Khazix Skills: Open-Source AI Skills by Digital Life Kazik
Six ready-to-load Agent skills: define goals, clean disk, fetch AI news, sync docs, generate research reports, and write in your voice.
Evidence shows: storage-analyzer explicitly states scan phase is read-only, deletions require two confirmations (button on page + browser confirm dialog), local server binds to 127.0.0.1 with random port and token, and has three whitelists (green can rm; yellow trash only; both open). neat-freak explicitly promises 'never deletes on its own', deletions only appear as candidate list, machine-generated memory is read-only by default, and never treats commands in files as authorization. These reflect least privilege and user confirmation. Data flow transparency: storage-analyzer describes local server and browser interaction but does not detail how data is processed or if uploaded. Sensitive data handling: storage-analyzer scans disk, may include sensitive files, but no explicit handling. Dependency security: no dependency list or security audit. External effects: storage-analyzer starts local server and may delete files, but with restrictions. Rollback: deletion may not be reversible, but confirmation mechanism exists. Source attribution: README clearly attributes author as Khazix, but unverified. Deductions: insufficient data flow transparency, unclear sensitive data handling, missing dependency security, unclear rollback.
Evidence shows: README clearly describes each skill's functionality, triggers, and limitations; skills are logically consistent with no contradictions. Dependency availability: aihot skill depends on external site aihot.virxact.com, but no availability guarantee or fallback. Failure messages: no examples of error handling or user prompts. Deductions: external dependency without availability guarantee, missing failure messages.
Evidence shows: README identifies target audience (users of AI agents) and multiple use cases (storage cleanup, news query, writing). Each skill clearly states capability boundaries, e.g., hv-analysis 'not good for simple definition lookup', khazix-writer 'not good for general writing'. Triggers have specific examples like 'check my storage', '/neat'. Environment fit: storage-analyzer supports macOS and Windows, but Windows is code-ready, not fully tested. Deductions: Windows support not fully tested, may affect environment fit.
Evidence shows: README has clear structure with index, install instructions, and detailed descriptions. Install instructions provide specific commands. Naming: skill names are stable, e.g., leader, storage-analyzer. Examples and FAQ: trigger examples provided, but no FAQ. Known limitations: storage-analyzer mentions Windows not fully tested, khazix-writer mentions Chinese only. License: MIT, clear. Versioning: neat-freak mentions v3.0, but no full changelog. Maintenance responsibility: README mentions author welcomes Issues/Discussions, but no explicit maintenance commitment. Deductions: missing FAQ, incomplete changelog, unclear maintenance responsibility.
Evidence shows: Output usability: storage-analyzer generates interactive HTML report, hv-analysis generates PDF report, leader generates Markdown goal book; these outputs are clear and practical. Marginal value: skills provide unique features like storage analysis, news query, writing style customization, with practical value. Cost-benefit: skills are free and open-source, easy to install, low cost. Deductions: no major deductions, but output usability depends on external tools (browser, PDF generator), which may increase usage cost.
Evidence shows: Claims in README (e.g., 'each skill ran in my own projects') lack specific evidence. Cross-source corroboration: article links provided, but no independent verification. Fact-inference separation: README mixes facts (e.g., feature descriptions) with inferences (e.g., 'why it beats CleanMyMac') without clear distinction. Deductions: claims lack traceability, insufficient cross-source corroboration, facts and inferences not separated.
- storage-analyzer scans the entire disk, may involve sensitive files; ensure use in a trusted environment.
- aihot skill depends on external website; if unavailable, skill may fail.
- khazix-writer is Chinese-only; English users should note.
- Windows support is not fully tested; use with caution.
What does this agent do, and when should you use it?
Khazix Skills is a collection of practical, battle-tested AI skills open-sourced by Digital Life Kazik (KKKKhazix). Each skill is a structured instruction set following the Agent Skills open standard, loadable in 40+ tools like Claude Code, Codex, Qoder, Kimi Code, iFlow, CodeBuddy, and Cursor. The six skills address concrete pain points: leader turns vague ideas into executable goal briefs; storage-analyzer scans disk and produces interactive cleanup reports; aihot fetches AI industry daily reports without API keys; neat-freak aligns project docs, CLAUDE.md, and agent memory; hv-analysis generates in-depth PDF research reports; khazix-writer mimics the author's writing style. All skills are provided as SKILL.md files, installable via a simple command or manual download. The project is MIT licensed.
