obra/superpowers
A rulebook that forces AI coding agents to plan, test, and review before writing code
Superpowers is a set of reusable 'skills' plus starting instructions that make coding agents like Claude Code or Codex follow a disciplined workflow instead of jumping straight into code. It walks the agent through brainstorming a spec, writing a step-by-step plan, then having subagents implement each task with test-first coding and two-stage code review. The project works as a plugin across many different coding-agent tools, not just one.
What it does
- Problem: coding agents often start writing code immediately without a clear spec, plan, or tests, producing messy or wrong results.
- Approach: the framework chains together skills — brainstorming a design, breaking it into 2-5 minute tasks, dispatching subagents per task, enforcing RED-GREEN-REFACTOR test-driven development, and running spec/quality code reviews — that trigger automatically without the user invoking them manually.
- It installs as a plugin/extension across many agent harnesses: Claude Code, Antigravity, Codex App/CLI, Cursor, Devin CLI, Factory Droid, Gemini CLI, GitHub Copilot CLI, Grok Build CLI, Kimi Code, OpenCode, Pi, and Hermes Agent.
- Result: agents can reportedly work autonomously for a couple of hours following the agreed plan without deviating, since the plan and reviews keep them on track.
- The repo has 274,467 stars, is written mostly in Shell, and is maintained by Jesse Vincent and Prime Radiant, with optional anonymous telemetry (version number only) that can be disabled.
Why it matters
For anyone relying on AI coding agents, this offers a concrete process to reduce the risk of the agent generating unreviewed, untested, or off-spec code. It's also useful as a template for how to structure agent workflows across many different tools rather than being locked into one platform.
Terms in this repo
- skill · A reusable, self-contained instruction module the agent triggers automatically for a specific task
- RED-GREEN-REFACTOR · A test-driven development cycle: write a failing test (red), make it pass minimally (green), then clean up the code (refactor)
- subagent-driven development · Having a fresh AI subagent handle each individual task, with its work then reviewed before moving on
- YAGNI · 'You Aren't Gonna Need It' — a principle against building features not currently needed
- worktree · An isolated copy of a git repository's working directory, used here to keep development branches separate
Repository description (English)
An agentic skills framework & software development methodology that works.
Open on GitHubTrending repos
- cathrynlavery/diagram-designA skill that makes AI coding tools draw magazine-quality diagrams instead of generic rounded boxes
- public-apis/public-apisA giant crowd-curated directory of free APIs for developers
- semantica-agi/semanticaAn open-source graph infrastructure that lets AI agents show their work, not just their answers
- cactus-compute/needleA 14MB AI model small enough to run tool-calling on a phone or watch, without internet
- unslothai/unslothA desktop app that lets you run and train AI models on your own computer, no coding required
- macro-inc/macroAn all-in-one workspace where email, chat, docs, tasks, and CRM are cross-linked and share one AI memory
- harry0703/MoneyPrinterTurboAn open-source tool that turns a single topic or keyword into a finished short video, complete with script, footage, subtitles, and music
- basecamp/omarchyA ready-made, opinionated Linux setup built by DHH
Latest from METAL LAB
- Grok leaks conversations, location data via commands hidden in ciphertext
- Cursor lets agents handle long-running tasks with new "/goal" command
- NVIDIA releases Switchyard, an LLM routing proxy for coding agents
- Caveman Cuts Claude Code Token Usage by 33% Using Caveman-Speak
- Apple Keeps 2027 Target for Camera AirPods, Qwen3.8-27B Makes 80 Tool Calls Locally