comet

Why we included this project

Comet is a harness for teams that want to hand coding agents substantial, multi-file work rather than babysitting every edit. It wraps spec-driven development (built on OpenSpec and Superpowers) into a five-phase, resumable workflow, and adds an evaluation harness that scores skills with Rubric, Pass@k, and Pass^k metrics instead of gut feeling. The phase-guarded state machine is the part that stands out: the workflow only advances when verification evidence checks out, so an agent can't simply claim it finished. That combination helps when agents stall mid-task, and it gives you a way to prove a prompt or skill change actually improved results. Skill authors also get to package composed skills and ship them to several coding platforms with a single command.