메뉴
Use when writing README files, API documentation, user guides, or technical documentation following industry standards from Google, Microsoft, and GitLab style guides.
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/MadAppGang/magus --skill documentation-standards명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
SKILL.md 표시 중
SOC 직업 분류 기준
Picks the browser-use agent brain LLM per task. Use when choosing a model for retry_with_browser_use_agent or a cloud session, or the user mentions a cheap, slow, or smart model.
Reference for the Browser Use MCP tools — parameters, returns, session lifecycle. Use when navigating, clicking, typing, evaluating JS, or extracting page content.
Visual UI debugging — screenshot capture and analysis, responsive layout checking at multiple viewport sizes, CSS validation via DOM state, visual regression detection, before/after state comparison.
| name | documentation-standards |
| description | Use when writing README files, API documentation, user guides, or technical documentation following industry standards from Google, Microsoft, and GitLab style guides. |
| user-invocable | false |
Comprehensive documentation standards based on research from Google, Microsoft, GitLab, Red Hat style guides, and 60+ additional sources. Provides templates, best practices, and anti-pattern detection for AI agents writing documentation.
Why: Reduces cognitive load 20-30%, shortens sentences 15-25%
Implementation:
BAD (passive): The request is processed by the server.
GOOD (active): The server processes the request.
BAD (future): The program will save your file.
GOOD (present): The program saves your file.
Validation: Check for "is/are/was/were + past participle" patterns
Why: 70% of visitors are first-time users, time-to-first-success critical
Implementation:
## Quick Start
```bash
npm install my-tool
npx my-tool init
npx my-tool start
Visit http://localhost:3000
That's it!
**Validation**: Can a new user succeed in under 5 minutes with zero context?
---
#### 3. Use Progressive Disclosure (Three-Tier Structure)
**Why**: Accommodates all skill levels, reduces overwhelm
**Implementation**:
Tier 1: Quick Start (2-5 min) -> Copy-paste code, immediate success, no explanations
Tier 2: Tutorial (15-30 min) -> Step-by-step with explanations, build one feature
Tier 3: Reference (lookup) -> Complete API docs, searchable, advanced patterns
---
#### 4. Use Second Person ("You")
**Why**: Increases engagement, reduces word count 15-25%
**Implementation**:
```markdown
BAD: The user should configure the settings.
GOOD: Configure the settings.
BAD: One can install the plugin by...
GOOD: Install the plugin by...
Why: Comprehension drops 30% when sentences exceed 30 words
Implementation:
BAD (42 words): The configuration file, which should be located in the
root directory of your project and named .claude, contains all the settings
that are required for the plugin system to function correctly and load
the appropriate agents.
GOOD (18 words): Place the `.claude` configuration file in your project
root. It contains settings for the plugin system and agents.
Why: 70% faster comprehension vs paragraphs
Implementation:
## Prerequisites
Before starting, you need:
- [ ] Node.js 18+ installed
- [ ] Git access to GitHub
- [ ] API token configured
## Tool Comparison
| Tool | Best For | Stars | Learning Curve |
|------|----------|-------|----------------|
| Docusaurus | React | 55k+ | Medium |
| VitePress | Vue | 12k+ | Low |
Why: Deep language integration, automatic type inference
Decision Matrix:
Language -> Tool:
- TypeScript -> TSDoc comments + TypeDoc generator
- Python -> Google/NumPy docstrings + Sphinx
- Rust -> /// comments + rustdoc
- Go -> Plain comments + godoc
- Java -> Javadoc
Why: Clear taxonomy reduces confusion
Four Types:
| Type | Purpose | Length | When to Use |
|---|---|---|---|
| Tutorial | Teach by doing | 15-60 min | "I'm new, show me" |
| How-To | Solve problem | 5-15 min | "I need to do X" |
| Reference | Describe what exists | As needed | "What parameters?" |
| Explanation | Clarify concepts | 5-30 min | "Why does this work?" |
Why: Users know what success looks like
Implementation:
### Example: Fetching User Data
**Code:**
```typescript
const user = await api.getUserById('user-123');
console.log(user.name, user.email);
Expected Output:
John Doe john@example.com
Error Case:
const user = await api.getUserById('invalid-id');
// Throws: NotFoundError: User with ID 'invalid-id' not found
---
#### 10. Provide Troubleshooting Section
**Why**: 50%+ of support questions are error-related
**Template**:
```markdown
## Troubleshooting
### Issue: Command not found
**Symptom**: `/implement` returns "command not found"
**Cause**: Plugin not enabled
**Solution**:
1. Check `.claude/settings.json`
2. Run `/plugin reload`
3. Restart if needed
**Prevention**: Verify with `/plugin list`
Why: Users think in terms of goals, not architecture
Implementation:
## Guides
### How-To Guides (Task-Oriented)
- [How to deploy to production](deploy.md)
- [How to add authentication](auth.md)
- [How to handle file uploads](uploads.md)
Why: Eliminates invisible barriers for beginners
Template:
## Prerequisites
Before starting, ensure you have:
- [ ] Node.js 18+ installed ([Download](https://nodejs.org))
- [ ] Basic JavaScript knowledge
- [ ] Text editor (VS Code recommended)
Why: Prevents hallucination and outdated examples
Process:
Why: Prevents documenting non-existent features
Process:
Why: Users need to know if docs apply to their version
Template:
---
**Version**: v2.0.0
**Last Updated**: 2026-01-09
**Compatible With**: Claude Code 1.5+, TypeScript 5.0+
---
> **DEPRECATED**: This approach works but superseded by [New Approach](link)
These rules eliminate AI-detectable writing patterns and enforce professional technical writing standards. dev:docs MUST apply these rules in write and fix modes, and check for violations in analyze mode.
