🪴 Brain

Superpowers

3 min read

Superpowers

obra/superpowers — a complete software development workflow for coding agents, built on a set of composable “skills” + initial instructions that force the agent to use them. Author: Jesse Vincent (obra). Philosophy: the agent does not jump straight to code — first it teases out a spec from the conversation, shows it in small chunks, and only after sign-off writes a plan for an “enthusiastic junior with poor taste, no judgement, no project context, and aversion to testing”, then subagent-driven-development with true red/green TDD, YAGNI, and DRY.

đź”— Links

Description

Download or use

# Claude Code (official marketplace)
/plugin install superpowers@claude-plugins-official
 
# Cursor
/add-plugin superpowers
 
# Codex / OpenCode — fetch instructions
# from raw.githubusercontent.com/obra/superpowers/.../INSTALL.md
 
# GitHub Copilot CLI
copilot plugin marketplace add obra/superpowers-marketplace
copilot plugin install superpowers@superpowers-marketplace
 
# Gemini CLI
gemini extensions install https://github.com/obra/superpowers

🗒️ Description

🧩 The Basic Workflow (7 mandatory skills)

  1. brainstorming — fires before code is written; refines rough ideas through questions, alternatives, design in sections to validate, saves a design doc.
  2. using-git-worktrees — after the design is approved; isolated workspace on a new branch, project setup, clean test baseline.
  3. writing-plans — bite-sized tasks (2-5 min each), exact paths, complete code, verification steps.
  4. subagent-driven-development or executing-plans — fresh subagent per task; two-stage review (spec compliance, then code quality), or a batch with human checkpoints.
  5. test-driven-development — RED-GREEN-REFACTOR; failing test → watch it fail → minimal code → watch it pass → commit. Discards code written before the tests.
  6. requesting-code-review — between tasks, review against the plan, severity-graded, critical = block.
  7. finishing-a-development-branch — verify tests, options (merge/PR/keep/discard), clean up the worktree.

Mandatory workflows, not suggestions — the agent automatically checks the skills before every task.

🧩 Skills Library (beyond the Basic Workflow)

  • Testing: testing anti-patterns reference (within TDD).
  • Debugging: systematic-debugging (4-phase root cause: root-cause-tracing, defense-in-depth, condition-based-waiting), verification-before-completion.
  • Collaboration: dispatching-parallel-agents, receiving-code-review.

🧩 Position in the ecosystem

This is a methodology framework — not a set of roles like gstack, not a YAML engine like Archon, not a minimal 1-page CLAUDE.md like Karpathy Skills. Philosophy: process rigor > skill count. Strongest combined with TDD and worktree isolation.

✍️ Reasoning for

For me the value clusters around three things:

  1. Forced TDD — in my Claude Code setups on Qamera AI I often punt tests “for later” and pay for it; Superpowers does not let me.
  2. Subagent-driven-development with fresh context — eliminates “context rot” after 30+ tool calls (same pattern as loop nodes in Archon).
  3. Multi-host portability — CC + Cursor + Codex + OpenCode + Copilot + Gemini from a single source; fits my workflow of jumping between Claude Code and Cursor.

Conflict with gstack: both try to structure the same thing (think→plan→build→review→test→ship). I’ll probably pick one per repo rather than layering them.

Warning: the rigor is literal — “discards code written before the tests”. You have to accept that up front, otherwise frustration is guaranteed.

Alternatives considered

  • gstack — role-based virtual team (CEO/Designer/QA), more startup-shaped; Superpowers is engineering-rigor-shaped
  • Karpathy Skills — minimal methodology in 1 CLAUDE.md; Superpowers is a complete stack
  • Archon — YAML workflow engine + worktrees; Superpowers stays in native skills + standard git
  • Agent Skills standalone — just the skills, no enforced methodology

đź”— Resources

Template: tool

Graph View

Backlinks