AI Dev Effectiveness — AI skill for Claude Code

ai-dev-effectiveness


Measure how much AI co-programming actually accelerates software delivery — on any git repo.

`ai-dev-effectiveness` reads your git history, detects which commits were co-authored by AI coding agents (Claude, Copilot, Cursor, Codex, Aider, …), and produces an interactive HTML report (and a structured JSON sidecar) comparing your real delivery against a hypothetical traditional team. It triangulates three independent estimators — top-down specialist roles, bottom-up per-commit formulas, and an AI judge that reads each diff — so the productivity multipliers are credible, not just plausible.

Quickstart

# Install (macOS):
brew install pipx && pipx ensurepath
pipx install git+https://github.com/denn-gubsky/ai-dev-effectiveness

# Pick a folder DEDICATED to running analyses — NOT one of your project repos.
mkdir -p ~/dev-effectiveness && cd ~/dev-effectiveness

# One-time: install the bundled subagents (effort-judge + roles-architect).
ai-dev-effectiveness init-judge

# One-time per target: have Claude survey the repo and propose specialist roles.
ai-dev-effectiveness suggest-roles ~/work/your-repo --apply

# Every run: just go. Config auto-loads, JSON sidecar emits by default.
ai-dev-effectiveness analyze ~/work/your-repo --judge claude-cli \
    --team "1 developer + Claude Code (Opus 4.7)"
open ./your-repo/effectiveness-report.html

That's the full happy-path. Each step is explained below.

The analyzer **never writes inside the target repo** — bundled subagents, the judgment cache, and all reports live in your workspace. Each target gets its own subfolder so you can compare repos side by side.

What it measures

Three independent productivity multipliers triangulating the same question:

1. Top-down — what would a traditional specialist team need? Person-months by role.
2. Bottom-up — per-commit effort using lang