md-docs（医学博士文档）

Markdown Documentation Management

Overview

Manage project documentation for Claude Code workflows including context files, READMEs, and agent instructions. This skill provides structured automation for maintaining accurate, up-to-date documentation that aligns with actual codebase structure and functionality. Use this skill when initializing new projects, updating existing documentation, or ensuring context files accurately reflect current code.

The skill emphasizes verification and validation over blind generation—analyze the actual codebase structure, file contents, and patterns before creating or updating documentation. All generated content should be terse, imperative, and expert-to-expert rather than verbose or tutorial-style.

Prerequisites

Before using any documentation workflow, verify basic project structure:

git rev-parse --git-dir

Ensure the output confirms you are in a git repository. If not initialized, documentation workflows may still proceed but git-specific features will be skipped.

For update workflows, verify target files exist:

ls -la CLAUDE.md AGENTS.md DOCS.md README.md CONTRIBUTING.md

Check which files are present before attempting updates. Missing files will show errors, which helps identify what needs initialization. Note that DOCS.md is optional and only relevant for projects with APIs or public interfaces. CONTRIBUTING.md is optional and only relevant if the repo already defines contribution guidance.

Update Context Files

Verify and fix CLAUDE.md, AGENTS.md, and optionally DOCS.md and CONTRIBUTING.md against the actual codebase. This workflow reads existing context files, analyzes the codebase structure, identifies discrepancies, and updates documentation to match reality. DOCS.md and CONTRIBUTING.md are only processed if they exist in the repository.

Workflow Steps

Parse Arguments

Support the following arguments:

• --dry-run: Show what would change without writing files
• --preserve: Keep existing content structure, only fix inaccuracies
• --thorough: Perform deep analysis of all files (slower but comprehensive)
• --minimal: Quick verification focusing on high-level structure only

Verify Git Repository

Confirm working directory is a git repository. If not, warn the user but proceed with limitations (cannot analyze git history or branches).

Read Existing Context Files

Read current CLAUDE.md, AGENTS.md, DOCS.md, and CONTRIBUTING.md (if present) contents:

cat CLAUDE.md
cat AGENTS.md
cat DOCS.md         # if exists
cat CONTRIBUTING.md # if exists

Parse the structure and extract documented information including:

• Project description and purpose
• File structure and organization
• Build and test commands
• Custom tooling or scripts
• Agent configurations and triggers
• API endpoints and methods (from DOCS.md)
• Function signatures and parameters (from DOCS.md)
• Type definitions and interfaces (from DOCS.md)
• Contribution workflow and guidelines (from CONTRIBUTING.md)
• Code review process and standards (from CONTRIBUTING.md)

Analyze Codebase

Scan the project to gather accurate information:

• Directory structure (ls -la, tree if available)
• Package configuration (package.json, pyproject.toml, Cargo.toml, etc.)
• Build scripts and commands
• Test frameworks and configurations
• README badges and metadata

For --thorough mode, also analyze:

• File content patterns (imports, exports, interfaces)
• Code organization conventions
• Dependency relationships

Identify Discrepancies

Compare documented information against actual codebase:

• Outdated file paths or structure
• Incorrect build commands
• Missing or removed features
• Deprecated dependencies
• Stale agent configurations
• Outdated API endpoints or routes (DOCS.md)
• Changed function signatures (DOCS.md)
• Modified type definitions (DOCS.md)
• Outdated contribution guidelines (CONTRIBUTING.md)
• Stale branch naming or PR process (CONTRIBUTING.md)

Create Backups

Before overwriting, create backup files:

cp CLAUDE.md CLAUDE.md.backup
cp AGENTS.md AGENTS.md.backup
test -f DOCS.md && cp DOCS.md DOCS.md.backup
test -f CONTRIBUTING.md && cp CONTRIBUTING.md CONTRIBUTING.md.backup

Update Context Files

Write corrected versions maintaining the existing structure when --preserve is used, or reorganizing for clarity when not. For --dry-run, display the diff without writing:

diff -u CLAUDE.md.backup CLAUDE.md

Generate Report

Display a summary of changes.

When DOCS.md exists:

✓ Updated CLAUDE.md
  - Fixed outdated build command
  - Added new /api directory to structure

✓ Updated AGENTS.md
  - Updated test-runner trigger pattern

✓ Updated DOCS.md
  - Fixed outdated endpoint path /api/v1/users
  - Updated function signature for createUser()

✓ Updated CONTRIBUTING.md
  - Updated branch naming convention
  - Fixed outdated PR template reference

When optional files are absent:

✓ Updated CLAUDE.md
  - Fixed outdated build command

✓ Updated AGENTS.md
  - Updated test-runner trigger pattern

⊘ DOCS.md not found (skipped)
⊘ CONTRIBUTING.md not found (skipped)

For the complete update context files workflow with verification strategies, diff examples, and edge cases, refer to ./references/update-agents.md.

