skill-creator
How to create new skills for this HQ — format, frontmatter, content structure. Use when you need to add a new skill to this repository, or when reviewing whether an existing skill is well-formed.
sabahattink · 已有静态检查记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add sabahattink/antigravity-fullstack-hq --skill skill-creator技能说明
name: skill-creator description: How to create new skills for this HQ — format, frontmatter, content structure. Use when you need to add a new skill to this repository, or when reviewing whether an existing skill is well-formed.
Skill Creator
What is a Skill?
A skill is a reference document that gives an AI agent (or a developer) deep, actionable knowledge for a specific domain. It is not a README, a tutorial, or a one-liner tip. It contains real patterns, real code, and clear guidance on what NOT to do.
File Location
skills/
└── <skill-name>/
└── SKILL.md
Skill names are lowercase, hyphenated: api-design-patterns, auth-patterns, docker-patterns.
Required Structure
---
name: skill-name
description: One-line description. Use when [specific trigger condition].
---
# Title
## [Section 1]
[Content with real code examples]
## [Section 2]
...
## Forbidden Patterns
[What NOT to do — these are as important as the positive guidance]
Frontmatter Rules
---
name: skill-name # must match the directory name
description: > # one line: what it does + when to trigger
Generates Excel files with ExcelJS. Use when creating .xlsx exports
from database data in a NestJS app.
---
The description field is what an AI uses to decide whether to load this skill. Make the trigger condition specific:
# Good — specific trigger
description: JWT access/refresh tokens, Passport.js strategies. Use when implementing auth in a NestJS + Next.js app.
# Bad — too vague
description: Authentication patterns. Use when building auth.
Content Rules
1. Real code, not pseudocode
Every code block must be runnable with the imports shown. Never use ... to hide required setup.
// Good — complete, runnable snippet
import { Injectable } from '@nestjs/common'
import { InjectRepository } from '@nestjs/typeorm'
import { Repository } from 'typeorm'
import { User } from './user.entity'
@Injectable()
export class UsersRepository {
constructor(
@InjectRepository(User)
private readonly repo: Repository<User>
) {}
findById(id: number) {
return this.repo.findOne({ where: { id } })
}
}
// Bad — too abstract
class MyRepo {
// inject the repo here
findById(id) { ... }
}
2. Forbidden Patterns section is mandatory
The "Forbidden Patterns" section must be the last section. List 5-10 concrete things that cause real bugs, security issues, or maintenance problems.
## Forbidden Patterns
- Never store access tokens in localStorage — XSS can steal them
- Never use the same JWT secret for access and refresh tokens
- Never skip `@IsEmail()` validation on email fields
3. Aim for 200-400 lines
- Under 200 lines: too shallow, not useful as a reference
- Over 400 lines: likely covering too many topics, should be split
4. One topic per skill
Don't mix unrelated concerns. If you're writing a testing skill, don't include deployment. If the topic is naturally broad, split it:
# Too broad:
skills/backend/SKILL.md ← covers NestJS + TypeORM + auth + caching
# Better:
skills/backend-dev-guidelines/SKILL.md ← NestJS layering
skills/auth-patterns/SKILL.md ← JWT + Passport
skills/docker-patterns/SKILL.md ← Containerization
Section Patterns to Follow
Command blocks for setup
npm install exceljs
npm install -D @types/node
Configuration files
Show actual config files, not fragments.
Patterns with comparison (good vs bad)
// Bad: explanation of why it's bad
const bad = doWrongThing()
// Good: explanation of why it's better
const good = doRightThing()
Checklists for pre-flight checks
Production Deployment Checklist:
□ NODE_ENV=production set
□ Secrets validated at startup
□ Health checks responding
□ Rate limiting enabled
Quality Checklist
Before merging a new skill:
□ Frontmatter has name + description
□ name matches directory name
□ Description includes "Use when [specific condition]"
□ Code examples are complete (have imports, no ellipsis hiding setup)
□ At least 3 major sections with real content
□ "Forbidden Patterns" section exists and has 5+ entries
□ 200-400 lines total
□ No tutorial prose — reference-style only
□ TypeScript/language specifics match the project stack
Example: Creating a New Skill
# 1. Create the directory and file
mkdir skills/redis-patterns
touch skills/redis-patterns/SKILL.md
# 2. Write the skill following the template above
# 3. Commit
git add skills/redis-patterns/SKILL.md
git commit -m "feat: add redis-patterns skill"
Forbidden Patterns
- Never write a skill that's just a list of links — content must be here, not linked
- Never use placeholders like
[your code here]— write the actual code - Never mix two unrelated domains in one skill — split instead
- Never skip the "Forbidden Patterns" section — negative examples are the most valuable
- Never write in tutorial style ("first you need to...") — use reference style (show the code)
- Never omit the trigger condition from the description — it's what activates the skill