Khazix Skills bundles six battle-tested agent skills from a 19k-star repo — goal definition, disk cleanup, AI news, and personal-style writing for Claude Code.
帮我安装这个 skill:https://github.com/KKKKhazix/khazix-skills/tree/main/<skill-name>What It Is
Khazix Skills is a 19.4k-star collection of six AI agent skills open-sourced by 数字生命卡兹克 (Khazix), founder of Xushi Media, for daily personal use. The official README describes it plainly: “AI Skills I use every day, all open-sourced here — nothing fancy, just a few genuinely practical things.” Every skill is a structured instruction set an agent can load directly, following the Agent Skills open standard that Claude Code, Codex, Qoder, Kimi Code, iFlow, CodeBuddy, Cursor, and 40+ agents can install.
The six skills cover very different jobs: leader turns a vague idea into a formal goal document an agent can run autonomously for hours; storage-analyzer scans a whole Mac or Windows disk and produces a color-coded cleanup report in the browser; aihot pulls the daily AI HOT digest from aihot.virxact.com with one sentence and no API key; neat-freak realigns project docs, CLAUDE.md, and agent memory after every task; hv-analysis researches a product, company, or concept into a 10,000–30,000-word PDF report; and khazix-writer makes the agent write long-form WeChat articles in the author’s own voice and rhythm.
What Makes It Different
The first thing that stands out is that every skill was run in production before being open-sourced. The README’s stated filter is explicit: only things “that ran in my own projects for a while and genuinely saved me effort” make the cut. That’s a different quality bar than most skill repos, where skills are written speculatively and never exercised. Each one also ships with a real-world use case — the storage-analyzer’s README even documents how it identified a 97 GB UUID container as Bilibili offline video cache and recommended cleaning it inside the app rather than deleting it by hand.
Second, the leader skill treats “what not to do” as the core of goal-setting. The README’s argument is that the most important part of a goal is its harness: JFK’s moon speech mattered because of “and safely return them to Earth,” not because of “land on the Moon.” The skill encodes a “seven questions” framework (purpose, completion state, evidence, anti-cheating, boundaries, trade-offs, unknowns) plus a zeroth question — verify the map yourself before sailing. It even recommends pairing models: a strong planner (Claude Fable 5) to write the goal and a strong executor (GPT-5.6 Sol) to run it.
Third, safety is designed in rather than promised. The storage-analyzer is strictly read-only by default: every deletion requires a browser button click plus a second confirmation dialog, runs on 127.0.0.1 with a random port and token, and uses a three-tier whitelist (green = cache you can delete, yellow = user data you move to trash, red = system files you can only open). neat-freak likewise never deletes anything without a candidate list you confirm, and treats machine-generated memory as read-only by default.
Who It’s For
Khazix Skills fits three types of users. First, Claude Code and Codex users who want goal-driven long-running tasks. If you’ve ever wanted to paste a vague idea and let an agent work for hours, the leader skill’s goal documents are designed for exactly that — the README suggests triggers like “help me write a goal for the agent” or “let the agent run this project on its own,” and the output is pure Markdown that works even without a goal mode.
Second, writers who want a specific voice rather than generic prose. khazix-writer is opinionated by design: it refuses clichés like “empower, leverage, closed loop,” rejects “firstly… secondly” and “in today’s rapidly developing AI era,” and enforces the author’s own rhythm through a four-layer self-check (structure, pacing, content, wording). It’s the opposite of a generic “improve this text” prompt.
Third, solo operators doing research and housekeeping. hv-analysis produces structured PDF research reports for competitive analysis or concept deep-dives, aihot keeps you current on AI news without configuring an API key, and storage-analyzer replaces CleanMyMac-style utilities with an agent-driven, explainable cleanup flow.
Skip it if you want a single cohesive skill rather than a grab-bag — these are six independent tools with different purposes, and you should install only the ones you need. It’s MIT licensed.
Bottom Line
Khazix Skills is the rare skill repo where everything is proven in daily use: goal documents that let agents run for hours, an agent-driven disk cleaner with real safety rails, no-API-key AI news, project doc hygiene, deep-research PDFs, and a writing voice that actively rejects AI-slop phrasing. If you use Claude Code or Codex daily, install the two or three that match your workflow — pair the goal-focused leader skill with the Baoyu Skills pipeline for long-running content tasks, or use khazix-writer alongside the HTML Anything skill to go from idea to published page. For a first-person take on AI writing workflows, see our guide on how AI writing assistants changed content creation.
Frequently Asked Questions
What is Khazix Skills: 6 Practical Agent Skills for Claude Code?
How to install Khazix Skills: 6 Practical Agent Skills for Claude Code?
帮我安装这个 skill:https://github.com/KKKKhazix/khazix-skills/tree/main/<skill-name>