Skip to content
UXClaim
Design Ops

Propose

Run three parallel design agents with different philosophies to generate ranked approach proposals for any design decision.

What it does

/propose is a Claude Code skill that breaks the first-idea trap by generating multiple design approaches in parallel. When you’re stuck on a design decision, invoke /propose and three subagents spin up simultaneously—each with a distinct philosophy: a Pragmatist seeking the smallest viable change, an Architect building for extensibility, and a Wildcard challenging your assumptions. Each agent reads your actual codebase, grounds proposals in real files, and returns a four-part analysis: Approach, How It Works, Tradeoffs, and Effort estimate.

How it works

After you describe a design question, Claude launches three independent subagents in parallel. Each agent reviews your codebase and proposes a solution without making changes. They return their findings in a consistent format, which Claude then ranks by fit for your situation, flags a clear winner or hybrid option, and asks which direction you’d like to pursue.

Use cases

  • Unblock design decisions when you’re torn between approaches
  • Explore alternative architectures before committing to one path
  • Challenge your assumptions by forcing lateral thinking
  • Ground proposals in reality by having agents read actual code
  • Make faster decisions by seeing trade-offs side-by-side in one turn

Who benefits

Product designers, design ops leads, and technical leads who use Claude Code and want a structured way to explore multiple design directions without implementation overhead.

Frequently asked questions

How do I install the propose skill?
Copy `SKILL.md` from the GitHub repo into `~/.claude/skills/propose/SKILL.md`, then restart Claude Code. Type `/propose` to invoke it. No build step or external tools required.
What happens when I run /propose?
Claude launches three subagents in parallel—Pragmatist, Architect, and Wildcard—each analyzing your codebase and proposing a different approach. They return ranked proposals with tradeoffs and effort estimates, then ask which direction you'd like to pursue.
Do the agents make changes to my code?
No. The agents are research-only. They read your codebase and propose solutions, but never edit files. Nothing changes until you explicitly pick a direction and ask Claude to implement it.
Can I use /propose without describing the problem again?
Yes. `/propose` picks up the current design decision from your conversation context, so you can invoke it mid-discussion without re-explaining. You can also specify a new question: `/propose how should we handle offline sync?`
What are the three agent philosophies?
Pragmatist: smallest change that works, low risk, reuses existing patterns. Architect: robust, extensible, handles edge cases. Wildcard: unexpected, lateral, challenges the premise entirely.
What do I need to run /propose?
You need Claude Code CLI installed. The skill relies on Claude Code's built-in subagent support to run the three proposers in parallel. No external tools, scripts, or MCP servers required.
Can /propose work in an empty repository?
The agents are instructed to read actual source files to ground their proposals in reality. An empty repo won't exercise the skill fully—use it on a real project with actual code.
How does /propose rank the three proposals?
Claude presents them as a ranked list based on fit for your situation, notes if one is clearly best or if combining two would be ideal, then asks whether you want to proceed with one approach or a hybrid.

Glossary

Subagent
An independent AI agent spawned by Claude Code to run research or exploration tasks in parallel. In `/propose`, three subagents run concurrently to generate different design proposals.
Research-only
Agents that explore and analyze code without making changes. They read files, propose solutions, and report findings, but never edit or implement.
Claude Code skill
A self-contained Markdown file with YAML frontmatter that defines a reusable command or workflow for Claude Code. Skills are installed locally and triggered by typing `/skillname` in a conversation.
Effort estimate
A rough size categorization—Small, Medium, or Large—indicating the development effort required to implement a proposed approach.

More in Design Ops

All →
Design Ops

Claude Skills

Production-grade Claude AI skills for stock analysis, prompt engineering, meeting documentation, and UX design with visual-first outputs.

bhrpraju
Design Ops

SOTA Present

Generate polished HTML slides, editable PowerPoint, and Feishu whiteboards from one content description with coordinated design and anti-slop taste rules.

YMaxwellHayes
Design Ops

AI Code Refactoring

Disciplined 5-phase workflow for refactoring AI-generated code with safety nets, design systems, and tests.

chake111
Design Ops

Goal Workflow Designer

Interrogates you until your /goal prompt is precise enough to work, using a five-element framework and rubric SOP.