NEVER use these words or phrases. Each tier has escalating severity.
CRITICAL — instant disqualifier (AI generation artifacts):
HIGH — marketing superlatives and difficulty dismissers:
MEDIUM — corporate jargon and filler:
STRUCTURAL OVERHEAD — throat-clearing and meta-commentary:
HEDGING CASCADES — qualify-then-qualify patterns:
Vary sentence lengths for natural rhythm. Monotonous sentence length signals AI generation.
Targets:
Anti-monotony rule: No more than 3 consecutive sentences within ±5 words of each other.
Transitions: Words like "however", "therefore", "for example" help readers connect ideas. Use them — but don't open more than 40% of paragraphs with an explicit transition word.
AI-generated text has detectable patterns. Avoid these:
Paragraph variety: Don't lock every paragraph into topic-sentence → supports → conclusion. Start some paragraphs with:
List variety: Don't make every list exactly 3 or 5 items. Use 2, 4, 6, or 7 items when the content calls for it. Never pad a list with filler items to reach a round number.
Section length variety: Vary section lengths. Some sections need 50 words, others need 300. Don't make every section approximately the same length.
Target 40%+ code coverage (code blocks as a percentage of total content). Developer documentation should show, not tell. Every concept explanation should have an accompanying code example within 2 paragraphs.
Following Stripe/Twilio/Vercel patterns:
## Prerequisites followed by "Before starting, ensure you have the following installed:"## Prerequisites: Node.js 18+, plugin SDKLead with essentials. Layer details in 5 tiers:
Use collapsible sections (<details>) for advanced content that most readers skip.
Escape hatches: At each tier, provide a link forward for users who already know the basics ("Already familiar? Skip to the API reference →").
Every sentence must add new information. Delete:
Include at least one Mermaid diagram for architecture or flow documentation. Diagrams must:
Every code example must:
Target: <10% passive voice sentences. Use imperative mood for instructions.
Exception: Use passive voice when the actor is irrelevant or unknown:
Purpose: Project landing page with 5-minute setup Max Length: 80 lines Use When: Creating project README
# [Project Name]
[One-line description of what this does]
## Quick Start
```bash
# Step 1: Install
npm install my-tool
# Step 2: Initialize
npx my-tool init
# Step 3: Start
npx my-tool start
Visit http://localhost:3000
That's it!
MIT - See LICENSE for details
---
### Template 2: TSDoc Function Documentation
**Purpose**: Document TypeScript functions
**Format**: TSDoc standard
**Use When**: Documenting public APIs
```typescript
/**
* Retrieves a user by their unique identifier.
*
* @remarks
* This function includes user permissions and preferences in the response.
* Results are cached for 5 minutes to reduce database load.
