Comparing SDD Frameworks: spec-kit vs OpenSpec vs get-shit-done

I did try https://kiro.dev/ and https://antigravity.google/ but they were so slow as to be unusable.

Notes on three Spec Driven Development (SDD) frameworks I’ve been comparing:

• github/spec-kit
• Fission-AI/OpenSpec
• gsd-build/get-shit-done

It retrospect I wish I generated unique ANTHROPIC_API_KEY keys for these experiments to get a better understanding of the costs.

spec-kit

Using https://github.com/kaihendry/ai-check-guardrails

Not a perfect experience with Claude code, nonetheless I thought it did well:

• It does a branch when you do a /speckit-specify which makes me think this is the most “multi-player” ready
• It kindof auto commits
• There does seem to be some community
• default constitution read my mind

OpenSpec

https://github.com/kaihendry/asaguard

• https://github.com/kaihendry/asaguard/blob/main/openspec/config.yaml doesn’t actually seem to work, UPDATE: my schema was wrong https://github.com/Fission-AI/OpenSpec/blob/main/docs/customization.md (+1 helpful discord community)
• not sure why you have to /opsx:archive manually

get-shit-done

https://github.com/kaihendry/ai-siem-endpoint

• it’s use of multiple agents makes me think this is pretty token heavy
• seems the most bureaucratic, but that’s maybe because my expectations were that it shouldn’t be!
• discord feels dead
• i am confused by the numbering, e.g. /gsd-plan-phase 1 and /gsd-execute-phase 1

Takeaways

My “own way” of using beads and vibing doesn’t seem eclipsed in my mind by SDD. The frameworks may help clarify some large features, but normally I’ve already broken them down already in my mind.

I don’t feel any of these frameworks captured the initial prompting very well and it feels wrong to manually edit their outputs.