Weekly Trending

Graphify is an agent skill for Claude Code, Codex, OpenCode, Cursor, Gemini CLI, and related tools that turns code, schemas, documents, papers, images, and videos into a queryable knowledge graph.

destructive_command_guard is a local enforcement layer that blocks dangerous git and shell commands before coding agents can execute them.

Open-source product analytics is being reframed around self-driving products, AI observability, session replay, experiments, workflows, and MCP access.

Long-Horizon-Terminal-Bench evaluates terminal agents on long tasks using dense intermediate reward signals, measuring progress through planning, context handling, iterative debugging, multimodal analysis, and software-engineering work rather than only final success.

Everyday perception becomes a streaming, queryable memory substrate for phones and AI glasses, split into current, short-term, and long-term memory.

Agentic visual generation gets a search reasoner that decides when the image model lacks world knowledge, filters noisy references, and integrates evidence into generation.

A research paper about making evolving AI agent harnesses readable, navigable, and editable by linking prompts, state, tool calls, behavior traces, and orchestration code to safe edit targets.

SearchOS-V1 proposes a multi-agent open-domain search system with shared state: frontier tasks, an evidence graph, a coverage map, failure memory, middleware, and reusable hierarchical search skills.