xiaobai-skills: Curation Over Collection for Codex Skills
Not everyone using Codex needs to write Skills. But almost everyone using Codex gets burned by day three.
You install twenty Skills, a prompt fires, and three Skills jump out fighting for the job. Two frontend frameworks, three testing conventions, each claiming to know best. You delete one, and two days later discover that one was right. You keep them all, and Skill descriptions eat half your context budget.
Tyuts' xiaobai-skills exists to solve this mess. It's a curation tool, not a skill library. Its job isn't to give you more options—it's to help you make a choice while keeping an exit.
1. Curation, Not Collection
The current skill ecosystem: Vercel's skills.sh supports 50+ agents (Claude Code, Codex, Cursor) with one unified format. Matt Pocock's skills focus on engineering practices; Baoyu's skills cover frontend visuals. Each library is good—combined, they're a disaster.
xiaobai-skills' core design: "one need, one default, the rest as backup."
Unlike npm dumping everything into node_modules, it:
- Audits your installed Skills
- Identifies overlaps
- Lets you pick one as the global default
- Moves unselected ones to a backup folder instead of deleting them
- Generates a manifest explaining why each choice was made
- Scans name and description of all installed Skills
- Computes a "conflict score" from keyword overlap
- Lists high-conflict Skill pairs
- You pick one as primary; the other gets moved aside
- Use skills.sh to install Skills
- Use xiaobai-skills to decide which are globally active and which get backed up
- Early: no extensions, bare usage
- Explosion: extension counts explode; people running 200 are common
- Organization: curated sets emerge (like VS Code Extension Packs)
- Maturity: curated packs become default; ad-hoc installs become niche
- GitHub: https://github.com/Tyuts/xiaobai-skills
- Positioning: Skill curation and backup tool for Codex beginners
- Recommended set: 8 domains across frontend / image / content / engineering / curation
- Upstream dependencies: OpenAI official skills, Matt Pocock skills, Baoyu skills, skills.sh
- Install: PowerShell script, cross-platform support planned
- License: MIT
That manifest matters. Agents have no memory, but you can leave a map for the next session.
2. Quality of the Recommended Starter Set
| Area | Representative Skills | Source | |------|----------------------|--------| | Skill discovery | find-skills | OpenAI | | Skill creation | skill-creator | OpenAI | | Frontend implementation | frontend-master | OpenAI | | Image generation | imagegen | OpenAI | | Content workflows | URL→Markdown, translation, publishing | Baoyu curated | | Engineering practices | TDD, diagnostics, review | Matt Pocock curated | | Curation backup | xiaobai-skills itself | Tyuts |
The selection logic is pragmatic: OpenAI official skills as the base, Matt Pocock's engineering skills for depth, Baoyu's content skills for extension. No sprawling "framework-style" skills (like GSD, BMAD), since those seize control and are hard to debug when things go wrong.
The recommendation list explicitly marks gstack, superpowers, and some Matt Pocock sub-skills as "default backup candidates"—not because they're bad, but because they're too broad and conflict-prone.
3. Conflict Resolution Mechanism
This is where the tool has real technical substance.
When multiple Skills' trigger words overlap, Codex loads the full SKILL.md content of several at once. Context budget is consumed and agent confusion probability rises.
xiaobai-skills' solution is blunt but effective:
It doesn't make decisions automatically—only presents information. That's correct: automated choices easily guess wrong about your real preferences.
The backup directory ~/.codex/skills-backup/xiaobai-YYYY-MM-DD/ includes a README.md explaining why each Skill was moved. That detail reflects a design philosophy: today's bad decision should be easy for tomorrow's you to undo.
4. Relationship with skills.sh
Vercel's skills.sh is ecosystem infrastructure—unified install, update, removal. But it only cares whether something is installed, not whether it fits.
xiaobai-skills is the ecosystem's governance layer. It doesn't replace skills.sh; it builds on top of it:
The relationship is like npm and eslint: npm handles installation, eslint handles conventions. Both matter, at different layers.
5. A Prediction About the Agent Ecosystem
The Skill ecosystem is repeating VS Code extension history:
xiaobai-skills sits at the entrance of the "organization" phase. It assumes users know what Skills are but not which are worth installing and which should step aside.
One risk in that assumption: true beginners may not even realize curation is needed. Their likely behavior: install a bunch, hit problems, then blame the Agent.
So if xiaobai-skills could do one thing, its value would double: warn before conflicts happen. Not audit after twenty Skills are installed, but at the fifth install, remind you: "You already have three TDD-related Skills; pick one."
That requires hooks or interception of Codex install behavior. The project currently uses PowerShell scripts and can't reach that granularity yet—but it's a direction worth pursuing.
6. Limitations
1. Windows-centric: primarily PowerShell scripts; macOS/Linux installers are still TODO. But skill ecosystem power users skew heavily macOS/Linux.
2. No auto-updates: Skill libraries evolve constantly—Matt Pocock just added /prototype last week; Baoyu updates weekly. The recommended set is static and needs manual syncing.
3. Cold backups only: removed Skills require manual restoration; no elegant "three-day trial with auto-rollback" mechanism.
7. One-Line Verdict
If you've used Codex for over a week, installed more than ten Skills, and keep feeling "why did the Agent misunderstand again"—it's probably not the Agent. Your Skills are fighting each other.
xiaobai-skills doesn't fix the Agent's intelligence, but it fixes the chaos in your toolbox. At the turning point where the Skill ecosystem shifts from wild growth to careful cultivation, curation tools like this will multiply.
It's not about installing more—it's about installing right, with an exit.
---
Project Info