Update README

Generate or update README.md based on project structure, package metadata, and codebase analysis. This workflow creates comprehensive, accurate READMEs that reflect the actual state of the project.

Workflow Steps

Parse Arguments

Support the following arguments:

• --dry-run: Preview README content without writing
• --preserve: Keep existing sections, only update outdated information
• --minimal: Generate minimal README (title, description, installation, usage)
• --full: Generate comprehensive README with all optional sections

Analyze Project Structure

Gather information from multiple sources:

# Package metadata
cat package.json
cat pyproject.toml
cat Cargo.toml

# Git information
git remote get-url origin
git describe --tags

# Directory structure
ls -la

Extract:

• Project name and description
• Version number
• Repository URL
• License
• Dependencies
• Scripts/commands

Read Existing README

If README.md exists and --preserve is used:

cat README.md

Parse existing sections to preserve custom content while updating technical details.

Read CONTRIBUTING.md (Optional)

If CONTRIBUTING.md exists, read it and treat it as the source of truth for contribution guidance. Do not duplicate detailed contribution steps in README; link to CONTRIBUTING.md instead.

test -f CONTRIBUTING.md && cat CONTRIBUTING.md

Create Backup

Before overwriting existing README:

cp README.md README.md.backup

Generate README Content

Create structured content with appropriate sections:

• Title and badges (version, license, build status)
• Description (concise project summary)
• Installation (package manager commands)
• Usage (basic examples)
• Development (build, test, lint commands)
• Contributing (if applicable; link to CONTRIBUTING.md when it exists)
• License (based on package metadata)

For --minimal mode, include only title, description, installation, and usage.

For --full mode, also include:

• API documentation
• Examples directory listing
• Deployment instructions
• Troubleshooting section
• Credits and acknowledgments

Write README

Save the generated content. For --dry-run, display without writing.

Generate Report

Display summary:

✓ Updated README.md
  - Added installation section
  - Updated build commands to match package.json
  - Added badges for license and version

For the complete update README workflow with section templates, metadata extraction strategies, and formatting examples, refer to ./references/update-readme.md.

Update CONTRIBUTING

Update CONTRIBUTING.md based on current codebase tooling and workflows. This workflow only runs when CONTRIBUTING.md already exists in the repository. If CONTRIBUTING.md is absent, skip this workflow entirely—do not auto-create contribution guidelines.

Prerequisite Check

Before proceeding, verify the file exists:

test -f CONTRIBUTING.md && echo "exists" || echo "missing"

If missing, report to the user and stop. Do not create CONTRIBUTING.md unless explicitly requested.

Workflow Steps

Parse Arguments

Support the following arguments:

• --dry-run: Show what would change without writing files
• --preserve: Maximum preservation; only fix broken commands/links
• --thorough: Deep analysis; verify all links and commands work

Read Existing CONTRIBUTING.md

cat CONTRIBUTING.md

Parse the document structure:

• Section headings and organization
• Code blocks with commands
• URLs and file path references
• Mentioned tooling (test runners, linters, formatters)

Gather Codebase Intelligence

Detect current tooling and compare against documented content:

• Package manager (npm, pnpm, yarn, bun) from lock files
• Available scripts from package.json/Makefile/justfile
• Branch conventions (main vs master)
• Linting/formatting tools in use

Identify Discrepancies

Compare documented information against actual codebase:

• Outdated CLI commands (npm → pnpm)
• Incorrect branch references (master → main)
• Broken links to issue templates or docs
• Stale tooling references (Jest → Vitest)

Update Content

Fix technical inaccuracies while preserving:

• Contribution policies (CLA, DCO, licensing)
• Review processes and expectations
• Code of conduct references
• Governance decisions

Generate Report

Display summary:

✓ Updated CONTRIBUTING.md
  - Fixed package manager: npm → pnpm
  - Corrected branch reference: master → main
  - Updated test command

⊘ Policy sections preserved (CLA, review process)

For the complete update CONTRIBUTING workflow with verification strategies and examples, refer to ./references/update-contributing.md.

Initialize Context

Create project-specific CLAUDE.md from scratch based on codebase analysis. This workflow is ideal for new projects or repositories lacking context documentation.

Workflow Steps

Parse Arguments

Support the following arguments:

• --dry-run: Preview generated content without writing
• --minimal: Create minimal context file (project description, structure)
• --full: Create comprehensive context file with all relevant sections

Verify No Existing CLAUDE.md

Check if CLAUDE.md already exists:

test -f CLAUDE.md && echo "exists" || echo "missing"

If exists, warn the user and suggest using the update workflow instead. Allow override with --force flag.

Analyze Project

Gather comprehensive information:

• Language and framework (detect from files and package configs)
• Directory structure and organization patterns
• Build system (npm, cargo, poetry, gradle, etc.)
• Test framework (jest, pytest, cargo test, etc.)
• Linting and formatting tools
• Environment variables or configuration files

Generate CLAUDE.md Content

Create structured sections:

# Context

Brief project description and purpose.

