DOCSEARCHbyAlgolia
Submit Docs

Khazix Skills

repository·main·Indexed 12 days ago

https://github.com/kkkkhazix/khazix-skills

A collection of structured instruction sets for AI Agents following the Agent Skills open standard. Compatible with 40+ agents including Claude Code, Codex, and Cursor, it provides specialized capabilities such as the aihot skill for AI news, storage-analyzer for disk cleanup, leader for goal definition, and neat-freak for documentation alignment.

Tokens
36.5K
Snippets
49
Records
154
Agent score
97%

What's inside Khazix Skills

  1. Overview of Khazix Skills

    main

    Khazix Skills is a collection of open-sourced, structured instruction sets designed to be used by AI agents to perform specific, high-value tasks. Each skill is a proven workflow that can be loaded directly by compatible agents.

    Available Skills

    NameDescription
    leaderConverts vague ideas into clearly defined, actionable goals that an agent can execute autonomously to completion.
    storage-analyzerScans Mac/Windows drives to provide a three-tier cleanup plan and one-click trash management via the browser.
    aihotAllows agents to pull daily AI reports and news from aihot.virxact.com using a single Chinese sentence (no API key required).
    neat-freakReconciles documentation, CLAUDE.md, and agent memory after a session, and audits project rule compliance (triggered via /neat).
    hv-analysisPerforms Horizontal-Vertical Analysis on products, companies, or concepts to generate 10k–30k word PDF research reports.
    khazix-writerEnables agents to write long-form Chinese articles using a specific personal voice.

  2. Overview of available Khazix Skills

    main

    The repository contains several specialized skills:

    NameDescription
    leader（领导）Transforms vague ideas into clear, actionable goals for the AI to execute autonomously.
    storage-analyzer（清理垃圾）Scans Mac/Windows disks and provides a three-color classification for cleanup decisions via a web interface.
    aihot（AI HOT 资讯查询）Allows agents to fetch daily AI HOT reports and news from aihot.virxact.com without an API key.
    neat-freak（洁癖）Triggered by /neat, it automatically aligns project documentation, CLAUDE.md, and agent memory while auditing rule execution.
    hv-analysis（横纵分析法）Generates comprehensive research reports (up to 10,000 words in PDF format) about products, companies, or concepts.
    khazix-writer（卡兹克写作）Enables agents to write long-form articles in a specific tone and rhythm.

  3. Overview of the Storage Analyzer skill

    main

    The storage-analyzer is a read-only storage analysis assistant for macOS and Windows. It scans the entire disk to identify large files and directories, categorizes them into three safety levels (🟢 Auto-cleanable, 🟡 Manual judgment required, 🔴 Cautious cleanup), and generates an interactive HTML report. The report allows users to view space usage and, when running in server mode, perform one-click deletions (moving to trash or direct deletion) via a web interface.

    When to use this skill: Trigger this skill when a user mentions: "storage analysis", "disk cleanup", "disk full", "not enough space", "C drive full", "clean cache", or similar phrases regarding disk/storage space.

    Note: This skill is for storage/disk space, not RAM/Memory. If a user asks about process memory usage or RAM, do not use this skill.

  4. Query AIHOT Chinese AI news and hot topics

    main

    AIHOT provides a stable public v1 API to query Chinese AI news, curated highlights, current hot topics, and daily reports. It is designed to answer questions about recent AI news, model/product releases (OpenAI, Anthropic, Google, etc.), AI papers, and trending events.

    Key Features:

    • Anonymous & Read-Only: No API Key or MCP server required.
    • Curated Content: Uses a high-threshold selection pool (mode=selected) by default.
    • Time Windows: Supports 24h and 7d windows based on the AIHOT timeline.
    • No Training Memory: Does not rely on model training data; always fetches real-time data from the API to ensure accuracy.

  5. About Khazix Skills

    main
    Khazix Skills is a collection of personal productivity and AI-related tools developed by KKKKhazix (founder of Xushi Media). These skills are designed for daily use and are open-sourced for the community. The project is licensed under the MIT License, allowing for free use, modification, and redistribution.

  6. Quality check topics using the HKR Method

    main

    Before writing, evaluate every topic against the HKR Method. A top-tier (S-level) topic meets all three criteria. A passing topic meets at least two. If it meets only one, reconsider it; if it meets zero, abandon it.

    • H (Happy): Is it interesting or suspenseful? Do the title and opening spark curiosity?
    • K (Knowledge): Is there information density? Will the reader learn something new?
    • R (Resonance): Does it touch an emotion? Does it make the reader think, "Exactly, I feel the same way"?

  7. Follow the Release State Machine

    main

    All releases must progress through the following states. If a state is skipped, a reason allowed by project rules must be provided. If a failure occurs, report exactly at the state where it stopped; do not use terms like "basically complete" to cover incomplete states.

