Jesse Vincent’s Superpowers framework has crossed 174,000 GitHub stars seven months after its October 2025 launch, establishing itself as the most widely adopted agent skills framework for coding agents. The project ships fourteen SKILL.md files that enforce structured engineering practices, from TDD to code review to systematic debugging, across six different agent platforms.

What Superpowers Does

The framework’s core thesis, as analyzed by Marc Nuri, is that coding agents don’t lack capability. They lack discipline. An agent can lecture on test-driven development and then immediately write production code without a single test. Superpowers addresses this by injecting a session-start hook, reportedly under 2,000 tokens, that forces the agent to read and follow structured skill files before executing any task.

Each of the fourteen skills enforces a specific engineering practice. The TDD skill opens with “NO PRODUCTION CODE WITHOUT A FAILING TEST FIRST.” The verification skill requires “NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE.” The debugging skill demands “NO FIXES WITHOUT ROOT CAUSE INVESTIGATION FIRST.” Each rule is paired with a table of red flags listing the rationalizations agents most commonly use to skip the discipline, according to Nuri’s analysis.

A typical task moves through brainstorming, git worktree creation, a written plan broken into two-to-five-minute subtasks, subagent-driven implementation with TDD enforcement, a code review pass from a fresh agent, and a finishing step that closes the branch. Each phase has its own skill file. The next phase does not begin until the previous one completes.

Six Platforms, One Skills Directory

Superpowers is not locked to a single vendor. The same skills/ directory powers Claude Code, OpenAI Codex (CLI and app), Cursor, GitHub Copilot CLI, Gemini CLI, and OpenCode. Each host gets a thin manifest file (.claude-plugin/, .codex-plugin/, .cursor-plugin/, gemini-extension.json) that points its own discovery mechanism at the same markdown. The skills themselves are host-agnostic.

The Anthropic marketplace plugin shipped in early 2026. OpenAI Codex added native plugin support in its app sidebar. Cursor and Gemini CLI each added their own installation paths. The result: one repository governs agent behavior regardless of which coding tool a team standardizes on.

The Adoption Signal

The VoltAgent/awesome-agent-skills repository, a curated index of over 1,000 agent skills across platforms, lists Superpowers alongside its lab environment. Vincent published a fifth major release in March 2026 and a candid post on the “agentic slop PR” problem the project exposed at scale: agents generating plausible but sloppy pull requests that pass surface-level review.

174,000 stars in seven months places Superpowers among the fastest-growing developer tools on GitHub. The framework’s entire payload fits in a directory you can read in an hour. No fine-tuned model, no proprietary SDK, no agent platform subscription. Plain text discipline, distributed as git.

The Portability Question

The broader signal is that agent behavior standardization may not require new protocols or vendor lock-in. If fourteen markdown files can make coding agents consistently follow TDD, run verification before claiming completion, and refuse to fix bugs they haven’t diagnosed, the same pattern could extend to security runbooks, release management, and on-call procedures. The constraint is not whether agents can follow structured instructions. It is whether teams will invest in writing them.