A standard structure, naming conventions, and content expectations for the docs/ directory across all projects. Designed to serve both human contributors and AI coding agents.
Common Docs uses best practices from Anthropic and OpenAI to produce an LLM-friendly documentation approach.
This specification defines how to organize project documentation using a progressive disclosure model:
- Context → Architecture → Implementation → Operations
Every file has a purpose. Every directory has an entry point. Documentation is versioned alongside code — not scattered across wikis or buried in Notion.
The Specification describes the directory structure and how to use the format.
- Copy the
docs/structure into your project - Fill in
docs/index.mdwith your project summary - Work through sections in order:
context/→architecture/→ the rest - Add the reference block to your
AGENTS.mdorCLAUDE.md
docs/
├── index.md # Master entry point
├── context/ # Why we exist, who we serve, domain vocabulary
├── architecture/ # Current system design
├── adr/ # Architecture Decision Records
├── plans/ # Future work, roadmap, feature specs
├── api/ # API reference
├── configuration/ # Configuration reference
├── integrations/ # Third-party integrations
├── security/ # Security posture and practices
├── guides/ # How-to guides for humans and agents
├── operations/ # Production operations
├── migrations/ # Database and data migrations
└── changelog/ # Release history
- Context before detail — Build understanding progressively
- Every directory has an
index.md— Navigation without scanning - One concern per file — Split files over ~300 lines
kebab-casenaming — No spaces, lowercase- Frontmatter on every file — title, description, status, updated
- Plans describe future. ADRs record past. Architecture describes present. Never conflate them
docs/is curated, not a wiki — Stale docs are deleted- Reference
docs/index.mdfromAGENTS.md— Agents start there
This project includes three AI agent skills for working with Common Docs:
Scaffolds the standard docs/ directory structure in a project. Creates all section directories and stub files with correct frontmatter.
Use when: Setting up docs, scaffolding documentation, creating the docs structure, or adding docs to a project.
Usage: /common-docs-scaffold
The skill will:
- Check what exists in the project root
- Ask which sections to create (default: all)
- Create files using the content spec in
references/specification.md - Add the documentation reference table to
AGENTS.mdorCLAUDE.md
Writes the actual content of a specific documentation file. Runs a structured interview for the target file type, explores the codebase to fill in what it can, then produces a complete document meeting the spec's "Must contain" requirements.
Use when: Writing, filling in, or populating a doc — e.g. "write the architecture overview", "fill in the threat model", "write the getting started guide", "document our API endpoints", "write an ADR for X".
Usage: /common-docs-write
The skill will:
- Identify the target file and look up its content requirements in the spec
- Ask a set of targeted questions specific to that file type
- Explore the codebase to fill gaps without bothering you
- Write the complete file — no TODOs or placeholders
- Offer to update related files (index, ADR index, etc.)
Finds the right documentation file for any topic using the standard Common Docs structure. Maps questions and topics to canonical file paths.
Use when: Looking for where to find something, where to document something, which file covers a topic, or before reading/writing docs to know the correct location.
Usage: /common-docs-find
The skill will identify the topic from your question and map it to the correct file path using its lookup table.
Skills are available in two locations:
skills/— Packaged skills for distribution (referenced byskill.json)
Common Docs is available as a plugin Automatic.
This specification is versioned and maintained.
MIT License - see LICENSE