Photoshop in Rust and open-sourcing it — and AI-native tools need rethinking — ravisparikh · 2026-10-07