Khazix Skills - Setup Guide¶

Source: kkkkhazix/khazix-skills via skills.sh - 15.1K combined installs across 9 indexed listings; first documented November 4, 2026 (morning sweep) GitHub: 21,933 stars, MIT licence, last pushed 2026-11-03, 6 SKILL.md directories at the repository root Category: Agent workflow for builders: planning, research, writing and delivery skills wired for Claude Code and Hermes Quality Tier: 🟡 Yellow: skill-manager and github-to-skills both draw a Gen Agent Trust Hub Fail, storage-analyzer draws Warns from all three scanners, and three of the 9 indexed listings have no directory in the tree

Khazix Skills is the working set of a Chinese AI writer and builder, published as six MIT licensed skills at the repository root. The install index is wider than the tree: it carries 9 listings, while the tree holds 6 SKILL.md directories. The three without a tree counterpart are skill-manager (295 installs), Skill Evolution Manager (289) and github-to-skills (260). The README says the repository maintains the six skills it documents, that it no longer distributes the older skill-manage and self-evolution manager set, and that names from older tutorials should not be read as current install directories.

The six live skills split into three kinds of work. leader turns a vague request into a goal brief an agent can run unattended, neat-freak closes a task by reconciling documentation and memory with what the code now does, and hv-analysis runs a two-axis research report. On the production side, khazix-writer writes long-form articles in the author's voice, aihot pulls daily AI news without an API key, and storage-analyzer analyses disk usage with a tiered cleanup report.

Installation¶

The README's own route is conversational: you give the agent a tree URL for the skill you want, and it clones the directory into place. The pattern is the repository URL plus a skill name:

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

Swap the placeholder for a real directory such as neat-freak, hv-analysis or khazix-writer. The README is explicit that you should keep the whole skill directory, including references/, scripts/, assets/ and agents/ where present, because pasting SKILL.md alone keeps the method but loses the scripts and resources.

Claude Code can also install all six through the plugin directory:

/plugin marketplace add KKKKhazix/khazix-skills
/plugin install khazix-skills@khazix-skills

Updates take two commands, and the README warns against keeping a plugin install and a hand-copied directory of the same skill at once:

/plugin marketplace update khazix-skills
/plugin update khazix-skills@khazix-skills

Pinned versions and full download bundles are on the repository's Releases page. For Hermes, clone the six root directories into the profile skills directory and keep scripts/ intact; three of the skills run local tooling rather than only producing text.

What It Provides¶

Six live skills and three indexed listings with no directory in the tree.

Skill	Installs	Use For
neat-freak	3,432	Closeout that aligns docs, CLAUDE.md and AGENTS.md, agent memory and residue with the code
khazix-writer	3,173	Long-form article writing in the author's voice, with style rules and a four-layer self-check
hv-analysis	2,929	Two-axis research: narrative history plus competitor comparison, crossed into a report
aihot	2,552	Daily AI news, hot events, timelines and keyword search with no API key
storage-analyzer	1,634	Read-only disk analysis with green, amber and red tiers and an interactive HTML report
leader	536	Turns one vague sentence into a goal brief with a whitelist, anti-cheat checks and resume points
skill-manager	295	No directory in the current tree; the index keeps the historical install count
Skill Evolution Manager	289	No directory in the current tree; the README says the older self-evolution set is no longer distributed
github-to-skills	260	No directory in the current tree; no replacement skill is documented

All 9 indexed listings appear in the table above.

Why This Matters for Hermes Agents¶

Most skill libraries are domain knowledge. This one is workflow, and it aims at the parts of a long session that rot: the goal at the start, the knowledge at the end, the research in the middle and the news feed in the background.

leader is the design worth studying. It inspects the codebase before writing anything, asks at most five questions, and produces a brief capped at 4,000 characters with an explicit whitelist, anti-cheat acceptance criteria and resume points, on the reasoning that a long-running agent will take whatever shortcut you left unstated. Copy that structure into any dispatch prompt.

neat-freak treats documentation drift as the failure mode rather than an afterthought. After a task it reconciles the README, docs/, CLAUDE.md and AGENTS.md, and the agent's own memory, with what the code now does, then audits whether the project rules were actually followed.

hv-analysis is a reusable research shape: one axis follows the subject from origin to now as narrative, the other compares it against contemporaries, and the cross of the two produces the judgement neither axis shows alone. It outputs 10,000 to 30,000 characters and a PDF by default.

aihot shows the right pattern for a news skill. The skill only asks and relays, while fetching, ranking and summarising happen on the service side, so it needs no API key and no update when the feed changes.

