oss-write-docs

# Write Docs

Contribute documentation that a maintainer would write — accurate, audience-aware, and verified against the actual code. Docs contributions have different mechanics than code PRs: you need to understand who reads the docs, verify every claim against source code, and test every example.

## Purpose

Documentation contributions are undervalued by contributors and overvalued by maintainers. Most contributors skip docs because they don't feel like "real" contributions. But outdated docs cause more user frustration than most bugs, and maintainers rarely have time to keep docs current. This skill guides you through finding real documentation gaps, verifying accuracy against the code, and writing docs that match the repo's voice and style.

## When to Use

- You've found outdated setup instructions (you followed them and they didn't work)
- Public API methods or functions lack documentation
- README assumes knowledge that new users don't have
- Code examples in docs don't run or produce wrong output
- Error messages reference docs that don't exist
- **NOT** for writing docs for your own project
- **NOT** for adding inline code comments — that's part of code contributions
- **NOT** when the repo has no docs infrastructure at all — discuss with maintainers first

## Prerequisites

- Repo forked, cloned, and set up (from `oss-prep-to-contribute`)
- You've actually tried to USE the docs and found problems (not just skimming)
- Understanding of the area you're documenting (or willingness to build it)

## Process

### 1. Identify documentation gaps

The best doc contributions come from actually using the docs and finding where they fail.

```bash
# Find docs files
find . -name "*.md" -not -path "*/node_modules/*" -not -path "*/.git/*" | head -30
ls docs/ doc/ documentation/ wiki/ 2>/dev/null

# Check for docs framework
ls docusaurus.config.* mkdocs.yml conf.py book.toml .readthedocs.yml 2>/dev/null

# Find broken links in docs
grep -rn '\[.*\](.*\.md)' docs/ README.md | while read line; do
  file=$(echo "$line" | sed -n 's/.*(\([^)]*\.md\)).*/\1/p')
  [ -n "$file" ] && [ ! -f "$file" ] && echo "Broken link: $line"
done

# Find code examples and check if they reference current APIs
grep -rn '```' docs/ README.md --include="*.md" | head -20
```

**High-value doc gaps** (prioritize these):
- Setup instructions that don't work (you tried them)
- API methods with no docs but active usage in the codebase
- Examples that reference renamed or removed functions
- Config options documented nowhere
- Error messages that say "see docs" but the docs don't exist

**Low-value doc gaps** (skip these):
- Typos in low-traffic pages
- Style inconsistencies (unless extreme)
- Adding docs for internal/private APIs

### 2. Understand the docs system

Every repo has documentation conventions. Learn them before writing.

```bash
# Check how docs are built
cat package.json 2>/dev/null | grep -i "doc\|docs\|build.*doc"
cat Makefile 2>/dev/null | grep -i doc
cat pyproject.toml 2>/dev/null | grep -i doc

# Read the docs contribution guide (if separate from CONTRIBUTING.md)
cat docs/CONTRIBUTING.md docs/contributing.md 2>/dev/null

# Check the docs config
cat docusaurus.config.js mkdocs.yml conf.py 2>/dev/null | head -50
```

Document:
- **Docs framework**: Docusaurus, MkDocs, Sphinx, mdBook, rustdoc, JSDoc, plain markdown?
- **File structure**: how are docs organized? (by topic, by API, by tutorial?)
- **Build process**: how to preview locally?
- **Style guide**: formal or casual? second person ("you") or imperative? code-heavy or prose-heavy?
- **Versioning**: are docs versioned alongside releases?

### 3. Verify the gap by reading code

Every documentation claim must be traceable to actual code. Don't write docs from memory or assumption.

```bash
# If documenting a function, read its implementation
grep -rn "function functionName\|def functionName\|func functionName" src/ --include="*.ts" --include="*.py" --include="*.go"

# If documenting config options, find where they're read
grep -rn "config\.\|getenv\|process\.env\." src/ --include="*.ts" --include="*.py"

# If documenting CLI flags, find the argument parser
grep -rn "argparse\|commander\|clap\|flag\." src/ --include="*.py" --include="*.ts" --include="*.rs" --include="*.go"
```

For each claim you plan to write:
- Find the source code that implements the behavior
- Note the file and line number (for your own reference, not for the docs)
- Check if the behavior has changed since the last docs update

```bash
# When was this doc last updated vs when was the code last changed?
git log --oneline -5 -- "docs/path/to/page.md"
git log --oneline -5 -- "src/path/to/implementation.ts"
```

### 4. Thinking gate — user explains the audience

> "Before writing anything:
> 1. Who reads this documentation? (New users? API consumers? Contributors? Ops/deploy engineers?)
> 2. What do they already know when they arrive at this page?
> 3. What are they trying to accomplish? (Not 'learn about X' — what task are they doing?)
> 4. What's the single most important thing this page should communicate?"

Wait for their answer. If they say "everyone" or "developers in general," push back: "Look at the existing docs — who is the implicit audience? What level of knowledge do they assume?"

### 5. User writes the docs

The user drafts the documentation. The LLM helps with:
- Verifying claims against source code ("you wrote that this accepts 3 arguments, but the function signature shows 4")
- Checking code examples actually work
- Matching the repo's documentation voice and style
- Pointing to similar pages as format reference

**What the LLM DOES**:
- Fact-check every claim against the source code
- Test code examples mentally (or by running them if possible)
- Review for clarity — flag jargon the target audience wouldn't know
- Point to existing docs pages as style reference

**What the LLM DOES NOT DO**:
- Write the documentation for the user
- Add promotional language ("this amazin