*
* @param id - User unique identifier (UUID format)
* @param options - Optional fetch configuration
* @param options.includePermissions - Include user permissions (default: true)
*
* @returns Promise resolving to User object with permissions
*
* @throws {NotFoundError} If user with given ID doesn't exist
* @throws {UnauthorizedError} If caller lacks 'read:users' permission
*
* @example
* Basic usage:
* ```typescript
* const user = await getUserById('user-123');
* console.log(user.name, user.email);
* // Output: John Doe john@example.com
* ```
*
* @example
* Error handling:
* ```typescript
* try {
* const user = await getUserById('user-123');
* } catch (error) {
* if (error instanceof NotFoundError) {
* console.error('User not found');
* }
* }
* ```
*
* @see {@link createUser} for creating new users
* @since 2.0.0
* @public
*/
Purpose: Document common errors and solutions Organization: By symptom Use When: Creating troubleshooting guides
# Troubleshooting Guide
## Installation Issues
### Error: EACCES permission denied
**Symptom**: npm install fails with permission error
**Full Error Message**:
npm ERR! code EACCES npm ERR! syscall access npm ERR! path /usr/local/lib
**Cause**: Trying to install globally without proper permissions
**Solution**:
**Option 1 (Recommended)**: Install locally
```bash
npm install --save-dev my-tool
npx my-tool
Option 2: Fix npm permissions
mkdir ~/.npm-global
npm config set prefix '~/.npm-global'
export PATH=~/.npm-global/bin:$PATH
Prevention: Use local installation (avoid -g flag)
Symptom: Import fails with "Cannot find module 'X'"
Causes:
Solution:
Step 1: Verify package is installed
npm list my-package
Step 2: Check import path matches package name
// CORRECT
import { func } from 'my-package';
// WRONG
import { func } from 'my-package/lib/index';
Prevention: Always install dependencies before importing
---
### Template 4: Tutorial Structure
**Purpose**: Teach by doing
**Length**: 15-30 minutes
**Use When**: Creating beginner tutorials
```markdown
# Tutorial: Building Your First REST API
**What you'll learn**:
- How to create a basic REST API
- How to define routes and handlers
- How to connect to a database
- How to handle errors
**Prerequisites**:
- [ ] Node.js 18+ installed ([Download](https://nodejs.org))
- [ ] Basic JavaScript knowledge
- [ ] Text editor (VS Code recommended)
**Estimated time**: 30 minutes
---
## Step 1: Set Up Project
Create a new directory and initialize npm:
```bash
mkdir my-api
cd my-api
npm init -y
What this does: Creates package.json with default settings
Create src/server.ts:
import express from 'express';
const app = express();
const PORT = 3000;
app.get('/', (req, res) => {
res.json({ message: 'Hello, World!' });
});
app.listen(PORT, () => {
console.log(`Server running on http://localhost:${PORT}`);
});
What this does:
Start development server:
npm run dev
Expected output:
Server running on http://localhost:3000
Test the endpoint:
curl http://localhost:3000
Expected response:
{"message":"Hello, World!"}
Success! You've created your first API endpoint.
You now have a basic REST API with:
Issue: Server won't start
Solution: Check if port 3000 is already in use:
lsof -i :3000
kill -9 <PID>
---
### Template 5: Changelog
**Purpose**: Track version changes
**Format**: Keep a Changelog standard
**Use When**: Every release
```markdown
# Changelog
All notable changes documented here.
Format based on [Keep a Changelog](https://keepachangelog.com/),
adheres to [Semantic Versioning](https://semver.org/).
## [Unreleased]
### Added
- New feature X for user authentication (#123)
### Changed
- Improved performance of database queries (#145)
---
## [2.0.0] - 2026-01-09
### Breaking Changes
#### Agent Configuration Format Changed
**Old format:**
```json
{
"agents": ["architect", "developer"]
}
New format:
{
"agents": {
"architect": { "enabled": true },
"developer": { "enabled": true }
}
}
Migration:
npx @my-tool/migrate-config
oldAuthMethod() - use newAuthMethod() instead
---
### Template 6: Architecture Decision Record (ADR)
**Purpose**: Document significant architecture decisions
**Format**: Lightweight ADR
**Use When**: Technology choices, patterns, security decisions
```markdown
# ADR-001: Use PostgreSQL for Primary Database
## Status
Accepted (2026-01-09)
## Context
We need persistent storage for user data, transactions, and analytics.
**Requirements**:
- Store structured user data
- Support complex queries
- ACID compliance for transactions
- Handle 100K users, 1M records/month
## Decision
We will use **PostgreSQL 15+** as our primary database.