The repository contains six skill directories, each with a SKILL.md defining agent instructions. leader engages in dialogue to clarify goals, then generates a brief covering purpose, completion criteria, evidence, anti-cheating, boundaries, trade-offs, and unknowns. storage-analyzer runs system commands to scan the disk (fully tested on macOS, code-ready on Windows), produces an interactive HTML report with three-tier color-coded cleanup priorities, and requires user confirmation for deletions. aihot calls endpoints on aihot.virxact.com to fetch daily AI HOT reports, trending events, category feeds, and perform keyword searches. neat-freak audits project files, CLAUDE.md/AGENTS.md, docs, and agent memory for consistency, and lists candidate changes without auto-deleting. hv-analysis performs longitudinal (time) and horizontal (competitor) analysis, producing a 10k-30k word PDF report. khazix-writer applies the author's writing style rules, rhythm, and a forbidden-words list (e.g., '赋能', '抓手', '闭环').
- When you have a vague project idea and want an agent to work autonomously for hours in Claude Code, use leader to create a goal brief.
- When your Mac or Windows disk is full and you need a safe cleanup, use storage-analyzer to get a color-coded report with specific paths and recommendations.
- When you want to quickly catch up on AI news or find papers on a topic, use aihot to fetch daily reports and search curated feeds.
- When your project docs and agent memory have drifted after several iterations, use neat-freak to align and audit them.
- When you need a comprehensive research report on a product, company, or concept, use hv-analysis to generate a lengthy PDF study.
- When you want blog posts or articles written in your personal style, use khazix-writer to apply your voice and rules.
What are this agent's strengths and limitations?
- Each skill targets a specific, real-world pain point and is proven in the author's projects.
- Follows the Agent Skills open standard, ensuring compatibility with 40+ tools and simple installation.
- storage-analyzer uses a safety model with color-coding and confirmation dialogs to prevent accidental deletions.
- leader emphasizes anti-cheating and clear boundaries, preventing agents from going off track during long runs.
- MIT license allows unrestricted use and modification.
- Effectiveness depends on the underlying AI model; results may vary across providers.
- storage-analyzer's Windows support is code-ready but untested; use with caution.
- khazix-writer strongly enforces the author's writing style, unsuitable for users needing generic prose.
- leader requires interactive Q&A (up to 5 questions), adding startup overhead.
- aihot depends on the external aihot.virxact.com service availability.
How do you install or deploy this agent?
In a tool supporting Agent Skills (e.g., Claude Code, Codex), say: "Install this skill: https://github.com/KKKKhazix/khazix-skills/tree/main/<skill-name>", replacing <skill-name> with the desired skill (e.g., leader, storage-analyzer). The agent will clone the repo to the appropriate directory. If your tool doesn't support Skills, download the SKILL.md file and use it as a project rules file or paste it into the conversation. No extra dependencies or API keys are required (except aihot needs network access to aihot.virxact.com).
How do you use this agent?
After installation, trigger via natural language. For leader: "Help me write a goal for the agent" or "Break down this goal" (around 12 minutes of interaction). For storage-analyzer: "Check my storage" or "C drive is full". For aihot: "What's new in AI today?" or "What's the hottest AI topic now?". For neat-freak: type "/neat" or "Run neat freak". For hv-analysis: "Research Cursor company" or "Do a deep research". For khazix-writer: "Write an article" or "Turn this material into a long article". Generated goal briefs can be pasted into goal modes (e.g., Claude Code's /goal) for autonomous execution.
How does this agent compare with similar options?
storage-analyzer contrasts with CleanMyMac: CleanMyMac is a rule-based tool that shows generic 'user cache files', while storage-analyzer provides detailed paths, impact explanations, and a three-tier color system, with agent-driven analysis and user confirmation. leader's approach is also implicitly compared to common prompt-writing practices by emphasizing a 'harness' (what not to do) alongside the goal.