Tuesday, February 17, 2026

SkillsBench. New benchmark evaluates how well agent skills transfer across diverse tasks. arXiv

Agent Evaluation Study. Research examines whether AGENTS.md documentation actually helps coding agents perform better. arXiv

Long Context Focus Gap. Study reveals LLMs lose focus in longer contexts, impacting privacy and personalization capabilities. arXiv

Boundary Point Jailbreaking. New research explores vulnerabilities in black-box LLM security through boundary point attacks. arXiv

Jemini. Gemini-powered search interface specifically designed for exploring the Epstein Files. Jmail

Claude DevTools. Local log viewer that makes Claude Code's CLI output more readable and useful. GitHub

Forestry Diary Archive. Digitized collection of daily USFS work diaries spanning 1927-1945. Forestry Diary

Open Source Model Dominance. Four of the top five most-used models on OpenRouter this week are open source. Reddit

FunctionGemma Fine-Tuning. Fine-tuned 270M parameter model achieves 90-97% accuracy on multi-turn tool calling tasks. Reddit

Google's Intelligent AI Delegation. DeepMind releases new approach to multi-agent orchestration and task coordination. Reddit

Qwen3.5-397B Available. Latest large language model from Qwen is now accessible through HuggingChat. HuggingChat

Building Solo Projects with AI. Developer shares insights on creating and completing side projects for personal use with AI assistance. CodeMade

Neurons Beyond the Brain. Essay explores how neural networks exist throughout the human body, not just in the brain. Debug Your Pain