The Superpowers Skills System
In one line: Skills are the process — invoking one structures the session around a phase rather than merely reminding you to follow it.
What: Superpowers is an open-source Claude Code plugin providing 14 lifecycle skills, each governing a phase — brainstorming, planning, implementing, debugging, verifying, reviewing, finishing. Skills are invoked per-task via slash commands (/brainstorming, /writing-plans, /test-driven-development) and enforce the Section 3 process.
Why: Process enforcement through documentation (wikis, runbooks, READMEs) fails — it depends on someone reading and following it every time. Skills work because they are the process: /test-driven-development does not just remind you to write tests first, it structures the session around red-green-refactor with an anti-rationalization table that blocks skipping a failing test.
Without skills, the methodology is written principles applied by hand. With them, it is the default workflow and deviating takes deliberate effort.
The 14 Skills:
| Skill | Purpose | Lifecycle Phase |
|---|---|---|
/brainstorming | Structured problem exploration with constraint analysis | Design |
/writing-plans | One design artifact: specification + ordered tasks | Design |
/executing-plans | Execute plan tasks sequentially in current session | Implementation |
/subagent-driven-development | Execute plan via fresh subagent per task with two-stage review | Implementation |
/dispatching-parallel-agents | Run independent tasks concurrently across subagents | Implementation |
/test-driven-development | Red-green-refactor cycle with anti-rationalization table | Implementation |
/systematic-debugging | Four-phase investigation: root cause, pattern, hypothesis, fix | Bug Fix |
/verification-before-completion | Evidence-based completion gate (tests, coverage, linting) | Verification |
/requesting-code-review | Two-stage review: spec compliance then code quality | Review |
/receiving-code-review | Process review findings: evaluate, respond, track dispositions | Review |
/using-git-worktrees | Filesystem-isolated feature branches | Git |
/finishing-a-development-branch | Merge, clean up worktree, update memory | Git |
/using-superpowers | Meta-skill: when and how to invoke other skills | Meta |
/writing-skills | Author and verify new skills before deployment | Meta |
Installation:
claude plugins install superpowers@superpowers-marketplace
The plugin installs globally and is available in all projects. Skills are invoked by name and provide their own instruction context to Claude — each skill's SKILL.md file contains the process definition, decision trees, prompt templates, and output format specifications.
How skills integrate with agents, hooks, and memory:
Skills do not operate in isolation. They integrate with the other methodology components:
-
Skills + Agents: The
/subagent-driven-developmentskill dispatches fresh subagent sessions. Reviewer agents (Section 5) can be invoked during the review phases of skills like/requesting-code-review. The skill's prompt templates include guidance on which agent model to select based on task complexity. -
Skills + Hooks: The
/verification-before-completionskill and the Stop hook (Section 7) serve the same purpose at different levels. The skill provides the structured verification process; the hook provides the fallback check if the skill was not explicitly invoked. Together, they ensure that no task is marked complete without evidence. -
Skills + Memory: The
/finishing-a-development-branchskill includes a memory update step. After merging a feature branch, the skill prompts the AI to update memory files that should reflect the new state of the system — pillar progress, architecture changes, new patterns established. This keeps memory current with the codebase.
Evidence: The skill system's value is the collective guarantee that phases are not skipped — design before implementation, verification before completion, review before merge. Each SKILL.md encodes its phase as executable structure; the enforcement is the skill refusing to produce the next-phase output without the current-phase artifact.
How: Install the plugin with the command above. Skills become available immediately as slash commands. No additional configuration is required. For the full skill integration architecture, see the lifecycle diagrams in Section 3.