I built aidd - the AI Development Director - to make AI coding agents write features unattended. Its audit mode checks a codebase in over twenty categories, including security and performance, then gives me actionable findings. I have a feeling I’ll use that mode at least as much as the feature-building mode, but we’ll see…
That feature-building mode is why aidd exists at all. For the past month I’ve been building individual wrappers for three AI coding assistants (Claude Code, KiloCode, and OpenCode). Each wrapper could drive its own AI tool automatically, but they were separate projects and I was maintaining the same logic in three places. Today they became one system.
You define features in structured JSON files, point aidd at whichever AI coding tool you choose, and let it run. Most of the work went into the tedious parts: spotting when an agent gets stuck, nudging it back on track, retrying after failures, and knowing when to give up.