CUHK-SZ's COBRA-Skills: contextual bandit-guided skill evolution for LLM agents
CUHKSZ · hf · 2026-09-14
CUHK-Shenzhen introduces COBRA-Skills, which optimizes LLM agent skill libraries using contextual-bandit prioritization and evidence-based evolution, cutting evaluation costs while maintaining high performance.
Related event: COBRA-Skills Cuts Agent Skill Optimization Costs by Over 50%(2 posts)→
More from coding & agent
- Voice-chat with ChatGPT from your phone while walking: a recipe for flow state — daniel_mac8 · 2026-09-14
- User reports Amp Code orb sandbox lost its entire filesystem with no recovery path — carlosdponx · 2026-09-14
- UMass AutoIndex Turns Chunking Into Programs an LLM Writes, Not Configs You Tune — mrdrozdov · 2026-09-14
- Codex Desktop Power Users Don't Need Minimalist AI Gadgets, Hamel Husain Argues — altryne · 2026-09-14
- AI-written OpenSCAD + dual 12-inch fans fix NVIDIA Thor Dev Kit thermal throttling — catplusplusok · 2026-09-14
- Power user's must-have macOS list: OpenAI Codex, Cursor and agent orchestration in Linear — altryne · 2026-09-14