Devdocs CLI — AI skill for Claude Code

devdocs-cli

**NOTE: This was just a (failed) experiment. Probably https://backlog.md is what you want instead. The [beads](https://github.com/steveyegge/beads) issue tracker is a massive PITA, do not recommend**

Structure your AI coding projects. Sets up issue tracking and documentation patterns that help agents maintain context across sessions.

Install

First, install the dependencies:

• beads (bd) for issue tracking
• pi for the coding agent and slash commands

Then install devdocs-cli:

go install github.com/mrexodia/devdocs-cli@latest

Usage

cd your-project
devdocs init

This sets up:

• Beads (bd) for issue tracking with no-git-ops config
• devdocs/ directory for epics and reference documentation
• AGENTS.md with the methodology (appends if file exists)
• pi hooks for bd prime injection and slash commands

What is this?

A hybrid approach combining:

• bd (beads) - Dependency-aware issue tracking that lives in your repo
• devdocs/ - Epic tracking with design documents and archived learnings

When to Use What

Scope	Approach
Single-session work	Just do it
Multi-session issue	bd create
Large initiative with phases	Epic (devdocs/ + bd epic)
Permanent reference knowledge	devdocs/.md

The Methodology

1. Design First

Don't jump into coding. Research first in Claude/ChatGPT/Grok, then consolidate into `design.md`:

• Problem statement and goals
• API design or architecture
• Key decisions and rationale
• References to documentation

2. TDD for AI

Set up the test harness *before* implementation:

• Define what tests need to pass
• Use "golden master" testing where possible (compare against known-good output)
• Ensure make test or equivalent produces actionable feedback
• Then tell the agent: "work until the tests pass"

3. Scaf