Superpowers

1What it does

A skills package by Jesse Vincent (obra) — one of the better-known agent-skill systems in the ecosystem, built around the idea that an agent should have procedural memory: structured playbooks it activates for recurring jobs (writing, debugging, releasing, planning) rather than re-deriving the procedure each session. Superpowers ships as a set of SKILL.md packages plus the conventions for how the agent selects and chains them — brainstorming, systematic debugging, test-driven development, verification loops — the disciplines that separate a competent agent session from a flailing one. It was authored for the multi-agent world: the skills install across hosts (Claude Code, Codex, and Gemini CLI among them) through the standard skills directory, which is why it lands in this directory under the CLI's skills hub.

2When to use it

Reach for it when you want the agent to follow process, not just instructions — Superpowers' skills are about how to work (plan first, verify, debug systematically), which is the layer most skill libraries skip in favor of domain trivia. It fits developers who run long agent sessions and notice the agent improvising procedure badly — the skills give it a known-good playbook to activate instead. Skip it if your work is single-shot commands — the value compounds over multi-step tasks; a one-line ask gains nothing from a debugging methodology skill.

3How to get started

Install per the README — the package drops SKILL.md files into the skills directory your host reads (for Gemini CLI, the standard skills location the CLI discovers). After install, the activation test is a task that matches a skill's description — ask for help debugging a failing test and watch whether the agent invokes the systematic-debugging skill rather than free-styling. Read the top-level skill index first: knowing what disciplines exist is how you learn to phrase requests that trigger them.

4Configuration

Configuration is the skills you keep enabled — Superpowers arrives as a suite, and enabling all of them on a small project adds activation surface without payoff. The skills interoperate with your own SKILL.md files — they coexist in the same directory, and a project-specific skill can reference a superpowers discipline as its procedure. If your agent supports skill pinning or priority, the meta-skills (the process disciplines) deserve precedence over domain skills — they shape how the domain skill gets executed.

5Things to know

• Process skills are the differentiator — most libraries teach the agent about things; this one teaches it how to do things, which is rarer and harder to write.
• Multi-host by design — the SKILL.md standard is the shared format; skills installed for Gemini CLI travel to Claude Code or Codex unchanged.
• Authored and actively maintained — obra is a known agent-tooling author; the package reflects real usage, not a catalog scrape.
• The disciplines assume a capable model — skills that say "verify before declaring done" only help a model that can actually verify; the skill is a harness, not a fix.
• The skill index itself is worth reading as documentation — the disciplines it encodes (plan-first, systematic debug, verification) are a good working definition of agent competence.
• Because skills are host-portable, teams mixing CLI agents can standardize on one skill set — the same files serve Gemini CLI, Claude Code and Codex.
• The suite's real product is shared vocabulary — once a team adopts it, "run the debugging discipline" means something specific across every agent session, which is worth more than any single playbook.
• License and contribution model are standard open-source — forking a discipline and reshaping it for your stack is legitimate use, and upstream accepts refinements.
• Skill descriptions are the activation contract — Superpowers' are written tight, which is why its skills fire on the right requests and stay quiet otherwise.

6Related on the map

References

• Repository
• Gemini CLI skills docs