Epic framework for spec-driven development

The EPIC framework assesses whether specifications give AI coding agents enough precise, complete, and consistent information to implement features without guessing. It organizes 40 practices into 10 dimensions covering requirements, plans, and tasks, drawing on the ISO/IEC/IEEE 29148 requirements standard. In a study of 114 open-source spec-driven projects, the top-scoring third devoted 11.8% of commits to bug fixes, compared with 20.4% for the bottom third. This is an association, not proof that stronger specs cause fewer bugs, but it suggests that vague targets, undefined scope, missing requirement IDs, and unclear links between requirements and tests can have practical costs.

EPIC Score turns the framework into a free browser and command-line checker, available through ZeeSpec and usable with files from tools such as Spec Kit and Kiro, as well as plain PRDs and AGENTS.md. It scores specs from 0 to 30, identifies missing or partial practices, cites supporting text, flags vague wording, and suggests specific fixes. It can also produce reports and badges, create prompts for coding agents, and enforce a minimum score in CI. A clinic-booking spec with concrete timings and exclusions nevertheless scores just 6 because it lacks items such as acceptance criteria, requirement IDs, and a separate plan; sample specs range from 0 to 29. The checker is heuristic, favors English, and should be treated as a diagnostic checklist rather than a definitive verdict.