writing-plans: ask for execution mode *before* writing the plan, not after

The `writing-plans` skill currently writes the full plan first, then offers the execution mode choice at the end:

> "Plan complete. Two execution options: 1. Subagent-Driven (recommended) 2. Inline Execution"

This is backwards. The two modes require fundamentally different plan formats:

- **Agentic**: Each subagent starts with zero context, so the plan must be self-contained — full code blocks, exact commands, expected outputs, repeated context across tasks.
- **Inline**: Claude retains full session context from brainstorming, so a lightweight task checklist (task list + file list) is sufficient. Writing a full agentic-style plan for inline execution wastes significant tokens and produces a document nobody reads.

**Proposed fix:** Ask "Inline or agentic?" as the *first* step of `writing-plans`, before writing anything. Then write the appropriate format.

Same issue exists in `brainstorming/SKILL.md` step 9, which unconditionally transitions to `writing-plans` without capturing execution intent — that intent could be passed as context to avoid the question being asked twice.

Read the writing-plans skill and brainstorming/SKILL.md, especially the transition described in step 9. Trace how execution intent is currently requested and ensure the choice comes before plan generation, is passed between skills, and produces the appropriate plan format without asking twice.

Written by the indexing model from the issue text.