Top Picks

A research writeup from AgentSploit analyzing 1,384 tools across 28 MCP servers, highlighting how tool descriptions, source tools, and sink tools create an attack surface for agents.

A detailed benchmark-style article comparing Claude Code’s single-agent and multi-agent /code-review modes on hard pull requests, with measured cost, latency, and issue-detection results, plus a comparison to Quickchat Review.

A GitHub repository that provides a boilerplate for running Claude Code in a containerized, rule-driven software factory setup. It includes a sandboxed container, a command regulator, lint-based feedback hooks, and a workflow/orchestrator system for multi-agent task execution.

MIKE is an integrated Kanban environment and LLM harness for software development work, with gVisor-based sandbox isolation, custom extensions, and support for llama.cpp and OpenAI-compatible inference providers.

An agent skill that converts AI output into more readable formats such as simplified technical English, diagrams, interactive HTML, narrated explainers, and review summaries for human oversight.

A local-first todo board that connects to an AI assistant through MCP, letting the assistant read and update tasks on the same machine.

A lightweight open-source telemetry collector for agentic systems. It stores events in SQLite, serves a local dashboard, and provides HTTP endpoints to send and retrieve agent events.

A talk transcript and write-up on using LLMs to prioritize vulnerability research at scale, with references to two open-source tools: Slice for build-free static analysis and Raink for O(N) listwise ranking.

An arXiv paper describing a governed, local-first memory system for AI agents, with audit trails, role-based access, verified erasure, and fault-injection results. The page also links to associated code and a Zenodo DOI.

An essay about why AI agents should use project documentation instead of conversation memory, with a description of the author’s Operator Memory system for markdown-based agent memory.

A blog post documenting an agent-assisted prototype of a Choplifter-style game for Apple Vision Pro, including gameplay basics, Blender asset creation, and the author’s workflow using Astra and Claude Opus 5.5.

A command-wrapper tool that adds adaptive startup friction around agentic AI binaries like Claude Code or GitHub Copilot CLI, with Nix and non-Nix install paths and a stats command.

Cotyper is a coding agent/devtool aimed at fast iteration and working over existing GitHub projects. The page positions it as faster than Codex and lists supported tools including Claude Code and Codex, plus a GitHub connection flow.

KillSwitch is an esoteric programming language aimed at breaking LLMs and coding agents. The site includes a language overview, interpreter/specification/examples, and benchmark references.

A research note from OpenShell on using formal methods and Z3 to reason about and verify permission changes in long-running AI agent systems.

A Prompt One blog post about compiled workflow agents, design-time tool selection, and why runtime agents should not choose tools dynamically.

A paper page about FrogNano, a 4B coding agent trained with RL on roughly 1,500 SWE environments using synthetic tasks and no distillation from a larger model.

GitHub’s research preview on multi-model orchestration for coding tasks. The page describes adaptive workflow selection and includes benchmark comparisons against the Opus 5 baseline with estimated cost reduction claims.

AntiAgent is a human-in-the-loop planning tool that uses AI to ask sharp questions, surface priorities, and turn messy thoughts into a step-by-step execution plan.

Oconee Runtime is an enterprise AI governance product focused on policy enforcement for browser AI and coding agents. The page explains how it evaluates proposed agent actions against policy and supports allow, warn, and block outcomes.

An interactive quiz from Raq.com that asks users to identify which AI model created each of ten default website designs, with links to the live pages after each reveal.

A field note about testing a coding agent fully offline on a laptop with Ollama and Qwen3.8 27B, including the setup, failure modes, and the workaround that made the project finish.

A research page describing a verified circle-packing result produced by a frozen language model with an external memory loop, with published trace, attempts archive, and repository code/data.

An interactive simulation about how inference providers ration access to stronger LLMs when demand spikes. Users can compare the default industry rule with their own policy and explore who gets the strong model.