prd-taskmaster

Why we included this project

Claude Code users who want their agent to actually build things, not just answer inline questions, can point this engine at a one-line goal and get a real pipeline out of it. It interviews you like a product manager, writes a PRD that gets scored against deterministic quality checks, and compiles the spec into a dependency-ordered task graph kept as plain repo files. The execution model is the interesting part: a task only counts as done when it produces verification evidence, and a SHIP_CHECK_OK token is emitted only when every gate passes, so 'done' means proven rather than claimed. That appeals to teams burned by AI coding tools that report success without proof, and the vendor-neutral design means the workflow survives switching between Claude, Codex, and Gemini. It is pre-alpha and under active development, so pin a version if you need stability, but the deterministic core has a substantial test suite behind it.