One caution: storage-analyzer can delete files through a local web service. Install it with the tier rules in view.

Usage¶

You say	What happens
"Write a goal brief for the agent so it can run this alone"	leader inspects the repo, asks up to five questions and writes a capped goal brief
"Run /neat before I close this session"	neat-freak aligns docs, rule files and agent memory with the code and audits rule compliance
"Research this company properly"	hv-analysis builds the history axis and the competitor axis, then crosses them into a report
"What happened in AI today?"	aihot pulls the daily digest and the current hottest events from aihot.news
"My disk is full, what can I delete?"	storage-analyzer scans read-only, tiers the findings and opens an HTML report
"Write this material up as a long article"	khazix-writer applies the author's voice rules and runs a four-layer self-check
"Install the writing skill"	the agent clones the khazix-writer directory with its references and scripts

Verification¶

Confirm the skills landed in the agent skills directory:

ls ~/.claude/skills

Then fetch one raw file before installing. This path is neat-freak/SKILL.md on branch main and returned 200 when this guide was written:

curl -fsSL https://raw.githubusercontent.com/kkkkhazix/khazix-skills/main/neat-freak/SKILL.md | head -20

Frontmatter with a name and a description confirms the listing is backed by a real file. leader/SKILL.md is the other path worth checking. The same probe shows the drift: skill-manager/SKILL.md and github-to-skills/SKILL.md return 404 on branch main, which matches the README's note that the older set is no longer distributed. The install index is at https://www.skills.sh/kkkkhazix/khazix-skills.

Security¶

skills.sh verdicts for sampled skills (verified November 4, 2026) - re-check the security pages on skills.sh before production use:

Skill	Gen Agent Trust Hub	Socket	Snyk
neat-freak	Pass	Pass	Pass
hv-analysis	Warn	Pass	Warn
storage-analyzer	Warn	Warn	Warn
skill-manager	Fail	Pass	Warn
github-to-skills	Fail	Pass	Warn

neat-freak is the only skill in the sample that passes all three scanners, which is worth noting because it is also the one that edits documentation and rule files in your project. Both Fail verdicts land on listings with no directory in the current tree, and storage-analyzer carries a Warn from every scanner while shipping a local service that can delete files.

Integration with CorpusIQ¶

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects

Limitations¶

Snapshot data, verified November 4, 2026: 15,100 combined installs across 9 indexed listings; 21,933 GitHub stars; MIT licence; last pushed 2026-11-03. Counts drift over time.
The install index outruns the tree: 9 indexed listings against 6 SKILL.md directories. skill-manager (295), Skill Evolution Manager (289) and github-to-skills (260) have no directory on branch main, and the README says the older skill-manage and self-evolution manager set is no longer distributed.
Two of the nine listings carry a Gen Agent Trust Hub Fail (skill-manager and github-to-skills) and only neat-freak passes all three scanners. Neither failing listing has a tree directory, so the flagged code is not necessarily what you would install, but re-check the security pages rather than assuming.
All six live skills are written in Chinese, trigger phrases included. The repository links an English README, but the skill bodies themselves are not translated, so an English prompt may not fire a skill without editing its triggers.
hv-analysis targets 10,000 to 30,000 characters and a PDF by default. That is a large spend for a question a shorter path could answer, and the report format is the point rather than the length.
storage-analyzer supports macOS and Windows only. Its delete flow depends on a local service on 127.0.0.1 with a token, and the README says the service exits after 30 idle minutes.

Pro Tips¶

Install six, not nine. The three listings without a tree directory cannot be cloned, whatever the install count says.
Read leader even if you never run it. The goal-brief structure, with a whitelist and anti-cheat acceptance, is the most portable idea in the set.
Keep the scripts/ and references/ folders. storage-analyzer and neat-freak both depend on local tooling that a bare SKILL.md does not carry.
Run neat-freak at the end of a task rather than mid-way. It compares finished work against the docs, and a half-built tree gives it the wrong answer.
Pick one install route per skill. The README warns that a plugin install plus a hand-copied directory of the same skill leads to version confusion.

← Skills Catalog | Skills Marketplace → Powered by CorpusIQ