misaka-mikoto-tech/agent-skills

agent-skills is a personal collection of reusable skills and utilities for AI coding agents, built mainly for Codex. It gathers day-to-day habits the author has built up around web page reading, safe command line usage, automated workflows, and integrating outside tools. Most of what's here is a set of SKILL.md files that can be copied straight into an agent's skills folder, and the author plans to add MCP server configuration and examples over time as well. Two skills currently exist. The first, powershell-safe-invocation, is a set of rules for calling PowerShell safely from an agent on Windows. It focuses on how modern PowerShell 7 handles native command arguments, file paths, quoting, character encoding, and starting new processes, so an agent following it avoids common mistakes when running commands on a user's machine. The README recommends installing PowerShell 7 or newer and links to Microsoft's official installation guide and release page. The second skill, bilibili-page-reader, reads the content of a Bilibili video page. It first tries to pull the video's official subtitles, a summary of the scrolling on-screen comments, and the regular comment section. If no official subtitles exist, it falls back to downloading the audio and transcribing it locally. To do this it relies on a browser bridge tool that lets the agent see a real, logged-in browser page, a browser extension called Bilibili Evolved that exposes subtitle and comment data, Node.js to call Bilibili's video URL API and download audio, ffmpeg to convert and inspect that audio, and a speech recognition library called FunASR to turn audio without subtitles into text. To install a skill, a user copies its folder from this repository into their agent's skills directory, such as Codex's local skills folder, or wherever their own agent looks for skills. The repository is organized with a skills folder holding each skill in its own directory, and an mcp folder currently reserved for future MCP server work. The project is released under the MIT license.