Superpowers for Claude Code: A Practical Guide

This guide introduces the superpowers plugin for Claude Code — a collection of skills that bring structured workflows to AI-assisted development.

Workshop demo project under review. A brainstorm of candidate W1 demo projects (and their W2 extensions) is in progress at docs/workshops-brainstorming/2026-05-01-w1-demo-candidates-brainstorm.md. The PromptCraft-based materials below describe the current baseline; the brainstorm evaluates whether to keep, replace, or split it across the two workshops.

Who This Is For

• Data scientists building LLM-powered tools and pipelines
• AI engineers working with Claude Code daily
• Both newcomers to AI-assisted coding and experienced users looking to level up

What You'll Learn

Three key skills that transform chaotic prompting into systematic development:

1. Brainstorming — Turn rough ideas into validated designs before writing code
2. Code Review — Systematic review with technical rigor (not performative agreement)
3. Subagent-Driven Development — Fresh agents per task with two-stage review

Start Here

1. Introduction — Why structured workflows matter
2. Tutorial — Build a prompt engineering helper step-by-step

The Example Project

Throughout this guide, we build PromptCraft — a Streamlit app that helps refine LLM prompts. It's meta (using Claude to build Claude tools), practical, and demonstrates all three skills in action.

Proposed Workshops

A two-workshop curriculum built on top of this guide, aimed at a consultancy audience (data scientists, data engineers, AI engineers, BI employees) for shipping deployment-ready code with Claude Code. Each workshop is available in 60-min (vital) and 90-min (vital + bonus) versions; the 90-min is a strict superset.

Workshop 1 — Using Claude Code Well

Covers the everyday primitives, then pivots mid-workshop to superpowers as the integrating workflow. Demo arc: live brainstorm of PromptCraft → reveal of the existing solution branch as the cooked-turkey payoff.

Topics: CLAUDE.md · Context · Compaction · Plan Mode · Checkpoints · Permissions · Slash Commands · Plugins (install) · Brainstorming · Writing-plans · Subagent-driven development.

Workshop 2 — Power Tools + Authoring

~70% consume / 30% author. Tour of professional-grade MCPs (Context7, GitHub MCP, Snowflake MCP) followed by a short authoring snack — write a custom skill in 5 minutes. The 90-min bonus combines Snowflake MCP with a PreToolUse safety hook that blocks destructive SQL on prod databases.

Topics: MCP concept · Context7 · GitHub MCP · Snowflake MCP · Headless mode · Plugin marketplaces · Skills authoring · Hooks · Custom subagents (mention) · Plugin publishing (mention).

Workshop materials

The full design, plan, and content artifacts live in docs/superpowers/:

• Spec: docs/superpowers/specs/2026-04-15-claude-code-workshops-design.md
• Plan: docs/superpowers/plans/2026-04-15-claude-code-workshops.md
• Slide-by-slide content specs: workshop-1/slides.md, workshop-2/slides.md
• Presenter scripts: workshop-1/presenter-script.md, workshop-2/presenter-script.md
• Demo scripts (4 main + 2 bonus): workshop-1/, workshop-2/
• Engineered demo prompts: magic-prompts.md
• Pre-workshop attendee checklists: pre-workshop-w1.md, pre-workshop-w2.md
• Rehearsal + risk register: rehearsal-checklist.md
• Slide 3 rewrite: slide-3-rewrite.md (corrects the existing pptx)
• Demo asset files: demo-assets/ — example CLAUDE.md, custom skill, Snowflake safety hook

The current draft pptx (and PDF export) lives at docs/superpowers/workshops/Claude Code Capabilities Workshop.pptx. The user and a co-presenter build the actual decks themselves from the content specs above.

Status — paused for partner review

Workshop content is complete (17 of 21 plan tasks done). The remaining 4 tasks are paused pending partner review of the slide specs, presenter scripts, and demo scripts. Once approved, these can be resumed:

• Task 18 — custom Skill asset for the W2 Demo 2 authoring demo (demo-assets/w2-demo2-skill/.claude/skills/review-prompt-quality/SKILL.md)
• Task 19 — Snowflake safety hook asset for the W2 90-min bonus demo (demo-assets/w2-bonus-snowflake-hook/ — settings fragment + safety_hook.py + README)
• Task 20 — pre-staged demo branches (workshop-w1-demo, workshop-w2-demo) so demos can git checkout to known-good states
• Task 21 — workshops index README + final cross-link verification (docs/superpowers/workshops/README.md)

All content for Tasks 18–19 is fully specified inline in the plan. To resume, ask Claude to "resume the workshops plan."

Topic coverage map

Capability (slide-2 diagram)	Workshop
① CLAUDE.md	W1
② Permissions	W1
③ Plan Mode	W1
④ Checkpoints	W1
⑤ Skills	W1 (consume via superpowers) + W2 (author)
⑥ Hooks	W2 (90-min bonus)
⑦ MCP	W2
⑧ Plugins	W1 (install) + W2 (marketplace)
⑨ Context	W1
⑩ Slash Commands	W1
⑪ Compaction	W1
⑫ Subagents	W1 (used by superpowers) + W2 (mention)