Grill the talk on stage

Use when stage-grill is the current gate.

• Read repo instructions, PRESENTATION-BRIEF.md, STORY.md, STORYBOARD.md, GRILL-LOG.md, current speaker notes, and actual page renders before judging the talk.
• Always explore the files instead of asking the user for discoverable facts.
• Grill in rounds per ../make-presentation/references/grilling-rounds.md when consequential stage decisions are unresolved: ask the whole frontier in one round.
• Include a recommended answer with each question, worded so "yes" accepts it.
• Use references/autonomous-stage.md for the autonomous fallback when the user is unavailable (headless run) or the remaining stage decisions are resolvable from the deck and upstream artifacts — apply references/stage-questions.md yourself, resolve each finding with a concrete change and confidence level, and flag any low-confidence load-bearing edit for later human review instead of silently rewriting it.
• When the user refers to the page they're currently viewing without naming it, resolve it through the engine's "current page" mechanism (e.g. current-slide for open-slide) before judging.
• Use actual renders and current speaker notes, not outline-only summaries.
• Apply references/stage-questions.md to representative prototype pages first, then to the full deck when available.
• Enforce presenter-language constraints, timed trimming, and existing-versus-remaining distinction.
• Speakability & Spoken-First Phrasing: Speaker notes must be written in natural, fluent spoken phrasing that the presenter can deliver aloud comfortably without stumbling (no dense academic paper sentences).
• Strict 1:1 Note Parity: Every slide must have exactly one corresponding speaker note matching its dominant message.
• Adult Attention Management & Active Learning Resets: In presentations longer than 30 minutes, verify that the delivery incorporates structured attention reset switches (e.g. audience polling/temperature checks, buzz-pair sharing, dialogue choral rehearsal, silent reflection) to combat the natural 15–20 minute adult attention decay.
• Behavioral Scaffolding (Kirkpatrick Level 3 Transfer): Ensure actionable tools include verbal rehearsal opportunities and take-home scaffolds (checklists, pocket cards) so knowledge bridges into daily behavioral change.
• Language Purity: For international presentations (e.g. English summit), confirm 100% target language consistency across all notes and visible cues with zero stray translation leftovers.
• Constructive Context Framing ("Not starting from zero"): When presenting local policy or regional case studies, clearly separate what the community/system has already established from the next challenge, avoiding nihilistic or accusatory framing.
• Separate what the audience can see immediately from what only works after narration.
• Run node .agents/skills/make-presentation/scripts/unslop-check.mjs <work-root>/SPEAKER-SCRIPT.md <slide-folder> and rewrite flagged notes using ../make-presentation/references/unslop-copy.md. Spoken phrasing still comes first.
• Sort audience-challenge and trimming findings into Act on, Consider, Noted, or Dismissed using ../make-presentation/references/review-buckets.md.
• Produce SPEAKER-SCRIPT.md from templates/SPEAKER-SCRIPT.md with approved wording, timing, likely audience challenge, and Q&A risk.
• Return pages for revision when density, wording, transition logic, or defensibility breaks the live talk.
• Return passed only when the talk is readable, speakable, timed, defensible, and audience-appropriate.

Write only consequential decisions to GRILL-LOG.md from ../presentation-product-grill/templates/GRILL-LOG.md.

• Append new decisions without deleting or rewriting entries from other grill phases or earlier runs.
• Preserve existing decisions when revisiting the stage-grill gate.
• Log only changes that materially affect wording, timing, page order, trims, likely objections, or presenter language.