Built for the Enthusiastic Junior Engineer With Poor Taste
It forces your AI coding agent to ask what you're building, write a plan, and get sign-off before it writes a single line of code.
Superpowers imposes software engineering discipline on autonomous coding agents by treating them like eager junior developers who need strict process, not just bigger context windows.

What it does
Superpowers wraps your existing coding agent—Claude, Codex, Cursor, Gemini, or others—in a full software development methodology enforced through composable skills. Instead of generating code immediately, the agent enters a structured lifecycle: Socratic brainstorming to refine the spec, human sign-off on bite-sized design chunks, granular implementation plans, and only then autonomous execution. The skills trigger automatically, so the agent cannot skip stages like red-green TDD, systematic debugging, or two-stage code review even if it wanted to.
The interesting bit
The framework explicitly treats the agent as an “enthusiastic junior engineer with poor taste, no judgement, no project context, and an aversion to testing,” then builds guardrails to compensate. By dispatching fresh subagents for individual tasks and forcing a spec-compliance review before a code-quality review, it creates a self-managing assembly line that can run for hours without human intervention.
Key highlights
- Supports eight coding harnesses including Claude Code, Codex CLI, Cursor, and OpenCode
- Mandates RED-GREEN-REFACTOR test-driven development and deletes any code written before tests
- Breaks work into 2–5 minute tasks with exact file paths, complete code, and verification steps
- Uses git worktrees to isolate branches and enforces a merge/PR/keep/discard decision workflow
- Includes meta-skills for writing new skills, though the project rarely accepts external contributions
Caveats
- The maintainer states they “don’t generally accept contributions of new skills,” so don’t expect a community-driven skill marketplace
- Every skill must remain portable across all supported agents, which limits harness-specific optimizations
- Update behavior varies by coding harness, so consistency depends on the tool you’re using
Verdict
Use this if you want your agent to work autonomously for hours without deviating from the plan. Skip it if you prefer tight, iterative pair-programming where the agent reacts to every keystroke.
Frequently asked
- What is obra/superpowers?
- It forces your AI coding agent to ask what you're building, write a plan, and get sign-off before it writes a single line of code.
- Is superpowers open source?
- Yes — obra/superpowers is open source, released under the MIT license.
- What language is superpowers written in?
- obra/superpowers is primarily written in Shell.
- How popular is superpowers?
- obra/superpowers has 259.5k stars on GitHub and is currently cooling off.
- Where can I find superpowers?
- obra/superpowers is on GitHub at https://github.com/obra/superpowers.