**Implementation**:
- Host on AWS RDS
- Use connection pooling (pgBouncer)
- Enable read replicas
## Consequences
### Positive
- ACID compliance for data integrity
- Rich query capabilities (JSON, full-text search)
- Proven scaling path
### Negative
- Schema migrations require planning
- Not optimal for document storage
### Risks
- **Risk**: Database bottleneck at scale
- **Mitigation**: Early read replicas and caching
## Alternatives Considered
### MongoDB
**Pros**: Flexible schema, horizontal scaling
**Cons**: Limited transactions, weaker consistency
**Why rejected**: Financial data requires ACID
### DynamoDB
**Pros**: Fully managed, auto-scaling
**Cons**: Vendor lock-in, complex pricing
**Why rejected**: Team expertise in SQL
---
**Author**: [Name]
**Reviewed by**: [Tech Lead]
Purpose: Systematic problem-solving Organization: By category and symptom Use When: Creating comprehensive troubleshooting docs
# Troubleshooting Guide: [Component]
**Last Updated**: 2026-01-09
**Covers**: Version 2.0.0+
---
## Quick Diagnostics
**Before detailed troubleshooting**:
1. **Check prerequisites**:
- [ ] Dependencies installed
- [ ] Environment configured
- [ ] Services running
2. **Enable debug mode**:
```bash
DEBUG=* my-tool start
tail -f ~/.my-tool/logs/latest.log
Symptom: npm install fails
Root Cause: Insufficient permissions
Solution:
npm install --save-dev my-tool
npx my-tool
Prevention: Use local installation
Symptom: "Missing required config" error
Root Causes:
Solution:
Step 1: Verify .env exists
ls -la .env
Step 2: Check location (must be project root)
Step 3: Verify format
# CORRECT
API_TOKEN=abc123
# WRONG (no quotes)
API_TOKEN="abc123"
Level 1: Quick (5 min)
Level 2: Investigation (15 min) 4. Enable debug mode 5. Check logs 6. Search known issues
Level 3: Deep (30+ min) 7. Add debug logging 8. Use debugger 9. Profile performance
Before reporting:
Bug report includes:
my-tool --versionnode --version
---
## 52-Point Quality Checklist
Use this checklist to validate documentation quality (42-point base + 10-point anti-slop).
### Content Quality (8 points)
- [ ] **No Over-Marketing**: Technical content, not sales pitch
- [ ] **No Feature Hallucination**: All documented features exist
- [ ] **No Assumption Overload**: Prerequisites explicitly listed
- [ ] **No Code Duplication**: Comments add insight, not repeat code
- [ ] **No Copy-Paste Docs**: Single source of truth
- [ ] **Examples Tested**: All code examples work
- [ ] **Errors Documented**: Top 5 errors with troubleshooting
- [ ] **Version Tracked**: Version header and date
### Structure Quality (8 points)
- [ ] **Quick Start First**: In first 20 lines (README)
- [ ] **Progressive Disclosure**: Simple -> Complex
- [ ] **User Journey Clear**: Persona-based entry points
- [ ] **Consistent Formatting**: Follows template
- [ ] **Hierarchy Logical**: H1 -> H2 -> H3
- [ ] **Lists for Steps**: Tasks >3 steps numbered
- [ ] **Tables for Comparison**: Comparison in tables
- [ ] **Navigation Present**: "Next Steps" included
### Writing Style (8 points)
- [ ] **Active Voice**: "Server processes" not "is processed"
- [ ] **Present Tense**: "Program saves" not "will save"
- [ ] **Second Person**: "You configure" not "user configures"
- [ ] **Short Sentences**: Average <25 words
- [ ] **Short Paragraphs**: 3-5 sentences max
- [ ] **Plain Language**: Terms defined
- [ ] **No Jargon**: Acronyms defined
- [ ] **Scannable**: Headings every 200-300 words
### AI-Specific (8 points)
- [ ] **Source Code Verified**: Claims match implementation
- [ ] **API Signatures Correct**: Parameters accurate
- [ ] **Examples Work**: Copy-paste tested
- [ ] **Version Compatible**: Versions stated
- [ ] **Edge Cases Included**: Error cases documented
- [ ] **Human Reviewed**: Expert validated
- [ ] **No Over-Confidence**: Uncertain qualified
- [ ] **Citations Provided**: Sources included
### Completeness (6 points)
- [ ] **Prerequisites Listed**: Dependencies stated
- [ ] **Expected Output Shown**: Results included
- [ ] **Error Cases Covered**: Common errors documented
- [ ] **Troubleshooting Present**: Dedicated section
- [ ] **Next Steps Provided**: Guidance included
- [ ] **Search Optimized**: Keywords in headings
### Maintenance (4 points)
- [ ] **Date Stamped**: "Last Updated: YYYY-MM-DD"
