claude code skills frameworks, compared

GSD vs Superpowers

Two of the most popular Claude Code frameworks. Both fight the same enemy — context rot — with phased workflows and fresh subagents. But they make opposite bets on how much methodology to impose. Here's the honest comparison, and how to pick.

Short version: GSD is the lightweight, spec-driven one — keep contexts clean, stay out of your way. Superpowers is the opinionated one — it brings a full engineering methodology (TDD, disciplined debugging) and expects you to work its way. Neither is "better"; they suit different temperaments.

If you've run Claude Code on anything non-trivial, you've felt it: it starts sharp, then around the halfway mark of a long session the quality slides — requirements drift, corners get cut. That's context rot, and it's the problem both of these frameworks exist to solve. Both are free and open-source, both split work into phases, and both spawn fresh subagents with clean context per task instead of one long, degrading session. The difference is what they do inside that structure.

GSD (Get Shit Done)

Created by TÂCHES (gsd-build/get-shit-done), GSD is a lightweight meta-prompting and spec-driven system, and it's the more popular of the two by a wide margin (30k+ GitHub stars). Its whole philosophy is context hygiene: split a task into plan → execute → review, and give each phase its own clean context window and a single clear objective, so Claude never has to hold the entire project in a degrading buffer.

Superpowers

Created by Jesse Vincent / obra (obra/superpowers), Superpowers installs a set of structured "skills" into Claude Code — reusable mental models it invokes instead of improvising each time. Where GSD is minimal, Superpowers is a full engineering methodology with strong opinions: red-green-refactor TDD (tests must fail before you implement), a four-phase debugging process that demands root-cause analysis before any fix, and Socratic brainstorming to pin down requirements before code. Each task runs in a fresh subagent with a two-stage review — spec compliance first, then code quality.

Side by side

 GSD (Get Shit Done)Superpowers
Core ideaLightweight, spec-driven context engineeringOpinionated engineering methodology as skills
StructurePlan → execute → review, clean context per phaseBrainstorm → plan → execute, skills + subagents
MethodologyUnopinionated — no imposed practiceEnforces TDD + disciplined debugging
Context strategyFresh subagent, clean window per taskFresh subagent per task, two-stage review
WeightLight, fast to adoptHeavier, more prescriptive
CommunityVery large (30k+ stars)Growing, methodology-focused
Pick it ifYou want clean phases without a methodology imposedYou want rigor (tests-first, root-cause) enforced

Both are free and open-source and evolve quickly; check each repo for the current state before you commit.

So which one?

It comes down to how much opinion you want from your tools. Reach for GSD if you already have your own engineering habits and just want Claude to stop degrading over long sessions — it's the lighter touch and the bigger community. Reach for Superpowers if you want the guardrails — if "it forces me to write tests first and actually root-cause bugs" sounds like a feature, not friction. Plenty of people try both and keep the one that matches how they already like to work.

A third option: if you want the app, not the framework

Both GSD and Superpowers are things you install into the Claude Code terminal and drive yourself — prompts, skills, and commands you compose. If what you actually want is that same phased, review-gated discipline delivered as an integrated app instead of a framework you assemble, that's a different category — and it's what we build.

defract runs the whole lifecycle — story → design → architecture → review → ship — as a desktop app: the phases, fresh git-worktree parallelism, and agent-reviewed merges come built in, plus a visual design stage where agents render mockups you approve before any code is written. Same core insight as these frameworks (structure beats a single degrading session); different form factor (an app with a UI, not skills in a terminal). If you live in the CLI and love composing your own setup, GSD or Superpowers is your answer. If you'd rather it came out of the box, that's us.

Want the process built in, not assembled?

defract runs your whole build on parallel Claude Code agents — local-first, on your own Claude subscription.

Comparison of third-party tools (GSD, Superpowers) based on their public repos and docs as of July 2026; both are independent open-source projects, evolve quickly, and are not affiliated with defract. Corrections welcome.