AI DevKit vs Superpowers

Both AI DevKit and Superpowers solve the same core problem: AI coding agents are powerful, but they need structure. Without a clear process, results become inconsistent, context is lost between sessions, and important engineering steps get skipped.

These tools take different paths to fix that.

	AI DevKit	Superpowers
Type	Workflow layer for AI coding agents with memory and skills	Opinionated skills framework
Install	npm install -g ai-devkit	Agent-specific setup guide (see official repository)
Approach	One config across agents: workflow skills, memory, and verification	Behavior focused: process instructions injected into agent prompts
Memory	Built-in local SQLite memory system	Relies on plan documents and git history
Skills	Built-in core skills (dev-lifecycle, debug, simplify-implementation, document-code) plus community registry via skill add/remove/find	14 built-in composable skills
Agents supported	11+ (Cursor, Claude Code, Codex, Copilot, Gemini CLI, OpenCode, Antigravity, and more)	4 (Claude Code, Cursor, Codex, OpenCode)
Documentation	Phase-based directory structure (docs/ai/)	Design docs saved to docs/plans/
Execution model	Single agent per feature with persistent memory	Sub-agent dispatching with two-stage review
License	MIT	MIT
Best for	Teams that want one repeatable workflow across coding agents, plus memory and reusable skills	Teams that want strict, opinionated, multi-agent workflow enforcement

Note: Agent and skill counts can change over time. Check each project's repository for the latest numbers.

• Choose AI DevKit if you want one config across agents, persistent local memory, verification, review, and flexible workflows.
• Choose Superpowers if you want strict process gates and sub-agent orchestration.
• Use both if you want AI DevKit as the infrastructure layer and Superpowers as the behavioral layer.

npm install -g ai-devkit
ai-devkit init
# then in your AI editor
Use the dev-lifecycle skill to start requirements for <feature>

Follow the official setup guide for your agent at https://github.com/obra/superpowers, then run one planning skill and one implementation skill on a small test feature to evaluate fit.

AI DevKit is a workflow layer for AI coding agents that you install globally and initialize per project. It creates workflow docs, installs built-in skills like dev-lifecycle, structured-debug, simplify-implementation, and document-code, adds persistent local memory, and makes agents verify work before claiming completion.

npm install -g ai-devkit
ai-devkit init

After initialization, your project gets a docs/ai/ directory with subdirectories for requirements, design, planning, implementation, and testing. Your AI agent uses workflow skills such as dev-lifecycle, tdd, and verify to plan before code and review before push.

AI DevKit also provides a local memory system backed by SQLite. It stores coding standards, patterns, and decisions. This memory persists across sessions, so your AI agent can recall project-specific knowledge without you repeating it.

Superpowers is an agentic skills framework (skills strongly drive how the AI plans and executes work). It provides 14 composable skills across the development lifecycle.

When installed, Superpowers injects skill instructions into your AI agent's context. Skills like brainstorming, writing-plans, and test-driven-development enforce strict process gates. The agent must follow each phase before proceeding to the next.

A distinctive feature is sub-agent dispatching. Superpowers can spin up fresh, isolated AI agent instances per task. Each sub-agent's work goes through a two-stage review: first for specification compliance, then for code quality.

AI DevKit includes a dedicated memory system. You can store and search knowledge using MCP tools or CLI commands (ai-devkit memory store, ai-devkit memory search). Memory is scoped by global, project, or repository level and tagged for easy discovery. This means your AI agent can recall decisions, patterns, and conventions from previous sessions.

Superpowers does not have a built-in memory system. Context persists through plan documents saved to docs/plans/ and git commit history. Each sub-agent starts fresh, which prevents context pollution but also means prior knowledge must be explicitly included in task descriptions.

AI DevKit provides a repeatable workflow but remains flexible. Built-in skills like dev-lifecycle and structured-debug encourage consistency, while simplify-implementation and document-code keep solutions practical and reusable. You can still adapt the workflow to your project. The docs structure is guidance, not a hard gate.

Superpowers enforces strict process discipline. In its documented methodology, test-first development is treated as mandatory.

AI DevKit combines built-in workflow skills (dev-lifecycle, debug, simplify-implementation, document-code) with a community-driven skill registry. You can browse, install, and manage additional skills with CLI commands (ai-devkit skill add, ai-devkit skill find). This gives teams a solid default methodology plus extensibility for domain-specific capabilities.

Superpowers comes with 14 built-in skills that form one cohesive methodology. There is a writing-skills skill for contributing new skills, but the system is more monolithic, with skills designed to work together in one workflow.

AI DevKit supports 11+ AI coding environments, including Cursor, Claude Code, GitHub Copilot, Gemini CLI, OpenAI Codex, OpenCode, Antigravity, KiloCode, AMP, and Roo Code. Each environment gets tailored configuration.

Superpowers focuses on 4 platforms: Claude Code, Cursor, Codex, and OpenCode. Each has a dedicated installation method and configuration directory.

AI DevKit usually runs one agent per feature, enhanced by persistent memory and workflow docs. In practice, teams often use the built-in dev-lifecycle skill, which creates a new git worktree per feature by default. This lets multiple features run in parallel while each feature stays isolated at the branch/worktree level.

AI DevKit's preferred scaling pattern is to maximize context quality with one agent per feature, then scale out by feature (more worktrees), instead of splitting one feature across many sub-agents. This reduces context fragmentation and avoids extra context pressure from coordinating many agents on the same feature.

Superpowers uses a multi-agent architecture. The subagent-driven-development skill dispatches fresh sub-agents per task, each with two-stage review. This helps prevent context contamination, but it usually costs more coordination and compute.

• A workflow layer that works with the AI coding tools you already use
• Persistent memory across sessions without repeating yourself
• Support for a wide range of AI coding environments
• Flexibility to adopt structured practices gradually
• Built-in skills for lifecycle, debugging, simplification, and code documentation
• A community skill ecosystem to extend capabilities
• Simple installation via npm

• Strict process enforcement with hard gates between phases
• Multi-agent task execution with built-in code review
• An opinionated methodology where TDD is mandatory
• Sub-agent isolation to prevent context pollution
• A self-contained framework with all skills included

AI DevKit and Superpowers are not mutually exclusive. AI DevKit can provide the infrastructure layer (memory, skill management, scaffolding), while Superpowers can provide the behavioral layer (strict process rules and multi-agent orchestration). Teams that want both persistent memory and strict workflow enforcement can combine them.

Ready to make your AI coding agent follow a repeatable engineering workflow? Install AI DevKit and initialize your project:

npm install -g ai-devkit
ai-devkit init

Then ask the agent to use dev-lifecycle so it clarifies the feature before editing code, or explore the documentation to learn more.

• AI DevKit repository: https://github.com/codeaholicguy/ai-devkit
• Superpowers repository: https://github.com/obra/superpowers