Superpowers

An agentic skills framework and software-development methodology by Jesse Vincent (obra) that makes coding agents brainstorm a spec, plan with TDD, and run subagent-driven development autonomously — installed and enforced always-on in Kevin's agent config.

What it is

Superpowers (obra/superpowers, 246K+ GitHub stars as of 2026-07-04) is "a complete software development methodology for your coding agents, built on top of a set of composable skills and some initial instructions that make sure your agent uses them." The skills trigger automatically, so the agent gains the methodology without the user doing anything special. Source: obra/superpowers README, https://github.com/obra/superpowers, 2026-07-04 Source: X/@shannholmberg, 2026-03-14

Current source snapshot on 2026-07-04: main HEAD d884ae04edebef577e82ff7c4e143debd0bbec99, latest tag v6.1.1 (peeled commit d884ae04edebef577e82ff7c4e143debd0bbec99), MIT license, 246,143 stars, 21,824 forks. Source: GitHub API; git ls-remote, 2026-07-04

The core loop the framework imposes from the moment you start a coding agent: Source: obra/superpowers README, https://github.com/obra/superpowers, 2026-06-12

  1. Don't jump to code. Step back and ask what you're really trying to do.
  2. Tease out a spec, then show it back in chunks short enough to actually read.
  3. After sign-off, write an implementation plan clear enough for "an enthusiastic junior engineer with poor taste, no judgement, no project context, and an aversion to testing" — emphasizing red/green TDD, YAGNI, and DRY.
  4. On "go," run subagent-driven development — agents work each task, inspecting and reviewing as they go, often running autonomously for a couple of hours without deviating from the plan.

Cross-harness

Superpowers installs per harness (install it separately for each): Claude Code, Antigravity, Codex App, Codex CLI, Cursor, Factory Droid, Gemini CLI, GitHub Copilot CLI, Kimi Code, OpenCode, and Pi. On Claude Code it ships through Anthropic's official plugin marketplace; on Codex App/CLI it ships through the official Codex plugin marketplace. Source: obra/superpowers README, https://github.com/obra/superpowers, 2026-06-30

# Claude Code (official marketplace)
/plugin install superpowers@claude-plugins-official

Where it fits Kevin's stack

Superpowers is not just referenced in Kevin's AGENTS.md — it is enforced. Its component skills run always-on across every session: Source: AGENTS.md Superpowers Enforcement, kevin-wiki, 2026-06-12

Skill When it fires
using-superpowers Always — skills-first check on every message
brainstorming Before any creative or implementation work
writing-plans After brainstorming, before implementation
systematic-debugging Before proposing fixes for bugs
test-driven-development Before writing implementation code
verification-before-completion Before claiming work is done

This makes Superpowers the upstream of Kevin's whole "brainstorm before building / verify before completion" discipline — the same gate-shape as the The Eval Loop (Slop Is an Output Problem) and Spec-Driven Development, distributed via Skills.sh - Agent Skills Registry and composing with the while-loop-plus-invariants view in Harness Engineering. New skills are still authored through Browser Testing Skills; Superpowers supplies the methodology scaffolding around them.

The current upstream basic workflow adds a fuller branch lifecycle: brainstorming, using-git-worktrees, writing-plans, subagent-driven-development or executing-plans, test-driven-development, requesting-code-review, and finishing-a-development-branch. Kevin's repo does not need to mirror every upstream skill name; the local invariant is that each serious task has a design/spec, isolated work path when needed, executable plan, review loop, verification proof, and cleanup/merge decision. Source: obra/superpowers README, 2026-06-30

It is also, per the marketing-angle bookmarks, broadly applicable beyond engineering — the brainstorm -> spec -> tasks -> subagents -> review pipeline maps onto any structured knowledge work, not just code. The reviewed image makes this concrete for marketing: brainstorm campaign angles/questions/brief, plan bite-sized deliverables/file paths, execute with researcher/writer/editor/publisher agents, review against brand foundation/rubric, then verify final voice and publish to channels. Source: X/@shannholmberg thread, 2026-03-14; Source: local asset /wiki/assets/x-bookmarks/2032892199751528486/image-01.jpg, 2026-06-30


Timeline