Cognitive Core Skills taxonomy released with 159 skill cards, benchmarks, and CI
A universal, industry-neutral taxonomy of cognitive core skills for LLMs, SLMs, AI agents, and world models has been released on GitHub. It includes schemas, 159 skill cards, benchmarks, and continuous integration, covering perception, memory, reasoning, planning, action, verification, learning, and governance.
Entities
Related
Developer completes Python-for-AI course covering agents and LLM evals
A developer completed a comprehensive Python-for-AI course that covered core Python, data structures, tooling, and agent-specific material including LLM evals, the Analyze-Measure-Improve cycle, and building a basic AI coding agent from scratch. The course also covered first-principles agent architecture with intelligence layer, memory, tools, validation, and control.
Meta releases SWE-Together benchmark measuring coding agent steering difficulty
Meta introduced SWE-Together, a new benchmark that evaluates coding agents on interactive, multi-turn tasks rather than single-shot problem solving. The benchmark measures how much human steering an agent requires, which correlates strongly with its capability. This addresses a key limitation of SWE-bench, which tests agents on frozen tickets alone.
ChatGPT Work: persistent AI agent for complex tasks
ChatGPT Work is an OpenAI tool that performs ongoing, complex tasks autonomously, persisting for hours or days. It uses LLMs to plan and execute multi-step workflows, solving the problem of AI agents that stop after a few minutes. Aimed at professionals needing long-running automated assistance.
Community compares local LLMs for agentic workflows using tool-eval-bench
A GitHub user published an interactive comparison report evaluating local LLMs for agentic workflows, using the tool-eval-bench benchmark (84 scenarios, 16 categories, 8 trials). The report targets single DGX Spark or other 96-128GB rigs and covers multi-turn tool orchestration, function calling, and autonomous planning as exercised by Hermes Agent.
UniClawBench benchmark proposed for evaluating proactive AI agents in real-world tasks
Researchers introduced UniClawBench, a universal benchmark for evaluating proactive agents that operate everyday tools in real-world environments. Unlike existing benchmarks that rely on sandboxed settings and single-turn evaluations, UniClawBench aims to isolate specific model capabilities to identify root causes of agent failures.