The archive · AI & Models · Product decision · 2025-2026
Prime Radiant's Superpowers: veteran dev's agent-skills framework hits 147k GitHub stars
Jesse Vincent's Superpowers encodes senior-dev discipline into coding agents; launched Oct 2025, it hit ~147k stars and GitHub Trending #2 by April 2026.
Prime Radiant
What the business is
Superpowers is an MIT-licensed agentic-skills framework and software-development methodology: a set of composable SKILL.md files that install into Claude Code, Codex, Cursor, Gemini CLI and other coding agents and gate progress behind structured brainstorming, written plans, red-green TDD, systematic debugging and subagent code review.
How it started
Jesse Vincent, the veteran open-source developer behind Request Tracker, Perl 6 leadership (2005-2008), Keyboardio and K-9 Mail, started using Claude Code on day one in February 2025 and spent a month watching it overbuild: it skipped tests, implemented features before understanding requirements and patched symptoms instead of root causes. He concluded the agent was not dumb, just undisciplined, and began months of prompt and workflow engineering, documenting every Claude.md change. On the morning of October 9, 2025, Anthropic shipped Claude Code's plugin system; Vincent shipped Superpowers the same day as a plugin marketplace install.
What happened
The framework grew from a personal toolkit into the most-used Claude Code plugin: 20+ composable skills (brainstorming, writing-plans, test-driven-development with an 'Iron Law' that code written before a failing test must be deleted, systematic-debugging, requesting-code-review, verification-before-completion) plus subagent-driven development that dispatches fresh agents per task and reviews each result twice. It was accepted into Anthropic's official plugin marketplace on 2026-01-15 and expanded beyond Claude Code to Codex CLI, Cursor, OpenCode, Gemini CLI, GitHub Copilot CLI and others as Anthropic, OpenAI and Google raced to ship native agent skills. Traction accelerated from roughly 27,000 stars in January 2026 to 147,000 by mid-April, with the repo ranking #2 on GitHub Trending on 2026-04-11; Vincent described the project as 'my hobby project from last fall went wildly out of control' on Heavybit's Open Source Ready podcast.
How it ended up
Still live and scaling as of September 2026: the repo passed roughly 265,000 stars by early September per Aliyun Developer's 2026-09-02 write-up, installation is documented across eight agent harnesses, and a community debate is underway about whether the full 'harness' pays for itself as models add native planning and verification features.
Background
Superpowers bets AI coding agents fail for lack of discipline, not intelligence. Jesse Vincent, creator of Request Tracker and veteran of Perl, Keyboardio and K-9 Mail, started using Claude Code in February 2025 and watched it overbuild: code before requirements, features before tests, fixes before root-cause diagnosis. He encoded the engineering process itself as skills - brainstorming before code, written plans, an 'Iron Law' of red-green TDD, systematic debugging and two-stage subagent review - and shipped on October 9, 2025, the morning Anthropic launched Claude Code's plugin system.
The format made the bet portable: skills are markdown files any agent that reads SKILL.md can follow, so the methodology installs into Claude Code, Codex CLI, Cursor, OpenCode, Gemini CLI and GitHub Copilot CLI. Anthropic accepted the plugin into its official marketplace on 2026-01-15, and the project rode the 2026 agent-skills wave. Termdock documented over 89,000 GitHub stars by 2026-03-16, five months after launch; by 2026-04-11 AgentConn reported 147,000+ stars, ranked #2 on GitHub Trending after adding 1,589 stars in a day.
The counter-position is equally visible: as models gain native planning, verification and subagent features, critics call the full harness a 'token tax,' and the most detailed 2026 reviews conclude the workflow survives while the all-in-one monolith is no longer the obvious default. Even skeptics concede the forced pause between intent and implementation is durable value; users who disabled Superpowers frequently rebuilt its core loop in their own configuration.
For Prime Radiant, Vincent's company, the project flipped a hobby into the most-used Claude Code plugin in the world within months, and turned a solo developer's process notes into the reference point for the argument that AI coding's battleground moved from models to skills.
What has to be true
- The pain was universal and visible: every early Claude Code user watched agents skip tests, guess architecture and patch symptoms, so the frustration Superpowers fixed was not niche.
- The insight was contrarian in 2025: the bottleneck was not model intelligence but absent process, which meant a markdown skills framework could compete with tools built by Anthropic and OpenAI.
- Portability multiplied distribution: because skills are plain files, one framework worked across Claude Code, Codex, Cursor and Gemini CLI instead of betting on a single vendor's IDE.
- The founder brought an existing audience: decades of open-source infrastructure made Vincent's launch post spread faster than an unknown developer's could.
- The risk was structural: native skills from platform owners kept absorbing the bottom of the stack, and the token cost of enforced ceremony could make the harness uneconomic as models improved.
What can be applied
When model capability is commoditized, the process layer becomes the product: encode senior-engineer discipline into skills an agent cannot skip, and make them portable so no vendor can lock users in.
Aftermath
As of 2026-09-05, obra/superpowers is live and still growing: Aliyun Developer's 2026-09-02 write-up put the repo at roughly 265,000 stars, with installs documented across eight coding-agent harnesses and Superpowers 5 (March 2026) adding visual brainstorming and adversarial spec review. No longer only a Claude Code plugin, it anchors the argument that agent skills, not models, are the next competitive layer, amid community debate over whether the full harness pays for itself as native model features improve. No funding or revenue figures have been disclosed for Prime Radiant.
Sources
- Superpowers: Skills Framework Reshaping AI Dev
- obra/superpowers: Claude Code Skills Framework Guide
- Open Source Ready | Ep. #36, Managing AI Coding Agents with Jesse Vincent
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card