PE hyperstack: reflect panel mode - every relevant lens at once

[KailasMahavarkar]

Adds panel mode to reflect: truly generalist reflection - the screen reviewed from every RELEVANT lens at once, each in its own voice, tensions surfaced with winners, blocker classes vetoing. Builds on the 8-archetype roster from #131.

Changes

File	Intent
src/plugins/reflect/snippets/rules/panel-protocol.txt	new MCP-served protocol: relevance-based assembly (3-5 lenses; Sam on every interactive-screen panel), own-voice rule, named collisions with per-product winners, class-owned blocker vetoes, verdict synthesis
src/plugins/reflect/tools/get-panel.ts	new reflect_get_panel tool (4th): protocol + voice rules + seated lens docs in one call; optional ids scope
skills/reflect/SKILL.md	Panel mode section + description trigger ("get everyone's read")
personas/reflect/{persona.json, LIFECYCLE.md}	multi-persona-panel capability + panel lifecycle
tests/reflect-behaviour.test.ts	get_panel scoping test (ids subset excludes unseated lenses; full call seats all)
bootstrap bullet + INDEX	recompiled/regenerated

Key protocol rules

• Relevance, not headcount - seat the 3-5 lenses the product actually has as customers; all 8 only on request.
• Own voice or it failed - if two lens lines could swap speakers, the panel is one lens repeated.
• Collisions get winners - speed vs trust, depth vs 8-second patience, polish vs accessibility: named, with who wins for THIS product.
• Vetoes belong to the class, not the seat - an accessibility/security/compliance breach fires even when its natural owner is unseated (gap found live by the pressure test, closed here).

Pressure-test evidence

Scenario: "get me everyone's read" on a devtool pricing page (card-required free tier, seat pricing, checkout-revealed annual billing, Escape-proof modal, pulsing badge).

• Seated Kenji + Sandra + Zoe + Sam exactly per the protocol's devtool prescription; Morgan/Diane/Max/Riley correctly left out.
• 4 distinct voices (self-report: only one near-swap, defensible convergence); 3 collisions named with winners; verdict table + one panel call.
• Sam's keyboard-trap veto fired per the veto line; the checkout-reveal billing exposed the seat-bound compliance-veto gap - protocol now class-owned.

Verification

• bun test: 87 pass, 0 fail (was 86; +1 panel test)
• tsc --noEmit: clean
• bun run compile:context + skills:index: regenerated
• em-dash sweep: clean