19 October 2026

An agent meta-harness for Claude Code and Codex

Ruflo adds agent swarms, adaptive memory, and vector RAG integration to Claude Code, Codex, and other MCP clients.

• Installed as Claude Code plugins: core (base tools), console (agent cockpit), and mods (routing and policy). Requires Claude Code 2.1.287 or later.
• For Codex or other MCP clients, run the setup wizard and connect to the Ruflo MCP server (npx -y ruflo@latest mcp start).
• The console also runs in a regular terminal via npx ruflo tui: Overview, Cost, Swarm, Events, and Missions pages.

Long-term memory shared across coding agents

Quit Claude Code in the middle of a task, open Codex in the same folder, and carry on without re-explaining the architecture, the failed attempts, or the open questions.

• More than 20 harnesses (Claude Code, Codex, Cursor, Gemini CLI, OpenCode, and others) write to one shared memory.
• Handoffs use a typed protocol: each handoff has an owner and can only be claimed once.
• A profile of working habits (package manager, test layout) follows you into each new repo when the session starts. Written in Rust.

One place to run many coding agents

Run Codex, Claude Code, OpenCode, or Pi side by side, each in its own git worktree, using your own subscriptions. Available on desktop, mobile, and remote.

• Parallel Worktrees: send one prompt to five agents, compare the results, then merge the best one.
• iOS and Android mobile app: get notified when an agent finishes and send follow-ups from your phone.
• Design Mode: click a UI element in a Chromium window to send its HTML, CSS, and a screenshot crop to the agent prompt.

Makes any software usable by agents through a CLI

CLI-Anything builds a CLI for existing software so that agents like Claude Code, Cursor, or OpenClaw can use it. CLI-Hub collects community-made CLIs.

• Install the hub with pip install cli-anything-hub, then install a CLI with cli-hub install <nama>.
• README demos: an agent uses generated CLIs to build CAD, 3D scenes, diagrams, gameplay, and subtitles.
• The registry keeps growing, for example CLIs for Obsidian and ArcGIS Pro were proposed in May 2026.

Official Android skills for AI agents

Modular instructions in the SKILL.md format that follow the developer.android.com guidance. The focus is on areas where evaluation results show LLMs are still weak.

• Install a single skill with android skills add r8-analyzer --project=., or all of them with --all.
• Available as a Claude Code plugin (/plugin install android-skills --marketplace android/skills) and for Codex.
• Follows the open agent skills standard from agentskills.io.

A free LLM course with a roadmap and Colab notebooks

The course is split into three parts: LLM Fundamentals (math, Python, neural networks), The LLM Scientist (building LLMs), and The LLM Engineer (building and deploying LLM applications).

• Ready-to-use notebooks include LazyMergekit for merging models and AutoQuant for quantizing to GGUF, GPTQ, EXL2, AWQ, and HQQ.
• The author also wrote the book LLM Engineer's Handbook based on this course. The course itself remains free.

A tool for restoring mosaic censoring in videos and images

Jasna finds mosaic areas in a video and restores them without touching the rest of the frame. Free, with an app and a command line, and runs on the GPU.

• Two models: the fast BasicVSR++ (free) and the more detailed LTX (supporters only, NVIDIA only).
• Requires an NVIDIA GTX 16-series or RTX 20-series or newer. AMD RX 7000/9000 is still experimental.

A node-based GUI, API, and backend for diffusion models

ComfyUI uses a node graph to create images, video, 3D models, and audio, with control over every model and parameter.

• Natively supports the latest open-source models.
• Can be used as a GUI, through the API, or as a backend.