Who superpowers is for#
Developer using Claude Code
Installs Superpowers into Claude Code so all AI-assisted features start with a proper spec review instead of jumping directly to implementation.
Team enforcing consistent agent methodology
Standardizes on Superpowers to ensure every agent run from any developer's machine follows the same TDD methodology without manual CLAUDE.md maintenance.
Solo developer on a long project
Uses Superpowers to maintain discipline on a long solo project, letting the agent run autonomously for multi-hour sessions while trusting the spec-first guardrails.
Agency with client sign-off requirements
Installs Superpowers for client projects where the client must review and sign off on the implementation plan before any development begins.
Developer evaluating a new agent harness
Installs Superpowers first to validate whether the harness actually follows structured prompts before committing to a new coding agent.
The problem it solves#
Coding agents left to their own defaults write code before understanding the problem, skip tests, violate YAGNI, and drift from the original spec after a few turns. Custom CLAUDE.md files and system prompt patches help but are fragile, hard to maintain across projects, and do not enforce a consistent workflow from requirement gathering through implementation. There is no standard way to install a shared set of development standards that works across multiple agent harnesses.
How it solves it#
Spec-first workflow with review gate
Spec-first workflow: the agent asks clarifying questions before writing any code and shows the spec in digestible chunks for review before implementation starts
Subagent-driven autonomous implementation
Subagent-driven implementation: after spec sign-off, the agent uses subagents to work through each engineering task, inspect results, and continue autonomously for hours without derailing
Enforces TDD, YAGNI, and DRY automatically
Enforces true red/green TDD, YAGNI, and DRY via skills that trigger automatically on recognized task patterns
Available on the Claude plugin marketplace
Available on the official Claude plugin marketplace: `/plugin install superpowers@claude-plugins-official`
Supports 8+ agent harnesses
Supports Claude Code, Codex CLI, Codex App, Cursor, Factory Droid, Gemini CLI, OpenCode, and GitHub Copilot CLI
MIT licensed, composable and forkable
MIT licensed; skills are composable markdown files you can fork and extend
Strengths and trade-offs#
Strengths
- Automatic triggering, no per-project configSkills trigger automatically without any per-project configuration once installed at the harness level
- One methodology across all your harnessesSupports multiple agent harnesses, so one methodology applies across your entire toolchain regardless of which agent you use on a given day
- Spec review catches misunderstandings earlyThe spec-review loop catches misunderstandings before long implementation runs, not after hours of agent work that misses the actual requirement
- MIT licensed with plugin marketplaceMIT licensed with active development and a plugin marketplace for extending with additional skill packs
Trade-offs
- -Spec gate is intentionally constrainingImposes a structured spec-first workflow: teams that want agents to start coding immediately without a requirements conversation will find it constraining by design
- -Effectiveness varies with model capabilityEffectiveness depends on the underlying agent following skill prompts accurately; lower-capability models may not adhere consistently to the TDD and YAGNI enforcement
- -Separate install step per harnessInstalling across multiple harnesses requires a separate install step for each; there is no single cross-harness install command
superpowers vs alternatives#
superpowers is an MIT-licensed open-source workflow layer for coding agents such as Claude Code, Codex, Cursor, Gemini CLI, OpenCode, and GitHub Copilot CLI.
vs Cursor: Cursor gives developers a proprietary AI editor with integrated chat, autocomplete, and agent workflows. superpowers does not replace that editor; it adds portable skills for brainstorming, planning, worktrees, test-driven development, subagent execution, and review, so teams can carry the same engineering process across Cursor and other agent harnesses.
vs Claude Code: Claude Code provides the proprietary agent runtime and model access. superpowers sits on top as a methodology plugin, with skills that force design review, implementation planning, red-green-refactor discipline, and code review checkpoints before an agent declares work complete.
vs managed coding agents: Managed agents such as Devin-style services aim to own more of the implementation loop for you. superpowers is better when you want your existing agent tools to follow a visible, repo-local process; managed agents are still better when you want a hosted service to handle orchestration, credentials, and long-running execution infrastructure.
What it's built on#
- Languages
- JavaScriptPythonTypeScript
FAQ#
Will Superpowers slow down my coding sessions by adding approval steps?
Yes, intentionally. The spec step adds an upfront conversation before implementation begins. In practice, this typically saves more time than it costs by preventing long agent runs that miss the actual requirement and need to be restarted.
Does it work with harnesses other than Claude Code?
Yes. Superpowers supports Claude Code, Codex CLI, Codex App, Cursor, Factory Droid, Gemini CLI, OpenCode, and GitHub Copilot CLI. Each harness requires a separate install step.
Are the skills visible and editable?
Yes. Skills are composable markdown files. Fork the repo, modify the skills, and install your customized version by registering your fork as a marketplace with your harness.
Similar open-source tools#
agentmemory
Persistent memory for AI coding agents
DeepSeek TUI
A coding agent that lives in your terminal.
Parallel Code
Run AI coding agents in parallel, each in its own worktree
orca
The ultimate IDE for coding agents
oh-my-pi
A coding agent with the IDE wired in
Agent Skills
Structured workflows for AI coding agents.