OpenSpec vs Spec Kit: workflows and code quality

The speaker compares OpenSpec and Spec Kit by building the same self-improving Python coding assistant with nine features and nine acceptance tests. Both runs used Claude Opus 4.7 in Claude Code with reasoning effort set to low. The results describe these specific runs rather than general performance benchmarks.

OpenSpec centers on propose, apply and archive. Spec Kit adds a project constitution and a specify, plan, tasks, analyze and implement cycle, with a feature branch for each epic. The speaker used Specify 0.8.7.dev and identifies 0.8.6 as the tagged stable release at the time. Neither framework split the PRD into phases automatically.

The reported OpenSpec run took about two and a half hours, while Spec Kit took roughly five; recording overhead contributed to elapsed time. Both passed all nine acceptance tests. The speaker found Spec Kit’s planning and sandbox implementation more rigorous, while OpenSpec had stronger data durability measures.

The generated memory layers differed. OpenSpec used offline TF-IDF retrieval. Spec Kit attempted Moonshot embeddings, but the provider lacked the required endpoint, leaving semantic retrieval unavailable. Neither generated app included automatic retries for transient LLM errors, and neither framework chained its full workflow autonomously.