## Structure

Directory organization and key files.

## Build

Commands for building the project.

## Test

Commands for running tests.

## Development

Conventions, patterns, and workflows.

Adapt sections based on project type. For --minimal, include only Context and Structure. For --full, add all applicable sections including deployment, troubleshooting, and custom tooling.

Write CLAUDE.md

Save generated content. For --dry-run, display without writing.

Generate Report

Display summary:

✓ Created CLAUDE.md
  - Detected Next.js project
  - Added npm scripts from package.json
  - Documented project structure
  - Added testing section for Jest

For the complete initialize context workflow with language-specific templates, detection strategies, and customization options, refer to ./references/init-agents.md.

DOCS.md Initialization

DOCS.md is optional and not created by default. Create DOCS.md manually when the project has:

• Public API endpoints requiring documentation
• Exported functions or classes intended for external use
• Complex type definitions users need to understand

The update context workflow will suggest creating DOCS.md if it detects significant APIs without corresponding documentation.

Common Patterns

Shared conventions and patterns used across all documentation workflows.

Argument Parsing

Standard arguments supported across workflows:

• --dry-run: Preview changes without writing files
• --preserve: Maintain existing structure, only fix inaccuracies
• --minimal: Generate minimal documentation
• --thorough/--full: Generate comprehensive documentation
• --force: Override safety checks

Parse arguments from user input and set appropriate flags for workflow execution.

Backup File Handling

Always create backups before overwriting existing files:

cp CLAUDE.md CLAUDE.md.backup
cp AGENTS.md AGENTS.md.backup
test -f DOCS.md && cp DOCS.md DOCS.md.backup                 # only if exists
test -f CONTRIBUTING.md && cp CONTRIBUTING.md CONTRIBUTING.md.backup  # only if exists

Inform the user when backups are created:

Created backup: CLAUDE.md.backup
Created backup: AGENTS.md.backup
Created backup: DOCS.md.backup (optional file)
Created backup: CONTRIBUTING.md.backup (optional file)

Never delete backups automatically. Let users manage backup cleanup manually. Note that DOCS.md and CONTRIBUTING.md are optional—skip backup and update operations if they don't exist.

Writing Style

Documentation should follow these conventions:

• Terse: Omit needless words, lead with the answer
• Imperative: Use command form ("Build the project") not descriptive ("The project is built")
• Expert-to-expert: Skip basic explanations, assume competence
• Scannable: Use headings, lists, and code blocks for easy navigation
• Accurate: Verify all commands and paths against actual codebase

Good:

## Build

Build the project:

\`\`\`bash
npm run build
\`\`\`

Run tests:

\`\`\`bash
npm test
\`\`\`

Bad:

## Building the Project

In order to build the project, you will need to use the npm build command. This command will compile all of the TypeScript files and generate the output in the dist directory. First, make sure you have installed all dependencies by running npm install.

Report Formatting

After completing operations, display a clear summary:

✓ Updated CLAUDE.md
  - Fixed build command
  - Added new directory structure

✓ Updated README.md
  - Added installation section
  - Updated badges

✓ Updated DOCS.md
  - Updated API endpoint documentation
  - Fixed function signature

✗ AGENTS.md not found
  - Skipped update

⊘ DOCS.md not found
  - Skipped (optional file)

⊘ CONTRIBUTING.md not found
  - Skipped (optional file)

Use checkmarks (✓) for successful operations, crosses (✗) for failed operations, and ⊘ for skipped optional files. Include indented details showing specific changes made.

File Detection

Detect project type and structure by checking for characteristic files:

# Node.js/JavaScript
test -f package.json

# Python
test -f pyproject.toml || test -f setup.py

# Rust
test -f Cargo.toml

# Go
test -f go.mod

Use detection results to customize documentation templates and commands.

Metadata Extraction

Read package configuration files to extract accurate metadata:

# Node.js
cat package.json | grep -E '"name"|"version"|"description"'

# Python
cat pyproject.toml | grep -E 'name|version|description'

Parse JSON or TOML appropriately to extract values. Never hardcode or guess metadata when it can be read directly from configuration files.

Additional Resources

For detailed workflows, examples, and implementation guidance, refer to these reference documents:

• ./references/update-agents.md - Complete context file update workflow including verification strategies, diff generation, and discrepancy detection
• ./references/update-readme.md - Complete README update workflow including section templates, metadata extraction, and formatting conventions
• ./references/update-contributing.md - Complete CONTRIBUTING.md update workflow including scope, templates, and validation (only when CONTRIBUTING.md exists)
• ./references/init-agents.md - Complete context initialization workflow including language-specific templates, detection strategies, and customization options

These references provide implementation details, code examples, and troubleshooting guidance for each workflow type.