25 October 2026

Agent skill for multi-phase security audits

A coding agent skill that runs a security audit in six phases, with an isolated agent in each phase. Findings are written as machine-readable records.

• Each candidate finding is handed to a fresh verifier that tries to refute it.
• Final results are split into confirmed, needs_validation, and rejected in findings.json.
• This skill became the seed of Cloudflare's vulnerability discovery harness.

Secure runtime for autonomous AI agents

OpenShell runs each agent in a sandbox. You write a policy for what the agent may touch, and OpenShell enforces it.

• The kernel restricts file access and system calls, and every network connection passes a policy check.
• The agent never sees the real credentials. Credentials are added only to requests going to approved endpoints.
• Policy changes are checked with formal verification before they are approved.

Official MCP server, skills, and plugins for AWS

The official AWS toolkit that lets coding agents build, deploy, and manage applications on AWS. Works in Claude Code, Codex, Cursor, and others.

• The aws-core plugin covers CDK/CloudFormation, serverless, containers, storage, and billing.
• There are separate plugins for agents on Bedrock, data analytics, and DevSecOps.
• It can also be used directly from the terminal through the AWS CLI.

Design skill so pages do not look AI-made

A design skill for Claude Code, Codex, Cursor, and OpenCode. It picks a theme, builds the page, then lints the result in a real browser.

• There are 48 themes. A build takes three themes from the catalog.
• The lint looks for signs of a machine-made page, such as an eyebrow on every heading.
• /hallmark redesign rebuilds a page, while /hallmark audit only reports.

Visual guide to Claude Code with templates

A guide with visual tutorials and copy-ready templates for Claude Code features, from the basics to agents, hooks, skills, and MCP.

• Mermaid diagrams explain how each feature works.
• There is a step-by-step learning path and a get-started-in-15-minutes section.
• The README is available in several languages, including Tiếng Việt and 中文.

Design skill for AI harnesses

A single skill with 24 commands for design guidance in coding agents. It has 62 detector rules for AI-made frontend design.

• /impeccable init records product context in PRODUCT.md.
• Commands include craft, critique, audit, and polish.
• The CLI and browser extension run the detector rules without an LLM and without an API key.

Small bipedal robot that walks with an RL policy

This repo is the brain of the Microduck robot: about 25 cm tall and 800 g, running on a Rockchip RK3566. A neural policy controls fifteen servos.

• A 50 Hz control loop loads an ONNX policy trained in the microduck_rl repo.
• The robot can walk with a gamepad, roll on wheels, and pick up objects.
• Software updates are verified and can be rolled back.

Open source AI agent in the terminal

An open source agent that brings Gemini to the terminal. The README says Gemini CLI is replaced by Antigravity CLI on 18 June 2026 for free tier users.

• Can be run without installation using npx.
• Available through npm, Homebrew, MacPorts, and Anaconda.
• Preview releases are published every Tuesday at 23:59 UTC.

Collection of e-book download links

A repo of e-book links grouped by category, with a README in Chinese. Each book is available in epub, mobi, and azw3 formats.

• There is a GitHub Pages site with multi-keyword search.
• Popular categories: literature, history, popular science, management.
• Coverage includes books from several Chinese reading apps.

Open source AI agent that is ready to use out of the box

Cindy combines several harnesses, models, and tools in one agent that runs locally. This repo is the desktop and mobile client.

• The first supported harnesses are Claude Code and Codex.
• Memory, skills, automation, and MCP can be configured by the user.
• A pnpm monorepo: Electron for desktop, Expo for mobile, Apache-2.0 license.