Salesforce has published an open-source collection of agent skills for building applications. The skills are optimized for Salesforce's Agentforce but designed to work with various AI tools.
A new paper presents 4DAnyone, a model that can generate a 4D representation (3D + time) of a person from a single casual monocular video. The approach shows significant quality improvement for dynamic 3D avatars.
Researchers have introduced AI4AI-Bench, a benchmark designed to evaluate LLM agents' ability to perform algorithmic design for recursive self-improvement. This targets a more advanced frontier of AI capability.
Community reports suggest Qwen3.8 27B packaged with Q6 quantization performs exceptionally well for agentic coding, outperforming other models of a similar size. Users are citing strong performance for a locally-runnable model.
This paper introduces FlashPrefill V2, a method for block-sparse prefill attention that promises faster and more efficient long-context LLM serving. It directly addresses computational bottlenecks in the prefill phase.
Apache Maka (Incubating) is a new local-first AI agent workspace that records all agent activity — messages, tool calls, permission decisions — as an append-only log. It's designed for transparency and auditability of agent operations.
AI-Infra-Guard is an open-source platform for red-teaming AI ecosystems, featuring Agent Scan, Skills Scan, MCP scan, and LLM jailbreak evaluation. It addresses security testing across the full AI stack.
A new open-source repository provides 817 structured cybersecurity skills for AI agents, mapped to six major frameworks including MITRE ATT&CK, NIST CSF 2.0, and D3FEND. The skills are compatible with Claude Code, GitHub Copilot, Codex CLI, Cursor, and 20+ other platforms.
NVIDIA's AVO system completed all 183 levels across all 25 public environments in the ARC-AGI-3 benchmark, achieving a perfect score without instructions, explicit rules, or stated goals. The benchmark tests an agent's ability to infer objectives and solve novel problems.