Brainstorm skill: add multi-level discussion modes with auto-detection

[rolker]

Summary

Enhance /brainstorm with multi-level discussion modes so the skill adapts to the level of abstraction needed. Inspired by gstack's hierarchical skill model (office-hours → CEO review → eng review → design review), but as modes within a single skill rather than separate skills.

Levels

Level	Focus	Key question
Vision	Problem validation, "is this worth building?"	"What problem are we really solving?"
Strategy	Scope, trajectory, design iteration	"Is this the right scope? Where are we headed in 12 months?"
Architecture	Structure, failure modes, tradeoffs	"Will this design hold? What are the failure modes?"
Design	Usability, quality, "what does 10/10 look like?"	"Is this the right UX/DX? What would make it great?"

Auto-detection

Rather than requiring the user to pick a level, the skill should triage based on signals:

• No design doc / no prior brainstorm → suggest Vision
• Design doc exists + issues accumulating against it → suggest Strategy
• Plan exists but architecture concerns → suggest Architecture
• Implementation in progress, UX/quality questions → suggest Design

The user can always override: "no, I want to zoom out" or "let's go deeper."

Connection to design iteration workflow

Roland's design iteration pattern is triggered by issue accumulation — when open issues suggest the higher-level design needs work. The auto-detection should recognize this signal and suggest shifting up a level proactively.

Roadmap

Phase C (Honest Advisor) — this is about how the agent thinks and at what level it engages.

Inspired by gstack's /office-hours, /plan-ceo-review, /plan-eng-review, /plan-design-review. See .agent/knowledge/inspiration_gstack_digest.md for details.

Part of #63.

---

Authored-By: Claude Code Agent
Model: claude-opus-4-6

[rolker]

Opened on wrong repo by mistake. Sorry!