Superpowers: The Agentic Dev Methodology Picking Up 1,800 Stars a Day

Why It Is Trending

obra/superpowers is picking up roughly 1,800 GitHub stars per day as of mid-May 2026, pushing past 190,000 total stars. That velocity points to a simple value proposition: most AI coding agents improvise their way through development. Superpowers refuses to let that happen. Built by Jesse Vincent and the Prime Radiant team, it installs as a plugin across eight major agent platforms and replaces ad-hoc prompting with a mandatory, repeatable seven-step workflow. The project reached v5.1.0 in early May 2026, and the sharp star acceleration suggests word of mouth from developers who adopted it and found it actually changed how their agents behave.

What It Does

Superpowers is a software development methodology packaged as composable skills for AI coding agents. It works by enforcing a structured loop that covers the full development lifecycle from intent clarification to branch cleanup - nothing is left to agent discretion.

The seven steps, drawn directly from the README:

1. Brainstorming - The agent asks clarifying questions, documents intent, and presents design in digestible sections before any code is written.

2. Git Worktrees - An isolated workspace is created, a branch is established, and a clean test baseline is confirmed.

3. Writing Plans - Work is decomposed into 2-5 minute tasks, each with exact file paths, complete code stubs, and verification steps.

4. Subagent-Driven Development - A fresh subagent handles each task, followed by a two-stage review: spec compliance first, then code quality.

5. Test-Driven Development - The RED-GREEN-REFACTOR cycle is enforced. Failing tests must be written before implementation code.

6. Code Review - Severity-based issue reporting gates progression. Critical issues block advancement to the next task.

7. Branch Completion - Tests are verified, then merge, PR, or worktree retention options are presented before cleanup.

The skills library covers testing, debugging, collaboration, and meta-skills including dispatching-parallel-agents, subagent-driven-development, using-git-worktrees, systematic-debugging, and requesting-code-review. Each skill is a plain text file that an agent reads and follows.

The repo is MIT licensed and primarily Shell (66.4%), which means the skill files are readable in any editor without tooling dependencies.

Install / Try It

Installation takes one command per platform.

Claude Code - official marketplace:

/plugin install superpowers@claude-plugins-official

Claude Code - Superpowers marketplace:

/plugin marketplace add obra/superpowers-marketplace
/plugin install superpowers@superpowers-marketplace

Cursor:

/add-plugin superpowers

Gemini CLI:

gemini extensions install https://github.com/obra/superpowers
gemini extensions update superpowers

Codex CLI: search for "superpowers" via the /plugins interface.

Factory Droid:

droid plugin marketplace add https://github.com/obra/superpowers
droid plugin install superpowers@superpowers

GitHub Copilot CLI:

copilot plugin marketplace add obra/superpowers-marketplace
copilot plugin install superpowers@superpowers-marketplace

OpenCode: tell the agent: "Fetch and follow instructions from https://raw.githubusercontent.com/obra/superpowers/refs/heads/main/.opencode/INSTALL.md"

After installation, run /using-superpowers to see the full command surface, then start any development task and the brainstorming skill activates automatically.

Who Should Use It

Superpowers targets developers who are already comfortable with AI coding agents but have hit the improvisation ceiling - the agent writes plausible-looking code, claims success, and then something breaks in a way that takes longer to debug than it would have taken to write the code manually.

It is a good fit for:

Teams shipping production software with agents. The mandatory code review step and severity gating mean the agent cannot declare victory without evidence. Critical issues block progression rather than get buried in a summary.

Developers committed to test-driven development. The RED-GREEN-REFACTOR enforcement is not optional. If you already work this way, Superpowers makes the agent follow the same discipline instead of skipping to green.

Anyone running parallel agent tasks. The dispatching-parallel-agents skill and Git worktree integration give you a structured way to coordinate multiple subagents without branch collisions or lost context.

Developers who work across platforms. Because Superpowers installs across Claude Code, Cursor, Gemini CLI, Codex CLI, and more, the same methodology follows you regardless of which tool you use that day.

It is a heavier investment than pasting a single prompt. The methodology expects engagement with all seven steps. If you want a lightweight wrapper, this is not it.

Relation to the DevDigest Ecosystem

The DevDigest site covers Claude Code skills and agent workflows closely. The skills directory at skills.developersdigest.tech catalogs reusable Claude Code skills - a direct parallel to what Superpowers builds: the idea that agent behavior should be defined by composable, version-controlled text files rather than improvised prompt sessions.

Superpowers takes the skills concept further by treating the entire development methodology as a skill set. The subagent-driven development pattern it enforces - fresh subagent per 2-5 minute task, two-stage review, spec compliance then code quality - is the kind of agent orchestration covered at subagent.developersdigest.tech.

The dispatching-parallel-agents skill in the Superpowers library maps to the parallel agent patterns that Claude Code hooks enable. If you already use hooks to automate pre and post-tool actions in Claude Code (see hooks.developersdigest.tech), combining them with Superpowers' planning step gives you a setup where agent scope is defined before execution and side effects are managed automatically.

The methodology gap Superpowers addresses - agents that improvise instead of follow process - is one the DevDigest community knows well. Watch this project as it matures past v5.

Honest Assessment

Superpowers is solving a real problem. AI agents are undisciplined by default, and the seven-step loop, mandatory TDD, and subagent-per-task design are all sound structural choices. The multi-platform support is genuinely useful, and plain-text skill files are easy to read, fork, and adapt.

The limitations are also real. The README provides no benchmarks - no numbers on how much the methodology reduces bug rates, how much overhead each step adds, or how two-stage code review compares to none. You are taking the methodology on faith until you run it on your own codebase.

The mandatory enforcement model is also a cultural fit question. If your team prefers lighter agent interactions, the rigid seven-step structure may feel like overhead rather than guardrail. And because each task runs in a fresh subagent, projects with deep shared context require careful planning to avoid losing state between tasks.

Worth installing and testing on a real feature branch before committing to it across a team.

References

• obra/superpowers repository: https://github.com/obra/superpowers
• Superpowers v5.1.0 release (May 4, 2026): https://github.com/obra/superpowers/releases
• DevDigest skills directory: https://skills.developersdigest.tech
• DevDigest subagent guide: https://subagent.developersdigest.tech
• DevDigest hooks reference: https://hooks.developersdigest.tech