badpowers

Three Claude Code skills for rigorous planning, review, and design — with built-in anti-sycophancy.

Skills

/think — Brainstorm & Validate

Structured brainstorming with two modes:

• Product mode — Six forcing questions that expose demand reality, status quo gaps, and the narrowest viable wedge. Pushes until answers are specific and evidence-based.
• Builder mode — Enthusiastic design collaborator for side projects and hackathons. Finds the most exciting version.

Output: Design document with premises, alternatives, and a concrete next action.

/audit — Plan & Architecture Review

Rigorous plan review with adaptive depth:

• DEEP mode — 10 review sections, 4 scope modes (expansion/selective/hold/reduction), dream state mapping, temporal interrogation. For strategic decisions and major refactors.
• STANDARD mode — 4 review sections (Architecture, Code Quality, Tests, Performance). For well-scoped implementations and bug fixes.

Output: Error & Rescue Map, Failure Modes Registry, Completion Summary with Lake Score.

/critique — Design Review & Design System

Design review with teeth:

• Review mode — 7-pass rating system (0-10) covering information architecture, interaction states, user journey, AI slop risk, design system alignment, responsive/a11y, and unresolved decisions.
• Create mode — Proposes a complete, coherent design system (aesthetic, typography, color, spacing, motion) and generates DESIGN.md.

Output: Design ratings with fixes / DESIGN.md with preview page.

What makes these different

• Anti-sycophancy built in. No “that’s interesting” or “great idea.” Takes positions, names problems directly, pushes on vague answers.
• Structured failure analysis. Every review maps error paths, names specific exceptions, and flags silent failures as critical gaps.
• Cognitive patterns with attribution. Engineering instincts from Bezos (reversibility), Munger (inversion), Jobs (subtraction), Rams (less but better) — applied naturally, not as checklists.
• Skills chain together. Each skill tells you what to run next and names the specific artifact that carries over.

Installation

# Clone to your Claude Code skills directory
git clone https://github.com/deadjoe/badpowers.git ~/.claude/skills/badpowers

That’s it. The skills are available immediately as /think, /audit, and /critique in Claude Code.

Updating

cd ~/.claude/skills/badpowers && git pull

Structure

badpowers/
├── think/
│   └── SKILL.md          # Brainstorming & validation
├── audit/
│   └── SKILL.md          # Plan & architecture review
├── critique/
│   └── SKILL.md          # Design review & system creation
└── references/
    ├── anti-sycophancy.md # Pushback framework (shared by think + audit)
    ├── review-framework.md # Error maps, failure registries, test diagrams (audit)
    └── design-knowledge.md # Aesthetics, fonts, AI slop patterns (critique)

Quick reference

You’re thinking…	Use	Mode
“Is this worth building?”	/think	Product
“Weekend project, brainstorm with me”	/think	Builder
“Review my plan for adding auth”	/audit	STANDARD
“Big refactor, audit this deeply”	/audit	DEEP
“My UI looks generic”	/critique	Review
“New project, need a design system”	/critique	Create

Workflow

/think  ──→  /audit  ──→  /critique  ──→  code
  │            │             │
  design doc   audit report   DESIGN.md

Each skill’s output feeds the next. /think produces a design doc that /audit reviews. /audit’s findings inform /critique’s design decisions. /critique’s DESIGN.md becomes the reference for implementation.

License

MIT