ai-devkit

About

Developers running several coding agents end up maintaining parallel config trees, losing session state, and re-explaining conventions to each tool. ai-devkit establishes one local .ai-devkit.json as the source of truth, from which an init wizard writes per-agent directories (.claude/, .cursor/, and peers), nine built-in skills, and a docs/ai/ phase structure spanning requirements through testing. The CLI treats running agents as infrastructure: list and inspect sessions, watch them in a live TUI console, and inject prompts, logs, or test output into sessions with optional stdin and wait semantics, or drive them from a phone over a Telegram channel. Skills follow a dev-lifecycle spine (verify, memory, tdd, structured-debug, dev-commit, and more), and additional skills install from a catalog including Anthropic, Vercel, Supabase, Microsoft, and Google publishers. Everything runs locally — SQLite memory, no telemetry — under an MIT license.