What it does

This repo packages five “Skills” — structured instruction sets that Claude Code, Codex, OpenClaw, and similar agents can load and run. They cover housekeeping (neat-freak syncs your code changes with docs and agent memory), research (hv-analysis spits out 10k–30k-word PDF reports), ghostwriting (khazix-writer clones the author’s WeChat public account voice), news aggregation (aihot pulls daily AI briefings without API keys), and disk cleaning (storage-analyzer with a traffic-light safety model).

The interesting bit

The author, a Chinese AI content creator, dogfooded these daily before open-sourcing them. The standout is neat-freak: it fixes “brain rot” where agent memory and project docs drift from reality after iterative coding — something most agent tools ignore. The disk cleaner’s 🟢🟡🔴 permission model (agent can delete, move to trash, or just explain) is a rare bit of safety-first UX in a space that usually trusts AI with rm -rf.

Key highlights

• neat-freak aligns three audiences separately: agent memory, CLAUDE.md/AGENTS.md, and human-facing docs — the author explicitly notes Claude Code’s AutoDream only handles memory
• hv-analysis runs parallel vertical (historical) and horizontal (competitive) analysis threads, intersecting them for conclusions
• khazix-writer is opinionated: it refuses to write corporate buzzwords like “empowerment,” “closed loop,” or “in today’s fast-paced AI era”
• aihot works across 7+ agent platforms including Cursor, Cline, and Windsurf; offers a China-direct install script that bypasses the firewall
• storage-analyzer runs read-only scans, serves results via local 127.0.0.1 with random port + token, and requires browser-based double-confirmation for any destructive action

Caveats

• storage-analyzer is “fully tested on macOS”; Windows support is “code-ready” but the README advises caution on first use
• All skills are Chinese-first; English support exists but the voice, examples, and news sources are Sino-centric
• The repo is essentially high-quality prompt engineering + some tooling, not a framework — valuable if the specific skills fit your workflow, less so if you need extensibility

Verdict

Grab this if you use Claude Code/Codex daily and want battle-tested, opinionated workflows rather than building your own prompts. Skip if you need a general-purpose skill framework or don’t read Chinese.