Refactor app-architect-brainstorm for proportional, persistent agent context

[salim4n]

Goal

Evolve app-architect-brainstorm so it remains useful for serious architecture work without becoming the default heavyweight workflow for every coding task.

Why

The current skill is strong for greenfield and major refactors, but it can overproduce documentation and enforce architectural patterns too rigidly. For coding agents, the objective should be persistent, navigable context rather than loading a full architecture package every time.

Changes

• Add complexity detection at the start: trivial / feature / architectural change.
• Only run the full architecture workflow for genuinely architectural work.
• Separate outputs into AS-IS, DECISIONS, and PLANS to reduce documentation drift.
• Treat source code and tests as implementation ground truth.
• Replace absolute Clean Architecture rules with contextual guidance where appropriate.
• Generate a lightweight context index for future agents (docs/context/README.md).
• Distinguish active plans from archived/completed plans.
• Keep the existing strengths: Socratic questioning, challenge mode, greenfield/reverse-engineering modes, ADRs, diagrams, stack-from-requirements, and no production source-code generation.

Target structure

docs/context/
├── README.md
├── current-state.md
├── decisions/
└── plans/
    ├── active/
    └── archived/

Principle

Convert expensive one-off reasoning into concise, persistent, selectively retrievable project knowledge.

The skill should produce the minimum durable context needed for humans and coding agents to understand, decide, and implement correctly.