27 October 2026

A design skill against AI-looking interfaces

A design skill for Claude Code, Codex, Cursor, and OpenCode, made by Together AI. It includes 48 themes and lints pages in a real browser.

• Picks three themes from a catalog of 48, then builds a page around the visitor's task.
• sloplint checks the page at eight screen widths, then a reviewer with fresh context judges the result.
• /hallmark redesign rebuilds the page, while /hallmark audit only reports.

Financial trading framework with multiple LLM agents

A Python framework that uses several LLM agents for trading analysis and decisions. The latest release is v0.7.0 from October 2026.

• v0.7.0 can run on your own ChatGPT Plus or Pro plan and scores backtests from the first close after each decision.
• v0.6.0 saves reports as a single HTML page and gives each model tier its own provider.
• v0.5.2 runs analysts in parallel and has a CLI with flags such as --ticker and --date.

Agentic IDE built around the pi coding agent

A lightweight IDE from JetBrains with an interface built around the pi coding agent. Work is split into isolated workspaces, and project knowledge is stored through specs.

• Each workspace is created as a git worktree with its own branch and folder, so agents work in parallel without touching your checkout.
• A graph of SPEC.md files next to the code is read and updated by the agent through the spec_* tools.
• Includes a Monaco editor, a terminal, and PR and CI status in the top bar.

Ready-to-use open source AI agent

A local AI agent that combines several harnesses, models, and tools. The first supported harnesses are Claude Code and Codex.

• Runs on your machine with your real files and apps you are already logged into. It can control a browser, a computer, and a phone.
• The model and harness can be swapped mid-task, while the workspace, memory, skills, and tools carry on.
• This repo is the open source client (Apache-2.0), a pnpm monorepo.

Harness system for coding agents

An engineering system for agents: plan, test, review, remember, then turn the results into skills. It includes 71 agents, 311 skills, and 97 command shims.

• The flow is: plan, test, implement, review, verify, remember, improve.
• Works best with Claude Code. There are plugin paths for Codex and Qoder, and limited adapters for Cursor, OpenCode, Gemini, and others.
• Includes hooks, rules, memory, continuous learning, and the AgentShield security scanner.

Design guidance for coding agents

A single skill with 24 commands and 62 deterministic detector rules for AI-generated frontend design.

• /impeccable init writes PRODUCT.md with the product's audience, goals, and constraints.
• Commands such as polish, audit, critique, distill, and animate are invoked through /impeccable.
• The CLI and browser extension run the deterministic rules without an LLM and without an API key.

Security scanner for AI agent skills

Made by NVIDIA. Detects vulnerabilities, dangerous patterns, prompt injection, and data exfiltration in skills before they are installed.

• Of 31,132 skills in a research subset, 26.1% contain vulnerabilities and 5.2% appear malicious.
• Part of the NVIDIA Verified Skills pipeline, which scans, evaluates, and signs skills.
• Has extensions for Pi and OpenCode.

Small biped duck robot driven by an RL policy

This repo is the brain of a robot about 25 cm tall and 800 g, running on a Rockchip RK3566.

• A 50 Hz control loop drives fifteen servos from a neural policy.
• The policy is trained in the microduck_rl repo with MuJoCo and PPO, then exported to ONNX.
• The robot can walk, roll on wheels, pick up objects, and stand up by itself after a fall.

Desktop workspace for persistent AI coworkers

A local-first desktop app for persistent agents based on Codex and Claude Code, plus Grok CLI, OpenCode, Gemini, Cursor CLI, and Cline CLI through ACP.

• Each agent has its own workspace and conversation, with a local queue, file transfer, an embedded browser, and agent-to-agent messaging.
• Still a development preview: agents run with danger-full-access and approvalPolicy: never.
• Supports OpenAI-compatible endpoints and detects local model servers such as Ollama and LM Studio.

One CLI so agents can read the internet

A CLI that installs and tests the tools behind platforms such as Bilibili, V2EX, GitHub, YouTube, and the web, so an agent only needs one command.

• agent-reach channels lists the channels and what they do. agent-reach read <url> picks the channel from the link.
• Default channels: bilibili, v2ex, github, youtube, web. Reddit, Twitter, and others need a login.
• The README warns that the agent-reach package on PyPI is not this project.