Each skill was used in real projects long enough to prove useful before being open-sourced. The repository contains structured instruction sets that agents can load directly and follows the Agent Skills open standard. It works with Claude Code, Codex, Qoder, Kimi Code, iFlow, CodeBuddy, Cursor, and many other compatible agents.

📋 Skills

Name	What it does
🧭 leader	Turns a vague idea into a clearly defined goal an agent can execute autonomously for hours.
💽 storage-analyzer	Scans Mac or Windows storage and produces a three-tier cleanup plan with safe actions.
🔥 aihot	Lets an agent query AI HOT daily reports and AI news without an API key.
🧹 neat-freak	Reconciles project docs, CLAUDE.md / AGENTS.md, and agent memory after a work session.
🔭 hv-analysis	Produces deep horizontal-and-vertical research on a product, company, concept, or person.
✍️ khazix-writer	Helps an agent write long-form Chinese articles in the author's personal style.

📦 Install

In an agent that supports Agent Skills, ask it to install the desired skill from this repository:

Install this skill: https://github.com/KKKKhazix/khazix-skills/tree/main/<skill-name>

Replace <skill-name> with the skill you want. If the agent does not support the Agent Skills standard directly, download that skill's SKILL.md and provide it as a project rule file or conversation context.

🧭 leader

The leader skill focuses on defining goals correctly before an autonomous agent starts long-running work. It turns an incomplete idea into a goal task book with explicit purpose, definition of done, proof requirements, anti-cheating constraints, bounds, trade-offs, and unresolved decisions. The goal is to prevent a long-running agent from optimizing for an unintended shortcut while still giving it enough freedom to complete the work independently.

💽 storage-analyzer

storage-analyzer scans a computer drive and opens an interactive HTML report with disk usage, major space consumers, and prioritized cleanup suggestions. Items are classified into green, yellow, and red tiers so the user can distinguish safe caches from user data and system-critical files. Scanning is read-only and destructive actions require explicit confirmation.

🔥 aihot

aihot allows compatible agents to retrieve AI HOT daily reports, trending events, editorial selections, category feeds, time-window feeds, and keyword searches from aihot.virxact.com. It can also mirror the current selection locally and later fetch only changes. No API key or MCP server configuration is required.

🧹 neat-freak

neat-freak is designed for project closeout. It checks what changed during a session and aligns project documentation, root AI instruction files, and agent memory with the actual codebase. It can identify outdated documentation, broken references, missing rule files, and leftover temporary work. It never deletes files automatically; deletion candidates remain subject to user confirmation.

🔭 hv-analysis

hv-analysis researches a subject along two axes: a vertical historical view from origin to the present, and a horizontal comparison with current competitors or alternatives. Combining both perspectives is intended to reveal conclusions that would be missed by looking only at current state or only at history. Typical outputs are long-form research reports useful for competitor analysis, company research, and learning a new domain.

✍️ khazix-writer

khazix-writer encodes the author's long-form Chinese writing style, including rhythm, narrative preferences, rhetorical rules, banned phrases, self-checks, and style examples. It is intended primarily for users who want Chinese long-form output in that specific voice.

🌟 About

Khazix is the founder of Virxact and shares practical AI workflows and tools. These skills are the author's own day-to-day tools and are released under the MIT License.

Made by @KKKKhazix