Skip to content

Agent Doc Stack is a lightweight, agent-first documentation architecture that defines where truth lives and how coding agents generate, update, and enforce docs with minimal tokens and zero drift.

Notifications You must be signed in to change notification settings

jeronimodeleon/agent-doc-stack

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 

Repository files navigation

Agent Doc Stack

Agent Doc Stack is a lightweight, agent-first documentation architecture that defines where truth lives and how coding agents generate, update, and enforce docs with minimal tokens and zero drift.

This repository contains the Agent Doc Stack v1.0 specification and a single initialization prompt used to apply the framework to any codebase.


What This Is

Agent Doc Stack is a contract-based documentation specification for AI-assisted software development.

It defines:

  • Canonical documentation locations
  • Strict rules for how agents read, write, and update docs
  • Writing standards optimized for token efficiency
  • Enforcement rules to prevent documentation drift

This is a specification, not a template or a tool.

It works across modern coding agents including Claude Code, Codex CLI, Cursor, and Copilot Agent.


Canonical Repository Structure

Agent Doc Stack standardizes documentation into a small, predictable set of files.

README.md
ARCHITECTURE.md

docs/
  app-workflows.md
  dev-workflows.md
  agent-tools.md        # optional (MCP / tool access)
  features/
    _template.md
    <feature>.md

AGENTS.md
CLAUDE.md

plans/
  _template.md
  YYYY-MM-DD-short-title.md

Each file has a single, well-defined responsibility.

Agents are explicitly instructed where to read, write, and update documentation as code changes.


Repository Contents

This repository includes:

agent-doc-stack.md

The authoritative Agent Doc Stack v1.0 specification.

initialization-prompt.md

A reusable prompt that instructs a coding agent to initialize or normalize documentation in a repository using the spec.


How To Use

You should already be inside the repository whose documentation you want to create or fix.

Typical workflow:

  1. Open the target repository in your coding agent.
  2. Run the initialization-prompt.md.
  3. Allow the agent to fetch and read the Agent Doc Stack specification directly from GitHub.
  4. Answer clarifying questions if the agent requests missing information.

The agent will:

  • Treat the current folder as the repository root
  • Follow the Agent Doc Stack rules exactly
  • Generate, update, or clean documentation without inventing behavior
  • Ask questions when information is unclear or unavailable

You do not need to copy this repository into your project.


The Specification

The canonical framework lives in:

agent-doc-stack.md

All structure, rules, and behavior are defined there.

This file is the single source of truth.


Why This Exists

As coding agents take on more responsibility, documentation must be:

  • Predictable
  • Enforceable
  • Low-noise
  • Safe from hallucination and drift

Agent Doc Stack treats documentation as a system contract, not prose.


Versioning

Agent Doc Stack follows semantic versioning.

  • The current version is defined in agent-doc-stack.md
  • Patch releases clarify wording without changing behavior
  • Minor releases may add rules while remaining backward compatible
  • Major releases may introduce breaking changes

Repositories that require strict stability should pin behavior to a specific version.

About

Agent Doc Stack is a lightweight, agent-first documentation architecture that defines where truth lives and how coding agents generate, update, and enforce docs with minimal tokens and zero drift.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published