Claude Code has rapidly emerged as a widely adopted AI coding tool, empowering individual developers, startups, and large engineering teams to significantly accelerate code writing, pull request reviews, debugging, and feature shipping. Central to its extensibility are Claude Skills—composable, auto-invoked instruction sets that shape how the AI agent plans, builds, and verifies. Anthropic's strategic consolidation of previously separate "slash commands" and "skills" into a unified skills format underscores their importance as the canonical method for extending the agent's behavior.
With Skills now foundational to the experience, the community has rallied around open-source frameworks like Superpowers and gstack, which package best practices into ready-made skill sets. However, attempting to install both can lead to conflicts, and haphazardly stacking frameworks often degrades system stability rather than improving it. This raises the critical question of how to choose and integrate them effectively.
This article proposes a stable three-layer practice to address not just the "Superpowers or gstack?" dilemma, but also the broader gaps in decision-making, durable context, and execution. The core insight is: gstack handles strategic thinking, GSD ensures context stability, and Superpowers drives execution.
In practice, a stable AI-assisted development workflow is best achieved through a clear division of labor across these three layers:
- Decision / Roles (gstack): Provides judgment from CEO, design, architecture, and quality assurance (QA) perspectives, extending beyond mere "how to code." It focuses on "what to do" and "whether it should be done."
- Context / Spec (GSD): Critical for preventing specification drift, status degradation, boundary decay, and context rot over long-running tasks. It ensures the work "does not drift."
- Execution (Superpowers): Manages the "how work gets done" loop, encompassing requirement clarification, planning, test-driven development (TDD), and acceptance. It provides a "smooth execution loop."
Specifically, each framework's distinct strengths are:
- Superpowers: Focuses on how work gets done; offers a smooth execution loop.
- gstack: Determines what to do and if it should be done; provides richer, role-based judgment.
- GSD: Prevents drift; maintains steadier specifications and context across extended task chains.