A contract-based, agent-first documentation specification for AI-assisted software development.
Humans steer. Agents execute.
Agent Doc Stack defines:
- Canonical doc locations with progressive disclosure
- A ~100-line agent table of contents that points to deep docs
- Planning as code with versioned execution plans
- Writing standards optimized for token efficiency
- Enforcement rules to prevent documentation drift
This is a specification, not a template or a tool.
Works with Claude Code, Codex CLI, Cursor, and Copilot Agent.
| File | Purpose |
|---|---|
agent-doc-stack.md |
The authoritative v2.0 specification — single source of truth for all structure, rules, and canonical file locations |
initialization-prompt.md |
Reusable prompt to initialize or normalize a repo's documentation using the spec |
- Open the target repository in your coding agent
- Run
initialization-prompt.md - The agent fetches and reads
agent-doc-stack.mddirectly from GitHub - Answer clarifying questions if the agent requests missing information
The agent will follow the spec to generate, update, or clean documentation — without inventing behavior.
You do not need to copy this repository into your project.
As coding agents take on more responsibility, documentation must be:
- Agent-legible — if the agent can't read it in-repo, it doesn't exist
- DRY — define once, link everywhere; no duplicated facts
- Enforceable — contracts, not prose
- Low-noise — minimum tokens, maximum signal
- Current version: v2.0 (defined in
agent-doc-stack.md) - Patch: clarify wording, no behavior change
- Minor: add rules, backward compatible
- Major: breaking changes
Pin to a specific version for strict stability.