- [ ] **Version Noted**: "Version: X.Y.Z"
- [ ] **Deprecation Warnings**: Old approaches marked
- [ ] **Links Valid**: All links work
### Anti-Slop Quality (10 points)
- [ ] **No CRITICAL Banned Words**: Zero AI artifacts or marketing superlatives (2pt)
- [ ] **No MEDIUM Banned Words**: No corporate jargon or filler phrases (1pt)
- [ ] **No Throat-Clearing**: No section openers like "In this section..." (1pt)
- [ ] **Sentence Rhythm Varies**: No 3+ same-length consecutive sentences (1pt)
- [ ] **Sentence Length**: Average 15-20 words, none exceeds 40 (1pt)
- [ ] **Structural Variety**: Paragraph openers, list lengths, section lengths vary (1pt)
- [ ] **Code-to-Prose Ratio**: ≥ 40% code blocks (1pt)
- [ ] **Heading Discipline**: Max 3 levels, sentence case, one H2 per 200-400 words (1pt)
- [ ] **Hedging Limited**: Max 2 hedge phrases per 1000 words (1pt)
---
## Quality Gate Thresholds
| Score | Verdict | Action |
|-------|---------|--------|
| 48-52 | PASS | Ready to publish |
| 42-47 | GOOD | Minor improvements optional |
| 31-41 | NEEDS_WORK | Fix HIGH issues before publish |
| <31 | FAIL | Major revision required |
---
## Tool Recommendations by Language
| Language | Inline | Generator | Site Builder |
|----------|--------|-----------|--------------|
| TypeScript | TSDoc | TypeDoc | Docusaurus |
| Python | Google docstrings | Sphinx | MkDocs |
| Rust | /// comments | rustdoc | rustdoc |
| Go | Plain comments | godoc | godoc |
| Java | Javadoc | javadoc | javadoc |
---
## Anti-Pattern Detection Patterns
### OVER_MARKETING
**Severity**: HIGH
**Description**: Marketing hype buries quick start 64+ lines deep
**Detection**: Search for words: "amazing", "revolutionary", "best", "incredible", "powerful". Check if quick start appears after line 30.
**Impact**: 30-minute barrier to first success (should be <5 min)
**Fix**: Move quick start to first 20 lines, remove superlatives
### STALE_DOCS
**Severity**: HIGH
**Description**: No version tracking or last-updated dates
**Detection**: Missing: "Last Updated:", "Version:", "As of v". No YAML frontmatter with date/version.
**Impact**: Users don't know if docs apply to their version
**Fix**: Add version header with date
### MISSING_ERROR_RECOVERY
**Severity**: CRITICAL
**Description**: Happy path only, no troubleshooting
**Detection**: Missing sections: "Troubleshooting", "Common Issues", "FAQ", "Error"
**Impact**: 50%+ of support questions are error-related
**Fix**: Add troubleshooting section with top 5 errors
### PASSIVE_VOICE
**Severity**: MEDIUM
**Description**: Passive voice increases cognitive load
**Detection**: Pattern: "is/are/was/were + past participle". Examples: "is processed", "was created", "are stored"
**Impact**: 20-30% increased cognitive load
**Fix**: Convert to active voice: "The server processes" not "is processed by"
### FRAGMENTED_INFO
**Severity**: HIGH
**Description**: Same topic in multiple places
**Detection**: Duplicate headings, repeated explanations
**Impact**: Users get confused, docs become stale
**Fix**: Consolidate to single source of truth
### MISSING_JOURNEY
**Severity**: HIGH
**Description**: No clear user learning path
**Detection**: No "Getting Started" or navigation
**Impact**: Users don't know where to begin
**Fix**: Add persona-based entry points
### INCONSISTENT_FORMAT
**Severity**: MEDIUM
**Description**: Different structures per file
**Detection**: Varying heading styles, list formats
**Impact**: Cognitive overhead for readers
**Fix**: Apply consistent template
### ASSUMPTION_OVERLOAD
**Severity**: HIGH
**Description**: Unexplained prerequisites
**Detection**: Missing "Prerequisites" section
**Impact**: Beginners blocked by invisible barriers
**Fix**: Add explicit prerequisites checklist
### COPY_PASTE_DOCS
**Severity**: MEDIUM
**Description**: Duplicated content
**Detection**: 80%+ similar paragraphs
**Impact**: Maintenance nightmare, drift between copies
**Fix**: Extract to single source, link to it
### AI_HALLUCINATION_RISK
**Severity**: CRITICAL
**Description**: Undocumented features
**Detection**: Compare docs with source code, flag mismatches
**Impact**: Users try features that don't exist
**Fix**: Verify all documented features exist in code
---
*Documentation standards based on 73+ authoritative sources with 98% factual integrity*