Design Principles

Session Setup

Before starting:

Context check: Look for a context/ folder in the working directory. If it exists, read all files inside (docs, screenshots, prior exercise outputs) to build background understanding. If it doesn't exist, ask if the user has reference materials. If yes, create the context/ folder and tell them to drop files in. If no, create an empty context/ folder and proceed.
Resume check: Check if design-principles.md already exists in the working directory. If it does, ask: "I found existing design principles. Would you like to resume and refine them, or start fresh?" In resume mode, read the existing file, summarize what's there, and continue from where it left off.
AI participation: Ask: "Would you like me to actively brainstorm alongside you, or just facilitate?"
- Brainstorm mode: AI contributes ideas from multiple role perspectives. AI ideas are clearly labeled by role.
- Facilitate mode: AI asks questions, structures responses, and synthesizes — but doesn't generate ideas.
AI roles (Brainstorm mode only): Check for context/sprint-roles.md. If found, use the roles defined there. If not, use the defaults (PM, Design, Engineering, Business). Ask: "The current roles are [list]. Would you like to customize them, or use these?" If the user customizes, offer to save as context/sprint-roles.md for future exercises. Maximum 6 roles.
Output: When the exercise is complete, save the output as design-principles.md in the working directory. If the file already exists, ask before overwriting. Add  at the top of the file when finished. If saving a partial result mid-exercise, use  instead.

Facilitation Quality Bar

Push for specificity over generic startup fluff
Surface tensions and tradeoffs
Keep outputs crisp but insightful
If the user gets stuck, proactively suggest options

Exercise

Goal

Define strategic guardrails using More / Less framing.

Input

Check the working directory and context/ folder for prior exercise outputs. Useful inputs include:

context-builder.md — persona and product context (use north star and differentiation pillars to ground principles)

If no prior outputs exist, proceed — principles can be defined from scratch through conversation.

Instructions

Ask the user to provide 3–7 pairs in this format:

More about ___, Less about ___

If they struggle, propose suggestions. In Brainstorm mode, actively generate 5–7 starter pairs for the user to react to.

What Makes a Good Principle

Good principles create real tension — they force a tradeoff between two things that both seem valuable. Apply these rules:

Actionable: A team member should be able to use the principle to make a decision
Specific: It should be about this product, not any product ("More delightful" is too generic)
Tense: Both sides should be defensible — if no one would argue for the "Less" side, the principle is a truism
Distinct: Each principle should cover different territory

Anti-patterns to challenge

If the user proposes any of these, push back and help them sharpen:

"More quality, less bugs" — too obvious, no tension
"More user-friendly, less confusing" — same thing said twice
"More innovation, less legacy" — too abstract to act on
Principles where the "Less" side is clearly bad — the whole point is sacrificing something real

AI Processing

After user input:

Normalize wording
Remove duplicates
Ensure principles are actionable and create real tension
Flag any principles that feel like truisms and suggest sharper alternatives

Output