// the find
Fission-AI/OpenSpec
Spec-driven development (SDD) for AI coding assistants.
OpenSpec is a CLI plus slash-command workflow that makes an AI coding assistant write a proposal, spec, design doc, and task checklist before touching code, then archives it once implemented. It's aimed at teams and solo devs who've been burned by an agent inventing requirements mid-session, and it plugs into 30+ existing tools (Cursor, Copilot, Codex, etc.) rather than requiring its own IDE.
Specs are plain Markdown with WHEN/THEN scenarios instead of a bespoke DSL, so there's nothing new to parse or learn and diffs stay readable in a normal PR review. It integrates into whatever tool you already use by generating the right slash-command syntax per tool, instead of locking you into one vendor the way Kiro does. The new 'Stores' feature lets a platform team own specs in their own repo and have product-repo agents read them read-only over git push, which is a real answer to the cross-repo spec problem rather than a wiki nobody updates.
The whole value proposition rests on the AI actually following the spec during /opsx:apply, and there's no enforcement mechanism in the tool itself — it's still trusting model discipline, and the README's own advice to use 'high-reasoning models' like Opus 4.7 for decent results is an implicit admission that weaker models will drift from the plan. Telemetry is on by default (command names + version), which is a minor flag but still something to check before rolling out org-wide. 'Stores' is explicitly beta, so the cross-repo pitch that's doing a lot of the marketing work right now isn't the stable core of the product yet. It's also a layer of ceremony — proposal, specs, design, tasks, archive — that's overkill for a one-off script or a solo side project, despite the README claiming it scales down fine.