AI-Agents

ARC-Skill is a lightweight agent skill that lets an unmodified Claude Code agent finish all 25 ARC-AGI-3 games and all 183 levels at a perfect 100.00 RHAE score, using just 7,645 actions versus a median human’s 17,135. It works by enforcing a single rule: before the agent presses any button, it must write down exactly what that press will do to the grid, and the harness refuses to act until a prediction is recorded and graded. This article explains how that one rule unlocks frontier-level performance, how ARC-Skill compares to heavier harness systems, and what it teaches about agentic skill engineering. ...

The DeepSeek Harness plugin hub is a decentralized, source-backed marketplace where every capability of the harness — models, tools, skills, sessions, sandboxes, storage, loops, scheduling, and the UI — is a swappable plugin. It grew from 0 to 7,740 plugins in under three weeks, making it one of the fastest-growing plugin ecosystems ever built. This guide explains how the hub works, how to install and publish plugins, and what to check before you trust one. ...

OpenSheet is an in-browser spreadsheet framework built specifically for AI agents, letting them read, query, and directly edit cells in CSV and Parquet files without sending your data to a cloud server. It runs on duckdb-wasm for local processing and evolved from the author’s earlier DataKit project. This review explains how it works, why spreadsheets are uniquely hard for LLMs, and how it compares to the growing field of spreadsheet agents. ...

ZhiShi is an open-source agent harness purpose-built for security research — a human-in-the-loop workbench for binary exploitation, penetration testing, whitebox code audit, and AI security — that layers a domain-specific research loop on top of the MIT-licensed @earendil-works/pi-agent-core engine. It uses a Tauri (Rust) desktop shell wrapping a TypeScript/Node core, and publishes unusual evidence-first proof of capability: 100% Pwnable.kr CTF completion, 98% Hacker101, and a fully reproduced exploit chain for CVE-2026-34621 (Adobe Acrobat Reader, CVSS 8.6). ...

A self-evolving agent harness is the layer of prompts, memory, skills, tools, and control logic that lets an AI agent improve its own behavior over time. Proteus is an open-source, MIT-licensed research-preview framework that lets you plug in any agent harness, evolve it across episodes, and measure exactly what changed. Instead of chasing a single benchmark score, Proteus asks what a self-evolving harness actually does and whether an initial condition leaves a permanent mark. ...

DeepSeek V4 Flash Vision Exp is an experimental multimodal model that accepts images alongside text, outputs text only, and pairs a 1M-token context window with an Intelligence Index score of 51 — ranked #3 out of 172 comparable models. At cache-hit input prices near $0.0028 per million tokens, it is shaping up as the cheapest capable vision path for high-frequency agent loops. Here is everything developers need to know before building with it. ...

MCP observability turns the Model Context Protocol into a two-way interface: AI agents don’t just call tools, they receive ground-truth telemetry from kernel tracepoints, eBPF programs, and kprobes. By exposing low-level system instrumentation through MCP servers, agents get a real-world model of the live system instead of hallucinated state — closing the observability gap that traditional APM leaves wide open. What Is MCP and Why It Needs Observability The Model Context Protocol (MCP) is an open standard that standardizes how AI agents discover and invoke tools, resources, and prompts. Instead of every agent building bespoke integrations with every service, MCP defines a common protocol: a host (the agent runtime) connects to MCP servers, which expose tools the model can call and resources it can read. ...

Simmis is a self-hosted workspace where people and AI agents share a single, versioned, queryable memory — chat rooms, wiki pages, knowledge bases, a double-entry ledger, and a code repository all live on one substrate instead of being scattered across siloed tools. Built in Clojure on the replikativ stack, it treats humans and agents as the same kind of participant, so @mentions, governance, and history work identically across both. It is an early, MIT-licensed “release early” project that is used daily but still has rough edges. ...

AgentBrain is a local-first, long-term memory tool for AI agents that stores everything as a plain Markdown vault. Instead of locking your agent’s memory into a proprietary database or cloud API, it gives you an append-only Markdown folder you can read, edit, grep, and version with Git. It uses index-first retrieval with CJK-aware BM25 to stay token-efficient, and enforces human-approved consolidation to prevent multi-agent write conflicts. What Is AgentBrain? A Local-First Markdown Memory Vault for AI Agents AgentBrain is an open-source Python tool (Python 3.10+) that gives AI agents a durable, local-first memory layer stored as plain Markdown files. The core idea is simple: an agent’s long-term knowledge should live in a human-readable folder on your own machine, not inside a black-box database or a third-party cloud service. ...

An agentic PostgreSQL investigation is an AI-driven workflow that explores a database to answer operational questions, and OpsGraph makes it evidence-first: it discovers an approved read-only schema, validates every SELECT against the PostgreSQL AST and policy bounds, and records a tamper-evident audit chain so no conclusion can be asserted without opening its exact evidence. Instead of giving a model direct credentials and “asking it anything,” OpsGraph constrains the investigation to bounded, read-only, auditable queries that fail closed on unsafe SQL. ...