Superpowers

Homonym resolved 2026-08-30. An unrelated project also called Superpowers — Sparklin Labs' open-source HTML5 game engine + IDE — lives in ../game-engines-wiki. It was slugged superpowers too until today; at the curator's decision it is now superpowers-game-engine, so a bare [[superpowers]] unambiguously means this page. Still two unrelated projects; still must not be merged. (Collision recorded in the 2026-07-28 quality cycle; resolved by rename 2026-08-30.)

"An agentic skills framework & software development methodology that works" — a library of composable agent skills plus the glue that makes a coding agent reach for them on its own, so it follows a structured process (brainstorm → plan → test → build → debug → review) instead of jumping from a prompt straight to a giant diff. By Jesse Vincent (obra) and the Prime Radiant team. MIT.

What it is

Superpowers packages a full development methodology as skills — small, composable instruction sets the agent invokes when a situation calls for one (the progressive-disclosure model of the open agent-skills standard). It is a concrete, opinionated instance of that standard: not one pasted rule, but a couple dozen skills covering the arc of building software.

Skills it ships (by phase):

Design / planning: brainstorming, writing-plans, executing-plans, spec work (cf. spec-driven-development).
Testing & verification: test-driven-development, verification-before-completion.
Debugging: systematic-debugging.
Review: requesting-code-review, receiving-code-review.
Parallelism / orchestration: dispatching-parallel-agents, subagent-driven-development, using-git-worktrees, finishing-a-development-branch (the multi-agent end).
Meta: writing-skills, using-superpowers (the bootstrap skill that teaches the agent to find and use the rest).

Distribution

Installed via official plugin marketplaces across harnesses — Claude Code (Anthropic's marketplace), Cursor, Codex, GitHub Copilot CLI, Kimi Code, Gemini CLI, and others — making it one of the more cross-harness skill packs rather than a single-tool plugin.

Adoption claim — flagged

GitHub's API puts this repo at 272,113 stars / 24,336 forks (checked live 2026-08-14; created 2025-10-09 — under a year old), up from the ≈242,000 recorded here at the 2026-06 snapshot. That growth rate is high enough to put it among the most-starred repositories anywhere for a ten-month-old project, so treat the figure as a dated, volatile number to re-verify each time, not a settled fact — the review that surfaced it (below) raised the same caution about quoting "200K+ stars" as gospel. See ecc for a same-corner repo whose count briefly looked close to this one on a stale comparison; live-checked side by side, the two are ~32k stars apart, not near-identical.

Source note (secondary review)

Surfaced via a member-only Level Up Coding / Medium review, "I Installed 'Superpowers' Into Claude Code. It Stopped Coding Like an Intern" by Abhishek Agarwal (https://levelup.gitconnected.com/superpowers-claude-code-plugin-536f8189922a) — a T4 practitioner write-up whose verdict is that Superpowers installs "a whole playbook" the agent follows, while cautioning it can be overkill for small tasks. The substantive claims here are grounded in the T1 repo above; the review is recorded for provenance and its overkill caveat.

Reflexive note

This is the skill framework this very wiki-hub session runs on — using-superpowers, brainstorming, test-driven-development, and the rest are live in the operating environment. Documenting it here is the spoke describing one of its own tools.

Nearest sibling

superclaude-framework packages a near-identical development arc for Claude Code only, as operator-typed slash commands rather than skills the agent self-invokes — the same methodology with the control point moved. Read the two together; neither has been evaluated.

superclaude-framework · agentskills-spec · anthropic-skills · agentic-coding-harness · spec-driven-development · agent-orchestration · claude-code-best-practices