spec_driven_develop

Why we included this project

Large rewrites and migrations tend to make coding agents improvise from scratch on each project, and this workflow exists to stop that. It's a set of Markdown skills that plug into Claude Code, Codex, OpenCode, Cursor, or any agent that reads custom skills, with no SDK or runtime dependencies. Every big change runs through the same loop: analyze the architecture first, decompose the work into phased tasks, track progress in GitHub Issues, PRs, and a MASTER.md index, then execute and archive the artifacts for traceability. Three workflows ship together: a spec-driven pipeline covering the whole lifecycle, a Deep Discuss flow for structured design discussion, and a findings-first code review for uncommitted changes and branch diffs. Adaptive control and a S.U.P.E.R. architecture health check keep execution on track. For teams that run AI agents beyond one-off prompts, a change this structured stays auditable, resumable across sessions, and much less likely to drift.