Documentation Standards

Self-Evolving Skill: This skill improves through use. If instructions are wrong, parameters drifted, or a workaround was needed — fix this file immediately, don't defer. Only update for real, reproducible issues.

Overview

Standards for writing markdown documentation optimized for both LLM consumption and conversion to professional PDFs using Pandoc. Ensures consistency across all documentation.

When to Use This Skill

Use when:

• Writing markdown documentation (README, skills, guides, specifications)
• Creating new skills that include markdown content
• Authoring content that may be converted to PDF
• Reviewing documentation for standards compliance

Core Principles

1. LLM-Optimized Documentation Architecture

Machine-Readable Priority: OpenAPI 3.1.0 specs, JSON Schema, YAML specifications take precedence over human documentation.

Why: Structured formats provide unambiguous contracts that both humans and LLMs can consume reliably. Human docs supplement, don't replace, machine-readable specs.

Application:

• Workflow specifications → OpenAPI 3.1.1 YAML in specifications/
• Data schemas → JSON Schema with examples
• Configuration → YAML with validation schemas
• Human docs → Markdown referencing canonical machine-readable specs

2. Hub-and-Spoke Progressive Disclosure

Pattern: Central hubs (like CLAUDE.md, INDEX.md) link to detailed spokes (skills, docs directories).

Structure:

CLAUDE.md (Hub - Essentials Only)
    ↓ links to
Skills (Spokes - Progressive Disclosure)
    ├── SKILL.md (Overview + Quick Start)
    └── references/ (Detailed Documentation)

Rules:

• Hubs contain essentials only (what + where to find more)
• Spokes contain progressive detail (load as needed)
• Single source of truth per topic (no duplication)

3. Markdown Section Numbering

Critical Rule: Never manually number markdown headings.

❌ Wrong:

## 1. Introduction

### 1.1 Background

### 1.2 Objectives

## 2. Implementation

✅ Correct:

## Introduction

### Background

### Objectives

## Implementation

Rationale:

• Pandoc's --number-sections flag auto-numbers all sections when generating PDFs
• Manual numbering creates duplication: "1. 1. Introduction" in rendered output
• Auto-numbering is consistent, updates automatically when sections reorganize
• Applies to ALL markdown: documentation, skills, project files, README files

Rule: If markdown might ever convert to PDF, never manually number headings. Use semantic heading levels (##, ###) and let tools handle numbering.

Standards Checklist

Use this checklist when creating or reviewing documentation:

Structure

• Follows hub-and-spoke pattern (essentials in main doc, details in references)
• Links to deeper documentation for progressive disclosure
• Single source of truth (no duplicate content across docs)

Markdown Formatting

• No manual section numbering in headings
• Semantic heading levels (##, ###, ####) used correctly
• Code blocks have language identifiers for syntax highlighting
• Links use markdown format [text](url), not bare URLs

Machine-Readable Content

• Workflows documented as OpenAPI 3.1.1 specs (when applicable)
• Data structures use JSON Schema (when applicable)
• Configuration uses YAML with validation (when applicable)
• Human docs reference canonical machine-readable specs

File Organization

• Documentation lives in appropriate location:
  • Global standards → docs/standards/
  • Skill documentation → skills/{skill-name}/references/
  • Project documentation → {project}/.claude/ or {project}/docs/
• Index files provide navigation (INDEX.md, README.md)

Related Resources

• ASCII Diagram Validation: ascii-diagram-validator - Validate ASCII diagrams in markdown
• Skill Architecture: See skill-architecture plugin for creating effective skills

Examples

Good Hub-and-Spoke Structure

Hub (CLAUDE.md):

## PDF Generation from Markdown

**Quick Start**: Use pandoc-pdf-generation skill

**Critical Rules**:

1. Never write ad-hoc pandoc commands
2. Always verify PDFs before presenting
3. See skill for detailed principles

Spoke (skill/SKILL.md):

• Quick start with examples
• Link to references/ for detailed documentation
• Progressive disclosure as needed

Good Machine-Readable Documentation

Workflow Specification (specifications/hook-prompt-capture.yaml):

openapi: 3.1.1
info:
  title: Hook Prompt Capture Workflow
  version: 1.0.0
paths:
  /capture-prompt:
    post:
      summary: Capture user prompt from hook
      # ... detailed spec

Human Documentation (README.md):

## Workflow

See [hook-prompt-capture.yaml](./specifications/hook-prompt-capture.yaml)
for complete workflow specification.

Quick overview: ...

Summary

Documentation standards ensure:

• Consistency across all workspace documentation
• LLM optimization through machine-readable formats
• Maintainability via hub-and-spoke + single source of truth
• PDF compatibility through proper markdown formatting

Follow these standards for all documentation.

---

Troubleshooting

Issue	Cause	Solution
Double section numbers in PDF	Manual numbering in markdown	Remove manual numbers, use --number-sections only
Broken links in PDF	Relative paths incorrect	Use repo-root paths for cross-document links
Code block no syntax color	Missing language identifier	Add language after opening triple backticks
Tables render poorly	Column widths too wide	Use shorter headers or pipe-table format
Hub doc too long	Too much detail in hub	Move details to spoke documents, link from hub
Duplicate content	Same info in multiple docs	Identify SSoT, remove duplicates, add links
YAML spec not rendering	Wrong file extension	Use .yaml extension for OpenAPI specs
Index navigation missing	No INDEX.md or README.md	Create navigation index in each directory

Post-Execution Reflection

After this skill completes, check before closing:

1. Did the command succeed? — If not, fix the instruction or error table that caused the failure.
2. Did parameters or output change? — If the underlying tool's interface drifted, update Usage examples and Parameters table to match.
3. Was a workaround needed? — If you had to improvise (different flags, extra steps), update this SKILL.md so the next invocation doesn't need the same workaround.

Only update if the issue is real and reproducible — not speculative.