    State Sequence:

    1. implemented
    2. locally verified
    3. pushed / PR opened
    4. CI + required backtest/visual review passed
    5. merged
    6. deployed
    7. live verified
    8. knowledge closed + receipt recorded
    9. full result reported while evidence is preserved
    10. user explicitly approved cleanup after the report
    11. workspace cleaned
    12. post-cleanup audit passed
    13. cleanup result appended

  8. Synchronization invariants and constraints

    main

    To ensure data integrity when syncing with the aihot service, adhere to these rules:

    Cursor Rules

    • Opaque: The cursor is an opaque token. Do not parse, increment, modify, or reuse it across different endpoints.
    • Separation of Concerns: The cursor (sync watermark) and nextPage (pagination) are distinct and cannot be used interchangeably.

    Data Integrity

    • Atomic Updates: Never claim a sync is complete if hasMore=true during a snapshot. Never save partial snapshots.
    • No Time-Window Syncing: Do not use publishedAt or discoveredAt timestamps to perform incremental syncs. They do not reliably capture edits or removals. Do not use overlapping time windows; only the changes endpoint reliably expresses remove operations.
    • Endpoint Semantics: Do not use /api/v1/items?mode=all as a substitute for a full snapshot; it represents the recent public pool, which has different semantics.

    Performance & Polling

    • Polling Interval: Normal polling should occur at least every 60 seconds (except when clearing a hasMore=true backlog).
    • ETag Usage: For persistent tasks, save the ETag of the full URL and send it via If-None-Match. If the server returns 304, maintain your current local state and cursor.

  9. Use the Change Impact Matrix for documentation routing

    main

    The Change Impact Matrix is a framework used to route updates to the correct audiences (documentation, architecture, operations, etc.) without forcing a rigid file structure. Instead of creating fixed filenames, use the matrix to identify existing architecture, integration, operations, decision, and release documentation, then ensure the corresponding audience is updated.

    When adding or changing a capability, evaluate it against these four responsibility types:

    1. How to use: README, API/integration guides, examples.
    2. How it works: architecture, schema, state machines, design boundaries.
    3. How to operate: runbooks, environment variables, smoke tests, rollback procedures, alerts.
    4. Current status/History: product changelog, release notes, incidents, or git history (only write changelogs if explicitly required by project rules/users).

    If a single document fulfills multiple responsibilities, update it in place rather than splitting it into multiple files for the sake of form.

  10. What are Khazix Skills and how do they work?

    main

    Khazix Skills are structured instruction sets designed for AI Agents. Each skill follows the Agent Skills open standard, making them compatible with over 40+ agents including Claude Code, Codex, Qoder, Kimi Code, iFlow, CodeBuddy, and Cursor.

    When an agent supports the Agent Skills standard, it can automatically load and execute these instructions to perform specialized tasks like disk analysis, research, or specific writing styles.

  11. Storage Analyzer execution workflow

    main

    The skill follows a four-step process:

    1. Scan (Read-only): Runs scan.py to collect disk usage data using du (macOS) or os.scandir (Windows). This step is strictly read-only.
    2. Analyze & Classify: An agent analyzes the scan JSON to identify Top 5 space consumers, investigate "mysterious large directories" (like UUID-named containers), and assign the three-light safety levels.
    3. Generate Report: An analysis.json is built and rendered into an interactive HTML report.
    4. Summary: A concise text summary is provided in the chat, highlighting total estimated reclaimable space, the top 2-3 priority items, and the highest risk item.

  12. L4 Human-Feel Final Audit: The Personality Layer

    main

    The most important and subjective layer. Read the entire article from a reader's perspective to answer: "Does this feel like a knowledgeable person talking to me about something that moves them, or an AI outputting information?"

    L4-1 Temperature (Emotional Resonance)

    • Use visceral, sensory emotional expressions (e.g., "I was stunned right then" or "my nose felt sore") instead of clinical descriptions (e.g., "I felt very shocked").
    • Characters should feel like they have "body temperature."

    L4-2 Uniqueness

    • Does the article offer an angle that "only Khazix would write"? It should not be something any AI blogger could produce.

    L4-3 Stance/Posture

    • The tone should be that of a "knowledgeable ordinary person talking seriously about something that moves them."
    • Avoid slipping into a "teacher lecturing a student" or "brand doing marketing" posture.

    L4-4 Flow (Mental State)

    • Check for any points where attention drops or logic becomes difficult to follow. These are rhythm issues that need fixing.

    Pass Standard: The overall feeling must be "written by a human." If any part feels "too AI-like," it must be reworked by rewriting the section in a more colloquial, personal, and imperfect way.